diff options
author | Daniel Baumann <daniel.baumann@progress-linux.org> | 2024-05-04 12:47:55 +0000 |
---|---|---|
committer | Daniel Baumann <daniel.baumann@progress-linux.org> | 2024-05-04 12:47:55 +0000 |
commit | 2aadc03ef15cb5ca5cc2af8a7c08e070742f0ac4 (patch) | |
tree | 033cc839730fda84ff08db877037977be94e5e3a /vendor/hashbrown/src | |
parent | Initial commit. (diff) | |
download | cargo-2aadc03ef15cb5ca5cc2af8a7c08e070742f0ac4.tar.xz cargo-2aadc03ef15cb5ca5cc2af8a7c08e070742f0ac4.zip |
Adding upstream version 0.70.1+ds1.upstream/0.70.1+ds1upstream
Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>
Diffstat (limited to 'vendor/hashbrown/src')
22 files changed, 19988 insertions, 0 deletions
diff --git a/vendor/hashbrown/src/external_trait_impls/mod.rs b/vendor/hashbrown/src/external_trait_impls/mod.rs new file mode 100644 index 0000000..01d386b --- /dev/null +++ b/vendor/hashbrown/src/external_trait_impls/mod.rs @@ -0,0 +1,6 @@ +#[cfg(feature = "rayon")] +pub(crate) mod rayon; +#[cfg(feature = "rkyv")] +mod rkyv; +#[cfg(feature = "serde")] +mod serde; diff --git a/vendor/hashbrown/src/external_trait_impls/rayon/helpers.rs b/vendor/hashbrown/src/external_trait_impls/rayon/helpers.rs new file mode 100644 index 0000000..070b08c --- /dev/null +++ b/vendor/hashbrown/src/external_trait_impls/rayon/helpers.rs @@ -0,0 +1,27 @@ +use alloc::collections::LinkedList; +use alloc::vec::Vec; + +use rayon::iter::{IntoParallelIterator, ParallelIterator}; + +/// Helper for collecting parallel iterators to an intermediary +#[allow(clippy::linkedlist)] // yes, we need linked list here for efficient appending! +pub(super) fn collect<I: IntoParallelIterator>(iter: I) -> (LinkedList<Vec<I::Item>>, usize) { + let list = iter + .into_par_iter() + .fold(Vec::new, |mut vec, elem| { + vec.push(elem); + vec + }) + .map(|vec| { + let mut list = LinkedList::new(); + list.push_back(vec); + list + }) + .reduce(LinkedList::new, |mut list1, mut list2| { + list1.append(&mut list2); + list1 + }); + + let len = list.iter().map(Vec::len).sum(); + (list, len) +} diff --git a/vendor/hashbrown/src/external_trait_impls/rayon/map.rs b/vendor/hashbrown/src/external_trait_impls/rayon/map.rs new file mode 100644 index 0000000..2534dc9 --- /dev/null +++ b/vendor/hashbrown/src/external_trait_impls/rayon/map.rs @@ -0,0 +1,721 @@ +//! Rayon extensions for `HashMap`. + +use super::raw::{RawIntoParIter, RawParDrain, RawParIter}; +use crate::hash_map::HashMap; +use crate::raw::{Allocator, Global}; +use core::fmt; +use core::hash::{BuildHasher, Hash}; +use core::marker::PhantomData; +use rayon::iter::plumbing::UnindexedConsumer; +use rayon::iter::{FromParallelIterator, IntoParallelIterator, ParallelExtend, ParallelIterator}; + +/// Parallel iterator over shared references to entries in a map. +/// +/// This iterator is created by the [`par_iter`] method on [`HashMap`] +/// (provided by the [`IntoParallelRefIterator`] trait). +/// See its documentation for more. +/// +/// [`par_iter`]: /hashbrown/struct.HashMap.html#method.par_iter +/// [`HashMap`]: /hashbrown/struct.HashMap.html +/// [`IntoParallelRefIterator`]: https://docs.rs/rayon/1.0/rayon/iter/trait.IntoParallelRefIterator.html +pub struct ParIter<'a, K, V> { + inner: RawParIter<(K, V)>, + marker: PhantomData<(&'a K, &'a V)>, +} + +impl<'a, K: Sync, V: Sync> ParallelIterator for ParIter<'a, K, V> { + type Item = (&'a K, &'a V); + + #[cfg_attr(feature = "inline-more", inline)] + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: UnindexedConsumer<Self::Item>, + { + self.inner + .map(|x| unsafe { + let r = x.as_ref(); + (&r.0, &r.1) + }) + .drive_unindexed(consumer) + } +} + +impl<K, V> Clone for ParIter<'_, K, V> { + #[cfg_attr(feature = "inline-more", inline)] + fn clone(&self) -> Self { + Self { + inner: self.inner.clone(), + marker: PhantomData, + } + } +} + +impl<K: fmt::Debug + Eq + Hash, V: fmt::Debug> fmt::Debug for ParIter<'_, K, V> { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + let iter = unsafe { self.inner.iter() }.map(|x| unsafe { + let r = x.as_ref(); + (&r.0, &r.1) + }); + f.debug_list().entries(iter).finish() + } +} + +/// Parallel iterator over shared references to keys in a map. +/// +/// This iterator is created by the [`par_keys`] method on [`HashMap`]. +/// See its documentation for more. +/// +/// [`par_keys`]: /hashbrown/struct.HashMap.html#method.par_keys +/// [`HashMap`]: /hashbrown/struct.HashMap.html +pub struct ParKeys<'a, K, V> { + inner: RawParIter<(K, V)>, + marker: PhantomData<(&'a K, &'a V)>, +} + +impl<'a, K: Sync, V: Sync> ParallelIterator for ParKeys<'a, K, V> { + type Item = &'a K; + + #[cfg_attr(feature = "inline-more", inline)] + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: UnindexedConsumer<Self::Item>, + { + self.inner + .map(|x| unsafe { &x.as_ref().0 }) + .drive_unindexed(consumer) + } +} + +impl<K, V> Clone for ParKeys<'_, K, V> { + #[cfg_attr(feature = "inline-more", inline)] + fn clone(&self) -> Self { + Self { + inner: self.inner.clone(), + marker: PhantomData, + } + } +} + +impl<K: fmt::Debug + Eq + Hash, V> fmt::Debug for ParKeys<'_, K, V> { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + let iter = unsafe { self.inner.iter() }.map(|x| unsafe { &x.as_ref().0 }); + f.debug_list().entries(iter).finish() + } +} + +/// Parallel iterator over shared references to values in a map. +/// +/// This iterator is created by the [`par_values`] method on [`HashMap`]. +/// See its documentation for more. +/// +/// [`par_values`]: /hashbrown/struct.HashMap.html#method.par_values +/// [`HashMap`]: /hashbrown/struct.HashMap.html +pub struct ParValues<'a, K, V> { + inner: RawParIter<(K, V)>, + marker: PhantomData<(&'a K, &'a V)>, +} + +impl<'a, K: Sync, V: Sync> ParallelIterator for ParValues<'a, K, V> { + type Item = &'a V; + + #[cfg_attr(feature = "inline-more", inline)] + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: UnindexedConsumer<Self::Item>, + { + self.inner + .map(|x| unsafe { &x.as_ref().1 }) + .drive_unindexed(consumer) + } +} + +impl<K, V> Clone for ParValues<'_, K, V> { + #[cfg_attr(feature = "inline-more", inline)] + fn clone(&self) -> Self { + Self { + inner: self.inner.clone(), + marker: PhantomData, + } + } +} + +impl<K: Eq + Hash, V: fmt::Debug> fmt::Debug for ParValues<'_, K, V> { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + let iter = unsafe { self.inner.iter() }.map(|x| unsafe { &x.as_ref().1 }); + f.debug_list().entries(iter).finish() + } +} + +/// Parallel iterator over mutable references to entries in a map. +/// +/// This iterator is created by the [`par_iter_mut`] method on [`HashMap`] +/// (provided by the [`IntoParallelRefMutIterator`] trait). +/// See its documentation for more. +/// +/// [`par_iter_mut`]: /hashbrown/struct.HashMap.html#method.par_iter_mut +/// [`HashMap`]: /hashbrown/struct.HashMap.html +/// [`IntoParallelRefMutIterator`]: https://docs.rs/rayon/1.0/rayon/iter/trait.IntoParallelRefMutIterator.html +pub struct ParIterMut<'a, K, V> { + inner: RawParIter<(K, V)>, + marker: PhantomData<(&'a K, &'a mut V)>, +} + +impl<'a, K: Sync, V: Send> ParallelIterator for ParIterMut<'a, K, V> { + type Item = (&'a K, &'a mut V); + + #[cfg_attr(feature = "inline-more", inline)] + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: UnindexedConsumer<Self::Item>, + { + self.inner + .map(|x| unsafe { + let r = x.as_mut(); + (&r.0, &mut r.1) + }) + .drive_unindexed(consumer) + } +} + +impl<K: fmt::Debug + Eq + Hash, V: fmt::Debug> fmt::Debug for ParIterMut<'_, K, V> { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + ParIter { + inner: self.inner.clone(), + marker: PhantomData, + } + .fmt(f) + } +} + +/// Parallel iterator over mutable references to values in a map. +/// +/// This iterator is created by the [`par_values_mut`] method on [`HashMap`]. +/// See its documentation for more. +/// +/// [`par_values_mut`]: /hashbrown/struct.HashMap.html#method.par_values_mut +/// [`HashMap`]: /hashbrown/struct.HashMap.html +pub struct ParValuesMut<'a, K, V> { + inner: RawParIter<(K, V)>, + marker: PhantomData<(&'a K, &'a mut V)>, +} + +impl<'a, K: Sync, V: Send> ParallelIterator for ParValuesMut<'a, K, V> { + type Item = &'a mut V; + + #[cfg_attr(feature = "inline-more", inline)] + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: UnindexedConsumer<Self::Item>, + { + self.inner + .map(|x| unsafe { &mut x.as_mut().1 }) + .drive_unindexed(consumer) + } +} + +impl<K: Eq + Hash, V: fmt::Debug> fmt::Debug for ParValuesMut<'_, K, V> { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + ParValues { + inner: self.inner.clone(), + marker: PhantomData, + } + .fmt(f) + } +} + +/// Parallel iterator over entries of a consumed map. +/// +/// This iterator is created by the [`into_par_iter`] method on [`HashMap`] +/// (provided by the [`IntoParallelIterator`] trait). +/// See its documentation for more. +/// +/// [`into_par_iter`]: /hashbrown/struct.HashMap.html#method.into_par_iter +/// [`HashMap`]: /hashbrown/struct.HashMap.html +/// [`IntoParallelIterator`]: https://docs.rs/rayon/1.0/rayon/iter/trait.IntoParallelIterator.html +pub struct IntoParIter<K, V, A: Allocator = Global> { + inner: RawIntoParIter<(K, V), A>, +} + +impl<K: Send, V: Send, A: Allocator + Send> ParallelIterator for IntoParIter<K, V, A> { + type Item = (K, V); + + #[cfg_attr(feature = "inline-more", inline)] + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: UnindexedConsumer<Self::Item>, + { + self.inner.drive_unindexed(consumer) + } +} + +impl<K: fmt::Debug + Eq + Hash, V: fmt::Debug, A: Allocator> fmt::Debug for IntoParIter<K, V, A> { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + ParIter { + inner: unsafe { self.inner.par_iter() }, + marker: PhantomData, + } + .fmt(f) + } +} + +/// Parallel draining iterator over entries of a map. +/// +/// This iterator is created by the [`par_drain`] method on [`HashMap`]. +/// See its documentation for more. +/// +/// [`par_drain`]: /hashbrown/struct.HashMap.html#method.par_drain +/// [`HashMap`]: /hashbrown/struct.HashMap.html +pub struct ParDrain<'a, K, V, A: Allocator = Global> { + inner: RawParDrain<'a, (K, V), A>, +} + +impl<K: Send, V: Send, A: Allocator + Sync> ParallelIterator for ParDrain<'_, K, V, A> { + type Item = (K, V); + + #[cfg_attr(feature = "inline-more", inline)] + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: UnindexedConsumer<Self::Item>, + { + self.inner.drive_unindexed(consumer) + } +} + +impl<K: fmt::Debug + Eq + Hash, V: fmt::Debug, A: Allocator> fmt::Debug for ParDrain<'_, K, V, A> { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + ParIter { + inner: unsafe { self.inner.par_iter() }, + marker: PhantomData, + } + .fmt(f) + } +} + +impl<K: Sync, V: Sync, S, A: Allocator> HashMap<K, V, S, A> { + /// Visits (potentially in parallel) immutably borrowed keys in an arbitrary order. + #[cfg_attr(feature = "inline-more", inline)] + pub fn par_keys(&self) -> ParKeys<'_, K, V> { + ParKeys { + inner: unsafe { self.table.par_iter() }, + marker: PhantomData, + } + } + + /// Visits (potentially in parallel) immutably borrowed values in an arbitrary order. + #[cfg_attr(feature = "inline-more", inline)] + pub fn par_values(&self) -> ParValues<'_, K, V> { + ParValues { + inner: unsafe { self.table.par_iter() }, + marker: PhantomData, + } + } +} + +impl<K: Send, V: Send, S, A: Allocator> HashMap<K, V, S, A> { + /// Visits (potentially in parallel) mutably borrowed values in an arbitrary order. + #[cfg_attr(feature = "inline-more", inline)] + pub fn par_values_mut(&mut self) -> ParValuesMut<'_, K, V> { + ParValuesMut { + inner: unsafe { self.table.par_iter() }, + marker: PhantomData, + } + } + + /// Consumes (potentially in parallel) all values in an arbitrary order, + /// while preserving the map's allocated memory for reuse. + #[cfg_attr(feature = "inline-more", inline)] + pub fn par_drain(&mut self) -> ParDrain<'_, K, V, A> { + ParDrain { + inner: self.table.par_drain(), + } + } +} + +impl<K, V, S, A> HashMap<K, V, S, A> +where + K: Eq + Hash + Sync, + V: PartialEq + Sync, + S: BuildHasher + Sync, + A: Allocator + Sync, +{ + /// Returns `true` if the map is equal to another, + /// i.e. both maps contain the same keys mapped to the same values. + /// + /// This method runs in a potentially parallel fashion. + pub fn par_eq(&self, other: &Self) -> bool { + self.len() == other.len() + && self + .into_par_iter() + .all(|(key, value)| other.get(key).map_or(false, |v| *value == *v)) + } +} + +impl<K: Send, V: Send, S, A: Allocator + Send> IntoParallelIterator for HashMap<K, V, S, A> { + type Item = (K, V); + type Iter = IntoParIter<K, V, A>; + + #[cfg_attr(feature = "inline-more", inline)] + fn into_par_iter(self) -> Self::Iter { + IntoParIter { + inner: self.table.into_par_iter(), + } + } +} + +impl<'a, K: Sync, V: Sync, S, A: Allocator> IntoParallelIterator for &'a HashMap<K, V, S, A> { + type Item = (&'a K, &'a V); + type Iter = ParIter<'a, K, V>; + + #[cfg_attr(feature = "inline-more", inline)] + fn into_par_iter(self) -> Self::Iter { + ParIter { + inner: unsafe { self.table.par_iter() }, + marker: PhantomData, + } + } +} + +impl<'a, K: Sync, V: Send, S, A: Allocator> IntoParallelIterator for &'a mut HashMap<K, V, S, A> { + type Item = (&'a K, &'a mut V); + type Iter = ParIterMut<'a, K, V>; + + #[cfg_attr(feature = "inline-more", inline)] + fn into_par_iter(self) -> Self::Iter { + ParIterMut { + inner: unsafe { self.table.par_iter() }, + marker: PhantomData, + } + } +} + +/// Collect (key, value) pairs from a parallel iterator into a +/// hashmap. If multiple pairs correspond to the same key, then the +/// ones produced earlier in the parallel iterator will be +/// overwritten, just as with a sequential iterator. +impl<K, V, S> FromParallelIterator<(K, V)> for HashMap<K, V, S, Global> +where + K: Eq + Hash + Send, + V: Send, + S: BuildHasher + Default, +{ + fn from_par_iter<P>(par_iter: P) -> Self + where + P: IntoParallelIterator<Item = (K, V)>, + { + let mut map = HashMap::default(); + map.par_extend(par_iter); + map + } +} + +/// Extend a hash map with items from a parallel iterator. +impl<K, V, S, A> ParallelExtend<(K, V)> for HashMap<K, V, S, A> +where + K: Eq + Hash + Send, + V: Send, + S: BuildHasher, + A: Allocator, +{ + fn par_extend<I>(&mut self, par_iter: I) + where + I: IntoParallelIterator<Item = (K, V)>, + { + extend(self, par_iter); + } +} + +/// Extend a hash map with copied items from a parallel iterator. +impl<'a, K, V, S, A> ParallelExtend<(&'a K, &'a V)> for HashMap<K, V, S, A> +where + K: Copy + Eq + Hash + Sync, + V: Copy + Sync, + S: BuildHasher, + A: Allocator, +{ + fn par_extend<I>(&mut self, par_iter: I) + where + I: IntoParallelIterator<Item = (&'a K, &'a V)>, + { + extend(self, par_iter); + } +} + +// This is equal to the normal `HashMap` -- no custom advantage. +fn extend<K, V, S, A, I>(map: &mut HashMap<K, V, S, A>, par_iter: I) +where + K: Eq + Hash, + S: BuildHasher, + I: IntoParallelIterator, + A: Allocator, + HashMap<K, V, S, A>: Extend<I::Item>, +{ + let (list, len) = super::helpers::collect(par_iter); + + // Keys may be already present or show multiple times in the iterator. + // Reserve the entire length if the map is empty. + // Otherwise reserve half the length (rounded up), so the map + // will only resize twice in the worst case. + let reserve = if map.is_empty() { len } else { (len + 1) / 2 }; + map.reserve(reserve); + for vec in list { + map.extend(vec); + } +} + +#[cfg(test)] +mod test_par_map { + use alloc::vec::Vec; + use core::hash::{Hash, Hasher}; + use core::sync::atomic::{AtomicUsize, Ordering}; + + use rayon::prelude::*; + + use crate::hash_map::HashMap; + + struct Dropable<'a> { + k: usize, + counter: &'a AtomicUsize, + } + + impl Dropable<'_> { + fn new(k: usize, counter: &AtomicUsize) -> Dropable<'_> { + counter.fetch_add(1, Ordering::Relaxed); + + Dropable { k, counter } + } + } + + impl Drop for Dropable<'_> { + fn drop(&mut self) { + self.counter.fetch_sub(1, Ordering::Relaxed); + } + } + + impl Clone for Dropable<'_> { + fn clone(&self) -> Self { + Dropable::new(self.k, self.counter) + } + } + + impl Hash for Dropable<'_> { + fn hash<H>(&self, state: &mut H) + where + H: Hasher, + { + self.k.hash(state); + } + } + + impl PartialEq for Dropable<'_> { + fn eq(&self, other: &Self) -> bool { + self.k == other.k + } + } + + impl Eq for Dropable<'_> {} + + #[test] + fn test_into_iter_drops() { + let key = AtomicUsize::new(0); + let value = AtomicUsize::new(0); + + let hm = { + let mut hm = HashMap::new(); + + assert_eq!(key.load(Ordering::Relaxed), 0); + assert_eq!(value.load(Ordering::Relaxed), 0); + + for i in 0..100 { + let d1 = Dropable::new(i, &key); + let d2 = Dropable::new(i + 100, &value); + hm.insert(d1, d2); + } + + assert_eq!(key.load(Ordering::Relaxed), 100); + assert_eq!(value.load(Ordering::Relaxed), 100); + + hm + }; + + // By the way, ensure that cloning doesn't screw up the dropping. + drop(hm.clone()); + + assert_eq!(key.load(Ordering::Relaxed), 100); + assert_eq!(value.load(Ordering::Relaxed), 100); + + // Ensure that dropping the iterator does not leak anything. + drop(hm.clone().into_par_iter()); + + { + assert_eq!(key.load(Ordering::Relaxed), 100); + assert_eq!(value.load(Ordering::Relaxed), 100); + + // retain only half + let _v: Vec<_> = hm.into_par_iter().filter(|(key, _)| key.k < 50).collect(); + + assert_eq!(key.load(Ordering::Relaxed), 50); + assert_eq!(value.load(Ordering::Relaxed), 50); + }; + + assert_eq!(key.load(Ordering::Relaxed), 0); + assert_eq!(value.load(Ordering::Relaxed), 0); + } + + #[test] + fn test_drain_drops() { + let key = AtomicUsize::new(0); + let value = AtomicUsize::new(0); + + let mut hm = { + let mut hm = HashMap::new(); + + assert_eq!(key.load(Ordering::Relaxed), 0); + assert_eq!(value.load(Ordering::Relaxed), 0); + + for i in 0..100 { + let d1 = Dropable::new(i, &key); + let d2 = Dropable::new(i + 100, &value); + hm.insert(d1, d2); + } + + assert_eq!(key.load(Ordering::Relaxed), 100); + assert_eq!(value.load(Ordering::Relaxed), 100); + + hm + }; + + // By the way, ensure that cloning doesn't screw up the dropping. + drop(hm.clone()); + + assert_eq!(key.load(Ordering::Relaxed), 100); + assert_eq!(value.load(Ordering::Relaxed), 100); + + // Ensure that dropping the drain iterator does not leak anything. + drop(hm.clone().par_drain()); + + { + assert_eq!(key.load(Ordering::Relaxed), 100); + assert_eq!(value.load(Ordering::Relaxed), 100); + + // retain only half + let _v: Vec<_> = hm.drain().filter(|(key, _)| key.k < 50).collect(); + assert!(hm.is_empty()); + + assert_eq!(key.load(Ordering::Relaxed), 50); + assert_eq!(value.load(Ordering::Relaxed), 50); + }; + + assert_eq!(key.load(Ordering::Relaxed), 0); + assert_eq!(value.load(Ordering::Relaxed), 0); + } + + #[test] + fn test_empty_iter() { + let mut m: HashMap<isize, bool> = HashMap::new(); + assert_eq!(m.par_drain().count(), 0); + assert_eq!(m.par_keys().count(), 0); + assert_eq!(m.par_values().count(), 0); + assert_eq!(m.par_values_mut().count(), 0); + assert_eq!(m.par_iter().count(), 0); + assert_eq!(m.par_iter_mut().count(), 0); + assert_eq!(m.len(), 0); + assert!(m.is_empty()); + assert_eq!(m.into_par_iter().count(), 0); + } + + #[test] + fn test_iterate() { + let mut m = HashMap::with_capacity(4); + for i in 0..32 { + assert!(m.insert(i, i * 2).is_none()); + } + assert_eq!(m.len(), 32); + + let observed = AtomicUsize::new(0); + + m.par_iter().for_each(|(k, v)| { + assert_eq!(*v, *k * 2); + observed.fetch_or(1 << *k, Ordering::Relaxed); + }); + assert_eq!(observed.into_inner(), 0xFFFF_FFFF); + } + + #[test] + fn test_keys() { + let vec = vec![(1, 'a'), (2, 'b'), (3, 'c')]; + let map: HashMap<_, _> = vec.into_par_iter().collect(); + let keys: Vec<_> = map.par_keys().cloned().collect(); + assert_eq!(keys.len(), 3); + assert!(keys.contains(&1)); + assert!(keys.contains(&2)); + assert!(keys.contains(&3)); + } + + #[test] + fn test_values() { + let vec = vec![(1, 'a'), (2, 'b'), (3, 'c')]; + let map: HashMap<_, _> = vec.into_par_iter().collect(); + let values: Vec<_> = map.par_values().cloned().collect(); + assert_eq!(values.len(), 3); + assert!(values.contains(&'a')); + assert!(values.contains(&'b')); + assert!(values.contains(&'c')); + } + + #[test] + fn test_values_mut() { + let vec = vec![(1, 1), (2, 2), (3, 3)]; + let mut map: HashMap<_, _> = vec.into_par_iter().collect(); + map.par_values_mut().for_each(|value| *value *= 2); + let values: Vec<_> = map.par_values().cloned().collect(); + assert_eq!(values.len(), 3); + assert!(values.contains(&2)); + assert!(values.contains(&4)); + assert!(values.contains(&6)); + } + + #[test] + fn test_eq() { + let mut m1 = HashMap::new(); + m1.insert(1, 2); + m1.insert(2, 3); + m1.insert(3, 4); + + let mut m2 = HashMap::new(); + m2.insert(1, 2); + m2.insert(2, 3); + + assert!(!m1.par_eq(&m2)); + + m2.insert(3, 4); + + assert!(m1.par_eq(&m2)); + } + + #[test] + fn test_from_iter() { + let xs = [(1, 1), (2, 2), (3, 3), (4, 4), (5, 5), (6, 6)]; + + let map: HashMap<_, _> = xs.par_iter().cloned().collect(); + + for &(k, v) in &xs { + assert_eq!(map.get(&k), Some(&v)); + } + } + + #[test] + fn test_extend_ref() { + let mut a = HashMap::new(); + a.insert(1, "one"); + let mut b = HashMap::new(); + b.insert(2, "two"); + b.insert(3, "three"); + + a.par_extend(&b); + + assert_eq!(a.len(), 3); + assert_eq!(a[&1], "one"); + assert_eq!(a[&2], "two"); + assert_eq!(a[&3], "three"); + } +} diff --git a/vendor/hashbrown/src/external_trait_impls/rayon/mod.rs b/vendor/hashbrown/src/external_trait_impls/rayon/mod.rs new file mode 100644 index 0000000..99337a1 --- /dev/null +++ b/vendor/hashbrown/src/external_trait_impls/rayon/mod.rs @@ -0,0 +1,4 @@ +mod helpers; +pub(crate) mod map; +pub(crate) mod raw; +pub(crate) mod set; diff --git a/vendor/hashbrown/src/external_trait_impls/rayon/raw.rs b/vendor/hashbrown/src/external_trait_impls/rayon/raw.rs new file mode 100644 index 0000000..612be47 --- /dev/null +++ b/vendor/hashbrown/src/external_trait_impls/rayon/raw.rs @@ -0,0 +1,230 @@ +use crate::raw::Bucket; +use crate::raw::{Allocator, Global, RawIter, RawIterRange, RawTable}; +use crate::scopeguard::guard; +use core::marker::PhantomData; +use core::mem; +use core::ptr::NonNull; +use rayon::iter::{ + plumbing::{self, Folder, UnindexedConsumer, UnindexedProducer}, + ParallelIterator, +}; + +/// Parallel iterator which returns a raw pointer to every full bucket in the table. +pub struct RawParIter<T> { + iter: RawIterRange<T>, +} + +impl<T> RawParIter<T> { + #[cfg_attr(feature = "inline-more", inline)] + pub(super) unsafe fn iter(&self) -> RawIterRange<T> { + self.iter.clone() + } +} + +impl<T> Clone for RawParIter<T> { + #[cfg_attr(feature = "inline-more", inline)] + fn clone(&self) -> Self { + Self { + iter: self.iter.clone(), + } + } +} + +impl<T> From<RawIter<T>> for RawParIter<T> { + fn from(it: RawIter<T>) -> Self { + RawParIter { iter: it.iter } + } +} + +impl<T> ParallelIterator for RawParIter<T> { + type Item = Bucket<T>; + + #[cfg_attr(feature = "inline-more", inline)] + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: UnindexedConsumer<Self::Item>, + { + let producer = ParIterProducer { iter: self.iter }; + plumbing::bridge_unindexed(producer, consumer) + } +} + +/// Producer which returns a `Bucket<T>` for every element. +struct ParIterProducer<T> { + iter: RawIterRange<T>, +} + +impl<T> UnindexedProducer for ParIterProducer<T> { + type Item = Bucket<T>; + + #[cfg_attr(feature = "inline-more", inline)] + fn split(self) -> (Self, Option<Self>) { + let (left, right) = self.iter.split(); + let left = ParIterProducer { iter: left }; + let right = right.map(|right| ParIterProducer { iter: right }); + (left, right) + } + + #[cfg_attr(feature = "inline-more", inline)] + fn fold_with<F>(self, folder: F) -> F + where + F: Folder<Self::Item>, + { + folder.consume_iter(self.iter) + } +} + +/// Parallel iterator which consumes a table and returns elements. +pub struct RawIntoParIter<T, A: Allocator = Global> { + table: RawTable<T, A>, +} + +impl<T, A: Allocator> RawIntoParIter<T, A> { + #[cfg_attr(feature = "inline-more", inline)] + pub(super) unsafe fn par_iter(&self) -> RawParIter<T> { + self.table.par_iter() + } +} + +impl<T: Send, A: Allocator + Send> ParallelIterator for RawIntoParIter<T, A> { + type Item = T; + + #[cfg_attr(feature = "inline-more", inline)] + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: UnindexedConsumer<Self::Item>, + { + let iter = unsafe { self.table.iter().iter }; + let _guard = guard(self.table.into_allocation(), |alloc| { + if let Some((ptr, layout, ref alloc)) = *alloc { + unsafe { + alloc.deallocate(ptr, layout); + } + } + }); + let producer = ParDrainProducer { iter }; + plumbing::bridge_unindexed(producer, consumer) + } +} + +/// Parallel iterator which consumes elements without freeing the table storage. +pub struct RawParDrain<'a, T, A: Allocator = Global> { + // We don't use a &'a mut RawTable<T> because we want RawParDrain to be + // covariant over T. + table: NonNull<RawTable<T, A>>, + marker: PhantomData<&'a RawTable<T, A>>, +} + +unsafe impl<T: Send, A: Allocator> Send for RawParDrain<'_, T, A> {} + +impl<T, A: Allocator> RawParDrain<'_, T, A> { + #[cfg_attr(feature = "inline-more", inline)] + pub(super) unsafe fn par_iter(&self) -> RawParIter<T> { + self.table.as_ref().par_iter() + } +} + +impl<T: Send, A: Allocator> ParallelIterator for RawParDrain<'_, T, A> { + type Item = T; + + #[cfg_attr(feature = "inline-more", inline)] + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: UnindexedConsumer<Self::Item>, + { + let _guard = guard(self.table, |table| unsafe { + table.as_mut().clear_no_drop(); + }); + let iter = unsafe { self.table.as_ref().iter().iter }; + mem::forget(self); + let producer = ParDrainProducer { iter }; + plumbing::bridge_unindexed(producer, consumer) + } +} + +impl<T, A: Allocator> Drop for RawParDrain<'_, T, A> { + fn drop(&mut self) { + // If drive_unindexed is not called then simply clear the table. + unsafe { + self.table.as_mut().clear(); + } + } +} + +/// Producer which will consume all elements in the range, even if it is dropped +/// halfway through. +struct ParDrainProducer<T> { + iter: RawIterRange<T>, +} + +impl<T: Send> UnindexedProducer for ParDrainProducer<T> { + type Item = T; + + #[cfg_attr(feature = "inline-more", inline)] + fn split(self) -> (Self, Option<Self>) { + let (left, right) = self.iter.clone().split(); + mem::forget(self); + let left = ParDrainProducer { iter: left }; + let right = right.map(|right| ParDrainProducer { iter: right }); + (left, right) + } + + #[cfg_attr(feature = "inline-more", inline)] + fn fold_with<F>(mut self, mut folder: F) -> F + where + F: Folder<Self::Item>, + { + // Make sure to modify the iterator in-place so that any remaining + // elements are processed in our Drop impl. + for item in &mut self.iter { + folder = folder.consume(unsafe { item.read() }); + if folder.full() { + return folder; + } + } + + // If we processed all elements then we don't need to run the drop. + mem::forget(self); + folder + } +} + +impl<T> Drop for ParDrainProducer<T> { + #[cfg_attr(feature = "inline-more", inline)] + fn drop(&mut self) { + // Drop all remaining elements + if mem::needs_drop::<T>() { + for item in &mut self.iter { + unsafe { + item.drop(); + } + } + } + } +} + +impl<T, A: Allocator> RawTable<T, A> { + /// Returns a parallel iterator over the elements in a `RawTable`. + #[cfg_attr(feature = "inline-more", inline)] + pub unsafe fn par_iter(&self) -> RawParIter<T> { + RawParIter { + iter: self.iter().iter, + } + } + + /// Returns a parallel iterator over the elements in a `RawTable`. + #[cfg_attr(feature = "inline-more", inline)] + pub fn into_par_iter(self) -> RawIntoParIter<T, A> { + RawIntoParIter { table: self } + } + + /// Returns a parallel iterator which consumes all elements of a `RawTable` + /// without freeing its memory allocation. + #[cfg_attr(feature = "inline-more", inline)] + pub fn par_drain(&mut self) -> RawParDrain<'_, T, A> { + RawParDrain { + table: NonNull::from(self), + marker: PhantomData, + } + } +} diff --git a/vendor/hashbrown/src/external_trait_impls/rayon/set.rs b/vendor/hashbrown/src/external_trait_impls/rayon/set.rs new file mode 100644 index 0000000..3de98fc --- /dev/null +++ b/vendor/hashbrown/src/external_trait_impls/rayon/set.rs @@ -0,0 +1,659 @@ +//! Rayon extensions for `HashSet`. + +use super::map; +use crate::hash_set::HashSet; +use crate::raw::{Allocator, Global}; +use core::hash::{BuildHasher, Hash}; +use rayon::iter::plumbing::UnindexedConsumer; +use rayon::iter::{FromParallelIterator, IntoParallelIterator, ParallelExtend, ParallelIterator}; + +/// Parallel iterator over elements of a consumed set. +/// +/// This iterator is created by the [`into_par_iter`] method on [`HashSet`] +/// (provided by the [`IntoParallelIterator`] trait). +/// See its documentation for more. +/// +/// [`into_par_iter`]: /hashbrown/struct.HashSet.html#method.into_par_iter +/// [`HashSet`]: /hashbrown/struct.HashSet.html +/// [`IntoParallelIterator`]: https://docs.rs/rayon/1.0/rayon/iter/trait.IntoParallelIterator.html +pub struct IntoParIter<T, A: Allocator = Global> { + inner: map::IntoParIter<T, (), A>, +} + +impl<T: Send, A: Allocator + Send> ParallelIterator for IntoParIter<T, A> { + type Item = T; + + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: UnindexedConsumer<Self::Item>, + { + self.inner.map(|(k, _)| k).drive_unindexed(consumer) + } +} + +/// Parallel draining iterator over entries of a set. +/// +/// This iterator is created by the [`par_drain`] method on [`HashSet`]. +/// See its documentation for more. +/// +/// [`par_drain`]: /hashbrown/struct.HashSet.html#method.par_drain +/// [`HashSet`]: /hashbrown/struct.HashSet.html +pub struct ParDrain<'a, T, A: Allocator = Global> { + inner: map::ParDrain<'a, T, (), A>, +} + +impl<T: Send, A: Allocator + Send + Sync> ParallelIterator for ParDrain<'_, T, A> { + type Item = T; + + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: UnindexedConsumer<Self::Item>, + { + self.inner.map(|(k, _)| k).drive_unindexed(consumer) + } +} + +/// Parallel iterator over shared references to elements in a set. +/// +/// This iterator is created by the [`par_iter`] method on [`HashSet`] +/// (provided by the [`IntoParallelRefIterator`] trait). +/// See its documentation for more. +/// +/// [`par_iter`]: /hashbrown/struct.HashSet.html#method.par_iter +/// [`HashSet`]: /hashbrown/struct.HashSet.html +/// [`IntoParallelRefIterator`]: https://docs.rs/rayon/1.0/rayon/iter/trait.IntoParallelRefIterator.html +pub struct ParIter<'a, T> { + inner: map::ParKeys<'a, T, ()>, +} + +impl<'a, T: Sync> ParallelIterator for ParIter<'a, T> { + type Item = &'a T; + + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: UnindexedConsumer<Self::Item>, + { + self.inner.drive_unindexed(consumer) + } +} + +/// Parallel iterator over shared references to elements in the difference of +/// sets. +/// +/// This iterator is created by the [`par_difference`] method on [`HashSet`]. +/// See its documentation for more. +/// +/// [`par_difference`]: /hashbrown/struct.HashSet.html#method.par_difference +/// [`HashSet`]: /hashbrown/struct.HashSet.html +pub struct ParDifference<'a, T, S, A: Allocator = Global> { + a: &'a HashSet<T, S, A>, + b: &'a HashSet<T, S, A>, +} + +impl<'a, T, S, A> ParallelIterator for ParDifference<'a, T, S, A> +where + T: Eq + Hash + Sync, + S: BuildHasher + Sync, + A: Allocator + Sync, +{ + type Item = &'a T; + + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: UnindexedConsumer<Self::Item>, + { + self.a + .into_par_iter() + .filter(|&x| !self.b.contains(x)) + .drive_unindexed(consumer) + } +} + +/// Parallel iterator over shared references to elements in the symmetric +/// difference of sets. +/// +/// This iterator is created by the [`par_symmetric_difference`] method on +/// [`HashSet`]. +/// See its documentation for more. +/// +/// [`par_symmetric_difference`]: /hashbrown/struct.HashSet.html#method.par_symmetric_difference +/// [`HashSet`]: /hashbrown/struct.HashSet.html +pub struct ParSymmetricDifference<'a, T, S, A: Allocator = Global> { + a: &'a HashSet<T, S, A>, + b: &'a HashSet<T, S, A>, +} + +impl<'a, T, S, A> ParallelIterator for ParSymmetricDifference<'a, T, S, A> +where + T: Eq + Hash + Sync, + S: BuildHasher + Sync, + A: Allocator + Sync, +{ + type Item = &'a T; + + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: UnindexedConsumer<Self::Item>, + { + self.a + .par_difference(self.b) + .chain(self.b.par_difference(self.a)) + .drive_unindexed(consumer) + } +} + +/// Parallel iterator over shared references to elements in the intersection of +/// sets. +/// +/// This iterator is created by the [`par_intersection`] method on [`HashSet`]. +/// See its documentation for more. +/// +/// [`par_intersection`]: /hashbrown/struct.HashSet.html#method.par_intersection +/// [`HashSet`]: /hashbrown/struct.HashSet.html +pub struct ParIntersection<'a, T, S, A: Allocator = Global> { + a: &'a HashSet<T, S, A>, + b: &'a HashSet<T, S, A>, +} + +impl<'a, T, S, A> ParallelIterator for ParIntersection<'a, T, S, A> +where + T: Eq + Hash + Sync, + S: BuildHasher + Sync, + A: Allocator + Sync, +{ + type Item = &'a T; + + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: UnindexedConsumer<Self::Item>, + { + self.a + .into_par_iter() + .filter(|&x| self.b.contains(x)) + .drive_unindexed(consumer) + } +} + +/// Parallel iterator over shared references to elements in the union of sets. +/// +/// This iterator is created by the [`par_union`] method on [`HashSet`]. +/// See its documentation for more. +/// +/// [`par_union`]: /hashbrown/struct.HashSet.html#method.par_union +/// [`HashSet`]: /hashbrown/struct.HashSet.html +pub struct ParUnion<'a, T, S, A: Allocator = Global> { + a: &'a HashSet<T, S, A>, + b: &'a HashSet<T, S, A>, +} + +impl<'a, T, S, A> ParallelIterator for ParUnion<'a, T, S, A> +where + T: Eq + Hash + Sync, + S: BuildHasher + Sync, + A: Allocator + Sync, +{ + type Item = &'a T; + + fn drive_unindexed<C>(self, consumer: C) -> C::Result + where + C: UnindexedConsumer<Self::Item>, + { + // We'll iterate one set in full, and only the remaining difference from the other. + // Use the smaller set for the difference in order to reduce hash lookups. + let (smaller, larger) = if self.a.len() <= self.b.len() { + (self.a, self.b) + } else { + (self.b, self.a) + }; + larger + .into_par_iter() + .chain(smaller.par_difference(larger)) + .drive_unindexed(consumer) + } +} + +impl<T, S, A> HashSet<T, S, A> +where + T: Eq + Hash + Sync, + S: BuildHasher + Sync, + A: Allocator + Sync, +{ + /// Visits (potentially in parallel) the values representing the union, + /// i.e. all the values in `self` or `other`, without duplicates. + #[cfg_attr(feature = "inline-more", inline)] + pub fn par_union<'a>(&'a self, other: &'a Self) -> ParUnion<'a, T, S, A> { + ParUnion { a: self, b: other } + } + + /// Visits (potentially in parallel) the values representing the difference, + /// i.e. the values that are in `self` but not in `other`. + #[cfg_attr(feature = "inline-more", inline)] + pub fn par_difference<'a>(&'a self, other: &'a Self) -> ParDifference<'a, T, S, A> { + ParDifference { a: self, b: other } + } + + /// Visits (potentially in parallel) the values representing the symmetric + /// difference, i.e. the values that are in `self` or in `other` but not in both. + #[cfg_attr(feature = "inline-more", inline)] + pub fn par_symmetric_difference<'a>( + &'a self, + other: &'a Self, + ) -> ParSymmetricDifference<'a, T, S, A> { + ParSymmetricDifference { a: self, b: other } + } + + /// Visits (potentially in parallel) the values representing the + /// intersection, i.e. the values that are both in `self` and `other`. + #[cfg_attr(feature = "inline-more", inline)] + pub fn par_intersection<'a>(&'a self, other: &'a Self) -> ParIntersection<'a, T, S, A> { + ParIntersection { a: self, b: other } + } + + /// Returns `true` if `self` has no elements in common with `other`. + /// This is equivalent to checking for an empty intersection. + /// + /// This method runs in a potentially parallel fashion. + pub fn par_is_disjoint(&self, other: &Self) -> bool { + self.into_par_iter().all(|x| !other.contains(x)) + } + + /// Returns `true` if the set is a subset of another, + /// i.e. `other` contains at least all the values in `self`. + /// + /// This method runs in a potentially parallel fashion. + pub fn par_is_subset(&self, other: &Self) -> bool { + if self.len() <= other.len() { + self.into_par_iter().all(|x| other.contains(x)) + } else { + false + } + } + + /// Returns `true` if the set is a superset of another, + /// i.e. `self` contains at least all the values in `other`. + /// + /// This method runs in a potentially parallel fashion. + pub fn par_is_superset(&self, other: &Self) -> bool { + other.par_is_subset(self) + } + + /// Returns `true` if the set is equal to another, + /// i.e. both sets contain the same values. + /// + /// This method runs in a potentially parallel fashion. + pub fn par_eq(&self, other: &Self) -> bool { + self.len() == other.len() && self.par_is_subset(other) + } +} + +impl<T, S, A> HashSet<T, S, A> +where + T: Eq + Hash + Send, + A: Allocator + Send, +{ + /// Consumes (potentially in parallel) all values in an arbitrary order, + /// while preserving the set's allocated memory for reuse. + #[cfg_attr(feature = "inline-more", inline)] + pub fn par_drain(&mut self) -> ParDrain<'_, T, A> { + ParDrain { + inner: self.map.par_drain(), + } + } +} + +impl<T: Send, S, A: Allocator + Send> IntoParallelIterator for HashSet<T, S, A> { + type Item = T; + type Iter = IntoParIter<T, A>; + + #[cfg_attr(feature = "inline-more", inline)] + fn into_par_iter(self) -> Self::Iter { + IntoParIter { + inner: self.map.into_par_iter(), + } + } +} + +impl<'a, T: Sync, S, A: Allocator> IntoParallelIterator for &'a HashSet<T, S, A> { + type Item = &'a T; + type Iter = ParIter<'a, T>; + + #[cfg_attr(feature = "inline-more", inline)] + fn into_par_iter(self) -> Self::Iter { + ParIter { + inner: self.map.par_keys(), + } + } +} + +/// Collect values from a parallel iterator into a hashset. +impl<T, S> FromParallelIterator<T> for HashSet<T, S, Global> +where + T: Eq + Hash + Send, + S: BuildHasher + Default, +{ + fn from_par_iter<P>(par_iter: P) -> Self + where + P: IntoParallelIterator<Item = T>, + { + let mut set = HashSet::default(); + set.par_extend(par_iter); + set + } +} + +/// Extend a hash set with items from a parallel iterator. +impl<T, S> ParallelExtend<T> for HashSet<T, S, Global> +where + T: Eq + Hash + Send, + S: BuildHasher, +{ + fn par_extend<I>(&mut self, par_iter: I) + where + I: IntoParallelIterator<Item = T>, + { + extend(self, par_iter); + } +} + +/// Extend a hash set with copied items from a parallel iterator. +impl<'a, T, S> ParallelExtend<&'a T> for HashSet<T, S, Global> +where + T: 'a + Copy + Eq + Hash + Sync, + S: BuildHasher, +{ + fn par_extend<I>(&mut self, par_iter: I) + where + I: IntoParallelIterator<Item = &'a T>, + { + extend(self, par_iter); + } +} + +// This is equal to the normal `HashSet` -- no custom advantage. +fn extend<T, S, I, A>(set: &mut HashSet<T, S, A>, par_iter: I) +where + T: Eq + Hash, + S: BuildHasher, + A: Allocator, + I: IntoParallelIterator, + HashSet<T, S, A>: Extend<I::Item>, +{ + let (list, len) = super::helpers::collect(par_iter); + + // Values may be already present or show multiple times in the iterator. + // Reserve the entire length if the set is empty. + // Otherwise reserve half the length (rounded up), so the set + // will only resize twice in the worst case. + let reserve = if set.is_empty() { len } else { (len + 1) / 2 }; + set.reserve(reserve); + for vec in list { + set.extend(vec); + } +} + +#[cfg(test)] +mod test_par_set { + use alloc::vec::Vec; + use core::sync::atomic::{AtomicUsize, Ordering}; + + use rayon::prelude::*; + + use crate::hash_set::HashSet; + + #[test] + fn test_disjoint() { + let mut xs = HashSet::new(); + let mut ys = HashSet::new(); + assert!(xs.par_is_disjoint(&ys)); + assert!(ys.par_is_disjoint(&xs)); + assert!(xs.insert(5)); + assert!(ys.insert(11)); + assert!(xs.par_is_disjoint(&ys)); + assert!(ys.par_is_disjoint(&xs)); + assert!(xs.insert(7)); + assert!(xs.insert(19)); + assert!(xs.insert(4)); + assert!(ys.insert(2)); + assert!(ys.insert(-11)); + assert!(xs.par_is_disjoint(&ys)); + assert!(ys.par_is_disjoint(&xs)); + assert!(ys.insert(7)); + assert!(!xs.par_is_disjoint(&ys)); + assert!(!ys.par_is_disjoint(&xs)); + } + + #[test] + fn test_subset_and_superset() { + let mut a = HashSet::new(); + assert!(a.insert(0)); + assert!(a.insert(5)); + assert!(a.insert(11)); + assert!(a.insert(7)); + + let mut b = HashSet::new(); + assert!(b.insert(0)); + assert!(b.insert(7)); + assert!(b.insert(19)); + assert!(b.insert(250)); + assert!(b.insert(11)); + assert!(b.insert(200)); + + assert!(!a.par_is_subset(&b)); + assert!(!a.par_is_superset(&b)); + assert!(!b.par_is_subset(&a)); + assert!(!b.par_is_superset(&a)); + + assert!(b.insert(5)); + + assert!(a.par_is_subset(&b)); + assert!(!a.par_is_superset(&b)); + assert!(!b.par_is_subset(&a)); + assert!(b.par_is_superset(&a)); + } + + #[test] + fn test_iterate() { + let mut a = HashSet::new(); + for i in 0..32 { + assert!(a.insert(i)); + } + let observed = AtomicUsize::new(0); + a.par_iter().for_each(|k| { + observed.fetch_or(1 << *k, Ordering::Relaxed); + }); + assert_eq!(observed.into_inner(), 0xFFFF_FFFF); + } + + #[test] + fn test_intersection() { + let mut a = HashSet::new(); + let mut b = HashSet::new(); + + assert!(a.insert(11)); + assert!(a.insert(1)); + assert!(a.insert(3)); + assert!(a.insert(77)); + assert!(a.insert(103)); + assert!(a.insert(5)); + assert!(a.insert(-5)); + + assert!(b.insert(2)); + assert!(b.insert(11)); + assert!(b.insert(77)); + assert!(b.insert(-9)); + assert!(b.insert(-42)); + assert!(b.insert(5)); + assert!(b.insert(3)); + + let expected = [3, 5, 11, 77]; + let i = a + .par_intersection(&b) + .map(|x| { + assert!(expected.contains(x)); + 1 + }) + .sum::<usize>(); + assert_eq!(i, expected.len()); + } + + #[test] + fn test_difference() { + let mut a = HashSet::new(); + let mut b = HashSet::new(); + + assert!(a.insert(1)); + assert!(a.insert(3)); + assert!(a.insert(5)); + assert!(a.insert(9)); + assert!(a.insert(11)); + + assert!(b.insert(3)); + assert!(b.insert(9)); + + let expected = [1, 5, 11]; + let i = a + .par_difference(&b) + .map(|x| { + assert!(expected.contains(x)); + 1 + }) + .sum::<usize>(); + assert_eq!(i, expected.len()); + } + + #[test] + fn test_symmetric_difference() { + let mut a = HashSet::new(); + let mut b = HashSet::new(); + + assert!(a.insert(1)); + assert!(a.insert(3)); + assert!(a.insert(5)); + assert!(a.insert(9)); + assert!(a.insert(11)); + + assert!(b.insert(-2)); + assert!(b.insert(3)); + assert!(b.insert(9)); + assert!(b.insert(14)); + assert!(b.insert(22)); + + let expected = [-2, 1, 5, 11, 14, 22]; + let i = a + .par_symmetric_difference(&b) + .map(|x| { + assert!(expected.contains(x)); + 1 + }) + .sum::<usize>(); + assert_eq!(i, expected.len()); + } + + #[test] + fn test_union() { + let mut a = HashSet::new(); + let mut b = HashSet::new(); + + assert!(a.insert(1)); + assert!(a.insert(3)); + assert!(a.insert(5)); + assert!(a.insert(9)); + assert!(a.insert(11)); + assert!(a.insert(16)); + assert!(a.insert(19)); + assert!(a.insert(24)); + + assert!(b.insert(-2)); + assert!(b.insert(1)); + assert!(b.insert(5)); + assert!(b.insert(9)); + assert!(b.insert(13)); + assert!(b.insert(19)); + + let expected = [-2, 1, 3, 5, 9, 11, 13, 16, 19, 24]; + let i = a + .par_union(&b) + .map(|x| { + assert!(expected.contains(x)); + 1 + }) + .sum::<usize>(); + assert_eq!(i, expected.len()); + } + + #[test] + fn test_from_iter() { + let xs = [1, 2, 3, 4, 5, 6, 7, 8, 9]; + + let set: HashSet<_> = xs.par_iter().cloned().collect(); + + for x in &xs { + assert!(set.contains(x)); + } + } + + #[test] + fn test_move_iter() { + let hs = { + let mut hs = HashSet::new(); + + hs.insert('a'); + hs.insert('b'); + + hs + }; + + let v = hs.into_par_iter().collect::<Vec<char>>(); + assert!(v == ['a', 'b'] || v == ['b', 'a']); + } + + #[test] + fn test_eq() { + // These constants once happened to expose a bug in insert(). + // I'm keeping them around to prevent a regression. + let mut s1 = HashSet::new(); + + s1.insert(1); + s1.insert(2); + s1.insert(3); + + let mut s2 = HashSet::new(); + + s2.insert(1); + s2.insert(2); + + assert!(!s1.par_eq(&s2)); + + s2.insert(3); + + assert!(s1.par_eq(&s2)); + } + + #[test] + fn test_extend_ref() { + let mut a = HashSet::new(); + a.insert(1); + + a.par_extend(&[2, 3, 4][..]); + + assert_eq!(a.len(), 4); + assert!(a.contains(&1)); + assert!(a.contains(&2)); + assert!(a.contains(&3)); + assert!(a.contains(&4)); + + let mut b = HashSet::new(); + b.insert(5); + b.insert(6); + + a.par_extend(&b); + + assert_eq!(a.len(), 6); + assert!(a.contains(&1)); + assert!(a.contains(&2)); + assert!(a.contains(&3)); + assert!(a.contains(&4)); + assert!(a.contains(&5)); + assert!(a.contains(&6)); + } +} diff --git a/vendor/hashbrown/src/external_trait_impls/rkyv/hash_map.rs b/vendor/hashbrown/src/external_trait_impls/rkyv/hash_map.rs new file mode 100644 index 0000000..fae7f76 --- /dev/null +++ b/vendor/hashbrown/src/external_trait_impls/rkyv/hash_map.rs @@ -0,0 +1,125 @@ +use crate::HashMap; +use core::{ + borrow::Borrow, + hash::{BuildHasher, Hash}, +}; +use rkyv::{ + collections::hash_map::{ArchivedHashMap, HashMapResolver}, + ser::{ScratchSpace, Serializer}, + Archive, Deserialize, Fallible, Serialize, +}; + +impl<K: Archive + Hash + Eq, V: Archive, S> Archive for HashMap<K, V, S> +where + K::Archived: Hash + Eq, +{ + type Archived = ArchivedHashMap<K::Archived, V::Archived>; + type Resolver = HashMapResolver; + + #[inline] + unsafe fn resolve(&self, pos: usize, resolver: Self::Resolver, out: *mut Self::Archived) { + ArchivedHashMap::resolve_from_len(self.len(), pos, resolver, out); + } +} + +impl<K, V, S, RandomState> Serialize<S> for HashMap<K, V, RandomState> +where + K: Serialize<S> + Hash + Eq, + K::Archived: Hash + Eq, + V: Serialize<S>, + S: Serializer + ScratchSpace + ?Sized, +{ + #[inline] + fn serialize(&self, serializer: &mut S) -> Result<Self::Resolver, S::Error> { + unsafe { ArchivedHashMap::serialize_from_iter(self.iter(), serializer) } + } +} + +impl<K: Archive + Hash + Eq, V: Archive, D: Fallible + ?Sized, S: Default + BuildHasher> + Deserialize<HashMap<K, V, S>, D> for ArchivedHashMap<K::Archived, V::Archived> +where + K::Archived: Deserialize<K, D> + Hash + Eq, + V::Archived: Deserialize<V, D>, +{ + #[inline] + fn deserialize(&self, deserializer: &mut D) -> Result<HashMap<K, V, S>, D::Error> { + let mut result = HashMap::with_capacity_and_hasher(self.len(), S::default()); + for (k, v) in self.iter() { + result.insert(k.deserialize(deserializer)?, v.deserialize(deserializer)?); + } + Ok(result) + } +} + +impl<K: Hash + Eq + Borrow<AK>, V, AK: Hash + Eq, AV: PartialEq<V>, S: BuildHasher> + PartialEq<HashMap<K, V, S>> for ArchivedHashMap<AK, AV> +{ + #[inline] + fn eq(&self, other: &HashMap<K, V, S>) -> bool { + if self.len() != other.len() { + false + } else { + self.iter() + .all(|(key, value)| other.get(key).map_or(false, |v| value.eq(v))) + } + } +} + +impl<K: Hash + Eq + Borrow<AK>, V, AK: Hash + Eq, AV: PartialEq<V>> + PartialEq<ArchivedHashMap<AK, AV>> for HashMap<K, V> +{ + #[inline] + fn eq(&self, other: &ArchivedHashMap<AK, AV>) -> bool { + other.eq(self) + } +} + +#[cfg(test)] +mod tests { + use crate::HashMap; + use alloc::string::String; + use rkyv::{ + archived_root, check_archived_root, + ser::{serializers::AllocSerializer, Serializer}, + Deserialize, Infallible, + }; + + #[test] + fn index_map() { + let mut value = HashMap::new(); + value.insert(String::from("foo"), 10); + value.insert(String::from("bar"), 20); + value.insert(String::from("baz"), 40); + value.insert(String::from("bat"), 80); + + let mut serializer = AllocSerializer::<4096>::default(); + serializer.serialize_value(&value).unwrap(); + let result = serializer.into_serializer().into_inner(); + let archived = unsafe { archived_root::<HashMap<String, i32>>(result.as_ref()) }; + + assert_eq!(value.len(), archived.len()); + for (k, v) in value.iter() { + let (ak, av) = archived.get_key_value(k.as_str()).unwrap(); + assert_eq!(k, ak); + assert_eq!(v, av); + } + + let deserialized: HashMap<String, i32> = archived.deserialize(&mut Infallible).unwrap(); + assert_eq!(value, deserialized); + } + + #[test] + fn validate_index_map() { + let mut value = HashMap::new(); + value.insert(String::from("foo"), 10); + value.insert(String::from("bar"), 20); + value.insert(String::from("baz"), 40); + value.insert(String::from("bat"), 80); + + let mut serializer = AllocSerializer::<4096>::default(); + serializer.serialize_value(&value).unwrap(); + let result = serializer.into_serializer().into_inner(); + check_archived_root::<HashMap<String, i32>>(result.as_ref()) + .expect("failed to validate archived index map"); + } +} diff --git a/vendor/hashbrown/src/external_trait_impls/rkyv/hash_set.rs b/vendor/hashbrown/src/external_trait_impls/rkyv/hash_set.rs new file mode 100644 index 0000000..c8a69cf --- /dev/null +++ b/vendor/hashbrown/src/external_trait_impls/rkyv/hash_set.rs @@ -0,0 +1,123 @@ +use crate::HashSet; +use core::{ + borrow::Borrow, + hash::{BuildHasher, Hash}, +}; +use rkyv::{ + collections::hash_set::{ArchivedHashSet, HashSetResolver}, + ser::{ScratchSpace, Serializer}, + Archive, Deserialize, Fallible, Serialize, +}; + +impl<K: Archive + Hash + Eq, S> Archive for HashSet<K, S> +where + K::Archived: Hash + Eq, +{ + type Archived = ArchivedHashSet<K::Archived>; + type Resolver = HashSetResolver; + + #[inline] + unsafe fn resolve(&self, pos: usize, resolver: Self::Resolver, out: *mut Self::Archived) { + ArchivedHashSet::<K::Archived>::resolve_from_len(self.len(), pos, resolver, out); + } +} + +impl<K, S, RS> Serialize<S> for HashSet<K, RS> +where + K::Archived: Hash + Eq, + K: Serialize<S> + Hash + Eq, + S: ScratchSpace + Serializer + ?Sized, +{ + #[inline] + fn serialize(&self, serializer: &mut S) -> Result<Self::Resolver, S::Error> { + unsafe { ArchivedHashSet::serialize_from_iter(self.iter(), serializer) } + } +} + +impl<K, D, S> Deserialize<HashSet<K, S>, D> for ArchivedHashSet<K::Archived> +where + K: Archive + Hash + Eq, + K::Archived: Deserialize<K, D> + Hash + Eq, + D: Fallible + ?Sized, + S: Default + BuildHasher, +{ + #[inline] + fn deserialize(&self, deserializer: &mut D) -> Result<HashSet<K, S>, D::Error> { + let mut result = HashSet::with_hasher(S::default()); + for k in self.iter() { + result.insert(k.deserialize(deserializer)?); + } + Ok(result) + } +} + +impl<K: Hash + Eq + Borrow<AK>, AK: Hash + Eq, S: BuildHasher> PartialEq<HashSet<K, S>> + for ArchivedHashSet<AK> +{ + #[inline] + fn eq(&self, other: &HashSet<K, S>) -> bool { + if self.len() != other.len() { + false + } else { + self.iter().all(|key| other.get(key).is_some()) + } + } +} + +impl<K: Hash + Eq + Borrow<AK>, AK: Hash + Eq, S: BuildHasher> PartialEq<ArchivedHashSet<AK>> + for HashSet<K, S> +{ + #[inline] + fn eq(&self, other: &ArchivedHashSet<AK>) -> bool { + other.eq(self) + } +} + +#[cfg(test)] +mod tests { + use crate::HashSet; + use alloc::string::String; + use rkyv::{ + archived_root, check_archived_root, + ser::{serializers::AllocSerializer, Serializer}, + Deserialize, Infallible, + }; + + #[test] + fn index_set() { + let mut value = HashSet::new(); + value.insert(String::from("foo")); + value.insert(String::from("bar")); + value.insert(String::from("baz")); + value.insert(String::from("bat")); + + let mut serializer = AllocSerializer::<4096>::default(); + serializer.serialize_value(&value).unwrap(); + let result = serializer.into_serializer().into_inner(); + let archived = unsafe { archived_root::<HashSet<String>>(result.as_ref()) }; + + assert_eq!(value.len(), archived.len()); + for k in value.iter() { + let ak = archived.get(k.as_str()).unwrap(); + assert_eq!(k, ak); + } + + let deserialized: HashSet<String> = archived.deserialize(&mut Infallible).unwrap(); + assert_eq!(value, deserialized); + } + + #[test] + fn validate_index_set() { + let mut value = HashSet::new(); + value.insert(String::from("foo")); + value.insert(String::from("bar")); + value.insert(String::from("baz")); + value.insert(String::from("bat")); + + let mut serializer = AllocSerializer::<4096>::default(); + serializer.serialize_value(&value).unwrap(); + let result = serializer.into_serializer().into_inner(); + check_archived_root::<HashSet<String>>(result.as_ref()) + .expect("failed to validate archived index set"); + } +} diff --git a/vendor/hashbrown/src/external_trait_impls/rkyv/mod.rs b/vendor/hashbrown/src/external_trait_impls/rkyv/mod.rs new file mode 100644 index 0000000..2bde6a0 --- /dev/null +++ b/vendor/hashbrown/src/external_trait_impls/rkyv/mod.rs @@ -0,0 +1,2 @@ +mod hash_map; +mod hash_set; diff --git a/vendor/hashbrown/src/external_trait_impls/serde.rs b/vendor/hashbrown/src/external_trait_impls/serde.rs new file mode 100644 index 0000000..0a76dbe --- /dev/null +++ b/vendor/hashbrown/src/external_trait_impls/serde.rs @@ -0,0 +1,220 @@ +mod size_hint { + use core::cmp; + + /// This presumably exists to prevent denial of service attacks. + /// + /// Original discussion: https://github.com/serde-rs/serde/issues/1114. + #[cfg_attr(feature = "inline-more", inline)] + pub(super) fn cautious(hint: Option<usize>) -> usize { + cmp::min(hint.unwrap_or(0), 4096) + } +} + +mod map { + use crate::raw::Allocator; + use core::fmt; + use core::hash::{BuildHasher, Hash}; + use core::marker::PhantomData; + use serde::de::{Deserialize, Deserializer, MapAccess, Visitor}; + use serde::ser::{Serialize, Serializer}; + + use crate::hash_map::HashMap; + + use super::size_hint; + + impl<K, V, H, A> Serialize for HashMap<K, V, H, A> + where + K: Serialize + Eq + Hash, + V: Serialize, + H: BuildHasher, + A: Allocator, + { + #[cfg_attr(feature = "inline-more", inline)] + fn serialize<S>(&self, serializer: S) -> Result<S::Ok, S::Error> + where + S: Serializer, + { + serializer.collect_map(self) + } + } + + impl<'de, K, V, S, A> Deserialize<'de> for HashMap<K, V, S, A> + where + K: Deserialize<'de> + Eq + Hash, + V: Deserialize<'de>, + S: BuildHasher + Default, + A: Allocator + Default, + { + fn deserialize<D>(deserializer: D) -> Result<Self, D::Error> + where + D: Deserializer<'de>, + { + struct MapVisitor<K, V, S, A> + where + A: Allocator, + { + marker: PhantomData<HashMap<K, V, S, A>>, + } + + impl<'de, K, V, S, A> Visitor<'de> for MapVisitor<K, V, S, A> + where + K: Deserialize<'de> + Eq + Hash, + V: Deserialize<'de>, + S: BuildHasher + Default, + A: Allocator + Default, + { + type Value = HashMap<K, V, S, A>; + + fn expecting(&self, formatter: &mut fmt::Formatter<'_>) -> fmt::Result { + formatter.write_str("a map") + } + + #[cfg_attr(feature = "inline-more", inline)] + fn visit_map<M>(self, mut map: M) -> Result<Self::Value, M::Error> + where + M: MapAccess<'de>, + { + let mut values = HashMap::with_capacity_and_hasher_in( + size_hint::cautious(map.size_hint()), + S::default(), + A::default(), + ); + + while let Some((key, value)) = map.next_entry()? { + values.insert(key, value); + } + + Ok(values) + } + } + + let visitor = MapVisitor { + marker: PhantomData, + }; + deserializer.deserialize_map(visitor) + } + } +} + +mod set { + use crate::raw::Allocator; + use core::fmt; + use core::hash::{BuildHasher, Hash}; + use core::marker::PhantomData; + use serde::de::{Deserialize, Deserializer, SeqAccess, Visitor}; + use serde::ser::{Serialize, Serializer}; + + use crate::hash_set::HashSet; + + use super::size_hint; + + impl<T, H, A> Serialize for HashSet<T, H, A> + where + T: Serialize + Eq + Hash, + H: BuildHasher, + A: Allocator, + { + #[cfg_attr(feature = "inline-more", inline)] + fn serialize<S>(&self, serializer: S) -> Result<S::Ok, S::Error> + where + S: Serializer, + { + serializer.collect_seq(self) + } + } + + impl<'de, T, S, A> Deserialize<'de> for HashSet<T, S, A> + where + T: Deserialize<'de> + Eq + Hash, + S: BuildHasher + Default, + A: Allocator + Default, + { + fn deserialize<D>(deserializer: D) -> Result<Self, D::Error> + where + D: Deserializer<'de>, + { + struct SeqVisitor<T, S, A> + where + A: Allocator, + { + marker: PhantomData<HashSet<T, S, A>>, + } + + impl<'de, T, S, A> Visitor<'de> for SeqVisitor<T, S, A> + where + T: Deserialize<'de> + Eq + Hash, + S: BuildHasher + Default, + A: Allocator + Default, + { + type Value = HashSet<T, S, A>; + + fn expecting(&self, formatter: &mut fmt::Formatter<'_>) -> fmt::Result { + formatter.write_str("a sequence") + } + + #[cfg_attr(feature = "inline-more", inline)] + fn visit_seq<M>(self, mut seq: M) -> Result<Self::Value, M::Error> + where + M: SeqAccess<'de>, + { + let mut values = HashSet::with_capacity_and_hasher_in( + size_hint::cautious(seq.size_hint()), + S::default(), + A::default(), + ); + + while let Some(value) = seq.next_element()? { + values.insert(value); + } + + Ok(values) + } + } + + let visitor = SeqVisitor { + marker: PhantomData, + }; + deserializer.deserialize_seq(visitor) + } + + #[allow(clippy::missing_errors_doc)] + fn deserialize_in_place<D>(deserializer: D, place: &mut Self) -> Result<(), D::Error> + where + D: Deserializer<'de>, + { + struct SeqInPlaceVisitor<'a, T, S, A>(&'a mut HashSet<T, S, A>) + where + A: Allocator; + + impl<'a, 'de, T, S, A> Visitor<'de> for SeqInPlaceVisitor<'a, T, S, A> + where + T: Deserialize<'de> + Eq + Hash, + S: BuildHasher + Default, + A: Allocator, + { + type Value = (); + + fn expecting(&self, formatter: &mut fmt::Formatter<'_>) -> fmt::Result { + formatter.write_str("a sequence") + } + + #[cfg_attr(feature = "inline-more", inline)] + fn visit_seq<M>(self, mut seq: M) -> Result<Self::Value, M::Error> + where + M: SeqAccess<'de>, + { + self.0.clear(); + self.0.reserve(size_hint::cautious(seq.size_hint())); + + while let Some(value) = seq.next_element()? { + self.0.insert(value); + } + + Ok(()) + } + } + + deserializer.deserialize_seq(SeqInPlaceVisitor(place)) + } + } +} diff --git a/vendor/hashbrown/src/lib.rs b/vendor/hashbrown/src/lib.rs new file mode 100644 index 0000000..06b1068 --- /dev/null +++ b/vendor/hashbrown/src/lib.rs @@ -0,0 +1,171 @@ +//! This crate is a Rust port of Google's high-performance [SwissTable] hash +//! map, adapted to make it a drop-in replacement for Rust's standard `HashMap` +//! and `HashSet` types. +//! +//! The original C++ version of [SwissTable] can be found [here], and this +//! [CppCon talk] gives an overview of how the algorithm works. +//! +//! [SwissTable]: https://abseil.io/blog/20180927-swisstables +//! [here]: https://github.com/abseil/abseil-cpp/blob/master/absl/container/internal/raw_hash_set.h +//! [CppCon talk]: https://www.youtube.com/watch?v=ncHmEUmJZf4 + +#![no_std] +#![cfg_attr( + feature = "nightly", + feature( + test, + core_intrinsics, + dropck_eyepatch, + min_specialization, + extend_one, + allocator_api, + slice_ptr_get, + maybe_uninit_array_assume_init, + strict_provenance + ) +)] +#![allow( + clippy::doc_markdown, + clippy::module_name_repetitions, + clippy::must_use_candidate, + clippy::option_if_let_else, + clippy::redundant_else, + clippy::manual_map, + clippy::missing_safety_doc, + clippy::missing_errors_doc +)] +#![warn(missing_docs)] +#![warn(rust_2018_idioms)] +#![cfg_attr(feature = "nightly", warn(fuzzy_provenance_casts))] + +#[cfg(test)] +#[macro_use] +extern crate std; + +#[cfg_attr(test, macro_use)] +extern crate alloc; + +#[cfg(feature = "nightly")] +#[cfg(doctest)] +doc_comment::doctest!("../README.md"); + +#[macro_use] +mod macros; + +#[cfg(feature = "raw")] +/// Experimental and unsafe `RawTable` API. This module is only available if the +/// `raw` feature is enabled. +pub mod raw { + // The RawTable API is still experimental and is not properly documented yet. + #[allow(missing_docs)] + #[path = "mod.rs"] + mod inner; + pub use inner::*; + + #[cfg(feature = "rayon")] + /// [rayon]-based parallel iterator types for hash maps. + /// You will rarely need to interact with it directly unless you have need + /// to name one of the iterator types. + /// + /// [rayon]: https://docs.rs/rayon/1.0/rayon + pub mod rayon { + pub use crate::external_trait_impls::rayon::raw::*; + } +} +#[cfg(not(feature = "raw"))] +mod raw; + +mod external_trait_impls; +mod map; +#[cfg(feature = "rustc-internal-api")] +mod rustc_entry; +mod scopeguard; +mod set; + +pub mod hash_map { + //! A hash map implemented with quadratic probing and SIMD lookup. + pub use crate::map::*; + + #[cfg(feature = "rustc-internal-api")] + pub use crate::rustc_entry::*; + + #[cfg(feature = "rayon")] + /// [rayon]-based parallel iterator types for hash maps. + /// You will rarely need to interact with it directly unless you have need + /// to name one of the iterator types. + /// + /// [rayon]: https://docs.rs/rayon/1.0/rayon + pub mod rayon { + pub use crate::external_trait_impls::rayon::map::*; + } +} +pub mod hash_set { + //! A hash set implemented as a `HashMap` where the value is `()`. + pub use crate::set::*; + + #[cfg(feature = "rayon")] + /// [rayon]-based parallel iterator types for hash sets. + /// You will rarely need to interact with it directly unless you have need + /// to name one of the iterator types. + /// + /// [rayon]: https://docs.rs/rayon/1.0/rayon + pub mod rayon { + pub use crate::external_trait_impls::rayon::set::*; + } +} + +pub use crate::map::HashMap; +pub use crate::set::HashSet; + +#[cfg(feature = "equivalent")] +pub use equivalent::Equivalent; + +// This is only used as a fallback when building as part of `std`. +#[cfg(not(feature = "equivalent"))] +/// Key equivalence trait. +/// +/// This trait defines the function used to compare the input value with the +/// map keys (or set values) during a lookup operation such as [`HashMap::get`] +/// or [`HashSet::contains`]. +/// It is provided with a blanket implementation based on the +/// [`Borrow`](core::borrow::Borrow) trait. +/// +/// # Correctness +/// +/// Equivalent values must hash to the same value. +pub trait Equivalent<K: ?Sized> { + /// Checks if this value is equivalent to the given key. + /// + /// Returns `true` if both values are equivalent, and `false` otherwise. + /// + /// # Correctness + /// + /// When this function returns `true`, both `self` and `key` must hash to + /// the same value. + fn equivalent(&self, key: &K) -> bool; +} + +#[cfg(not(feature = "equivalent"))] +impl<Q: ?Sized, K: ?Sized> Equivalent<K> for Q +where + Q: Eq, + K: core::borrow::Borrow<Q>, +{ + fn equivalent(&self, key: &K) -> bool { + self == key.borrow() + } +} + +/// The error type for `try_reserve` methods. +#[derive(Clone, PartialEq, Eq, Debug)] +pub enum TryReserveError { + /// Error due to the computed capacity exceeding the collection's maximum + /// (usually `isize::MAX` bytes). + CapacityOverflow, + + /// The memory allocator returned an error + AllocError { + /// The layout of the allocation request that failed. + layout: alloc::alloc::Layout, + }, +} diff --git a/vendor/hashbrown/src/macros.rs b/vendor/hashbrown/src/macros.rs new file mode 100644 index 0000000..eaba6be --- /dev/null +++ b/vendor/hashbrown/src/macros.rs @@ -0,0 +1,70 @@ +// See the cfg-if crate. +#[allow(unused_macro_rules)] +macro_rules! cfg_if { + // match if/else chains with a final `else` + ($( + if #[cfg($($meta:meta),*)] { $($it:item)* } + ) else * else { + $($it2:item)* + }) => { + cfg_if! { + @__items + () ; + $( ( ($($meta),*) ($($it)*) ), )* + ( () ($($it2)*) ), + } + }; + + // match if/else chains lacking a final `else` + ( + if #[cfg($($i_met:meta),*)] { $($i_it:item)* } + $( + else if #[cfg($($e_met:meta),*)] { $($e_it:item)* } + )* + ) => { + cfg_if! { + @__items + () ; + ( ($($i_met),*) ($($i_it)*) ), + $( ( ($($e_met),*) ($($e_it)*) ), )* + ( () () ), + } + }; + + // Internal and recursive macro to emit all the items + // + // Collects all the negated cfgs in a list at the beginning and after the + // semicolon is all the remaining items + (@__items ($($not:meta,)*) ; ) => {}; + (@__items ($($not:meta,)*) ; ( ($($m:meta),*) ($($it:item)*) ), $($rest:tt)*) => { + // Emit all items within one block, applying an appropriate #[cfg]. The + // #[cfg] will require all `$m` matchers specified and must also negate + // all previous matchers. + cfg_if! { @__apply cfg(all($($m,)* not(any($($not),*)))), $($it)* } + + // Recurse to emit all other items in `$rest`, and when we do so add all + // our `$m` matchers to the list of `$not` matchers as future emissions + // will have to negate everything we just matched as well. + cfg_if! { @__items ($($not,)* $($m,)*) ; $($rest)* } + }; + + // Internal macro to Apply a cfg attribute to a list of items + (@__apply $m:meta, $($it:item)*) => { + $(#[$m] $it)* + }; +} + +// Helper macro for specialization. This also helps avoid parse errors if the +// default fn syntax for specialization changes in the future. +#[cfg(feature = "nightly")] +macro_rules! default_fn { + (#[$($a:tt)*] $($tt:tt)*) => { + #[$($a)*] default $($tt)* + } +} +#[cfg(not(feature = "nightly"))] +macro_rules! default_fn { + ($($tt:tt)*) => { + $($tt)* + } +} diff --git a/vendor/hashbrown/src/map.rs b/vendor/hashbrown/src/map.rs new file mode 100644 index 0000000..edd1977 --- /dev/null +++ b/vendor/hashbrown/src/map.rs @@ -0,0 +1,8902 @@ +use crate::raw::{Allocator, Bucket, Global, RawDrain, RawIntoIter, RawIter, RawTable}; +use crate::{Equivalent, TryReserveError}; +use core::borrow::Borrow; +use core::fmt::{self, Debug}; +use core::hash::{BuildHasher, Hash}; +use core::iter::{FromIterator, FusedIterator}; +use core::marker::PhantomData; +use core::mem; +use core::ops::Index; + +/// Default hasher for `HashMap`. +#[cfg(feature = "ahash")] +pub type DefaultHashBuilder = core::hash::BuildHasherDefault<ahash::AHasher>; + +/// Dummy default hasher for `HashMap`. +#[cfg(not(feature = "ahash"))] +pub enum DefaultHashBuilder {} + +/// A hash map implemented with quadratic probing and SIMD lookup. +/// +/// The default hashing algorithm is currently [`AHash`], though this is +/// subject to change at any point in the future. This hash function is very +/// fast for all types of keys, but this algorithm will typically *not* protect +/// against attacks such as HashDoS. +/// +/// The hashing algorithm can be replaced on a per-`HashMap` basis using the +/// [`default`], [`with_hasher`], and [`with_capacity_and_hasher`] methods. Many +/// alternative algorithms are available on crates.io, such as the [`fnv`] crate. +/// +/// It is required that the keys implement the [`Eq`] and [`Hash`] traits, although +/// this can frequently be achieved by using `#[derive(PartialEq, Eq, Hash)]`. +/// If you implement these yourself, it is important that the following +/// property holds: +/// +/// ```text +/// k1 == k2 -> hash(k1) == hash(k2) +/// ``` +/// +/// In other words, if two keys are equal, their hashes must be equal. +/// +/// It is a logic error for a key to be modified in such a way that the key's +/// hash, as determined by the [`Hash`] trait, or its equality, as determined by +/// the [`Eq`] trait, changes while it is in the map. This is normally only +/// possible through [`Cell`], [`RefCell`], global state, I/O, or unsafe code. +/// +/// It is also a logic error for the [`Hash`] implementation of a key to panic. +/// This is generally only possible if the trait is implemented manually. If a +/// panic does occur then the contents of the `HashMap` may become corrupted and +/// some items may be dropped from the table. +/// +/// # Examples +/// +/// ``` +/// use hashbrown::HashMap; +/// +/// // Type inference lets us omit an explicit type signature (which +/// // would be `HashMap<String, String>` in this example). +/// let mut book_reviews = HashMap::new(); +/// +/// // Review some books. +/// book_reviews.insert( +/// "Adventures of Huckleberry Finn".to_string(), +/// "My favorite book.".to_string(), +/// ); +/// book_reviews.insert( +/// "Grimms' Fairy Tales".to_string(), +/// "Masterpiece.".to_string(), +/// ); +/// book_reviews.insert( +/// "Pride and Prejudice".to_string(), +/// "Very enjoyable.".to_string(), +/// ); +/// book_reviews.insert( +/// "The Adventures of Sherlock Holmes".to_string(), +/// "Eye lyked it alot.".to_string(), +/// ); +/// +/// // Check for a specific one. +/// // When collections store owned values (String), they can still be +/// // queried using references (&str). +/// if !book_reviews.contains_key("Les Misérables") { +/// println!("We've got {} reviews, but Les Misérables ain't one.", +/// book_reviews.len()); +/// } +/// +/// // oops, this review has a lot of spelling mistakes, let's delete it. +/// book_reviews.remove("The Adventures of Sherlock Holmes"); +/// +/// // Look up the values associated with some keys. +/// let to_find = ["Pride and Prejudice", "Alice's Adventure in Wonderland"]; +/// for &book in &to_find { +/// match book_reviews.get(book) { +/// Some(review) => println!("{}: {}", book, review), +/// None => println!("{} is unreviewed.", book) +/// } +/// } +/// +/// // Look up the value for a key (will panic if the key is not found). +/// println!("Review for Jane: {}", book_reviews["Pride and Prejudice"]); +/// +/// // Iterate over everything. +/// for (book, review) in &book_reviews { +/// println!("{}: \"{}\"", book, review); +/// } +/// ``` +/// +/// `HashMap` also implements an [`Entry API`](#method.entry), which allows +/// for more complex methods of getting, setting, updating and removing keys and +/// their values: +/// +/// ``` +/// use hashbrown::HashMap; +/// +/// // type inference lets us omit an explicit type signature (which +/// // would be `HashMap<&str, u8>` in this example). +/// let mut player_stats = HashMap::new(); +/// +/// fn random_stat_buff() -> u8 { +/// // could actually return some random value here - let's just return +/// // some fixed value for now +/// 42 +/// } +/// +/// // insert a key only if it doesn't already exist +/// player_stats.entry("health").or_insert(100); +/// +/// // insert a key using a function that provides a new value only if it +/// // doesn't already exist +/// player_stats.entry("defence").or_insert_with(random_stat_buff); +/// +/// // update a key, guarding against the key possibly not being set +/// let stat = player_stats.entry("attack").or_insert(100); +/// *stat += random_stat_buff(); +/// ``` +/// +/// The easiest way to use `HashMap` with a custom key type is to derive [`Eq`] and [`Hash`]. +/// We must also derive [`PartialEq`]. +/// +/// [`Eq`]: https://doc.rust-lang.org/std/cmp/trait.Eq.html +/// [`Hash`]: https://doc.rust-lang.org/std/hash/trait.Hash.html +/// [`PartialEq`]: https://doc.rust-lang.org/std/cmp/trait.PartialEq.html +/// [`RefCell`]: https://doc.rust-lang.org/std/cell/struct.RefCell.html +/// [`Cell`]: https://doc.rust-lang.org/std/cell/struct.Cell.html +/// [`default`]: #method.default +/// [`with_hasher`]: #method.with_hasher +/// [`with_capacity_and_hasher`]: #method.with_capacity_and_hasher +/// [`fnv`]: https://crates.io/crates/fnv +/// [`AHash`]: https://crates.io/crates/ahash +/// +/// ``` +/// use hashbrown::HashMap; +/// +/// #[derive(Hash, Eq, PartialEq, Debug)] +/// struct Viking { +/// name: String, +/// country: String, +/// } +/// +/// impl Viking { +/// /// Creates a new Viking. +/// fn new(name: &str, country: &str) -> Viking { +/// Viking { name: name.to_string(), country: country.to_string() } +/// } +/// } +/// +/// // Use a HashMap to store the vikings' health points. +/// let mut vikings = HashMap::new(); +/// +/// vikings.insert(Viking::new("Einar", "Norway"), 25); +/// vikings.insert(Viking::new("Olaf", "Denmark"), 24); +/// vikings.insert(Viking::new("Harald", "Iceland"), 12); +/// +/// // Use derived implementation to print the status of the vikings. +/// for (viking, health) in &vikings { +/// println!("{:?} has {} hp", viking, health); +/// } +/// ``` +/// +/// A `HashMap` with fixed list of elements can be initialized from an array: +/// +/// ``` +/// use hashbrown::HashMap; +/// +/// let timber_resources: HashMap<&str, i32> = [("Norway", 100), ("Denmark", 50), ("Iceland", 10)] +/// .iter().cloned().collect(); +/// // use the values stored in map +/// ``` +pub struct HashMap<K, V, S = DefaultHashBuilder, A: Allocator = Global> { + pub(crate) hash_builder: S, + pub(crate) table: RawTable<(K, V), A>, +} + +impl<K: Clone, V: Clone, S: Clone, A: Allocator + Clone> Clone for HashMap<K, V, S, A> { + fn clone(&self) -> Self { + HashMap { + hash_builder: self.hash_builder.clone(), + table: self.table.clone(), + } + } + + fn clone_from(&mut self, source: &Self) { + self.table.clone_from(&source.table); + + // Update hash_builder only if we successfully cloned all elements. + self.hash_builder.clone_from(&source.hash_builder); + } +} + +/// Ensures that a single closure type across uses of this which, in turn prevents multiple +/// instances of any functions like RawTable::reserve from being generated +#[cfg_attr(feature = "inline-more", inline)] +pub(crate) fn make_hasher<Q, V, S>(hash_builder: &S) -> impl Fn(&(Q, V)) -> u64 + '_ +where + Q: Hash, + S: BuildHasher, +{ + move |val| make_hash::<Q, S>(hash_builder, &val.0) +} + +/// Ensures that a single closure type across uses of this which, in turn prevents multiple +/// instances of any functions like RawTable::reserve from being generated +#[cfg_attr(feature = "inline-more", inline)] +fn equivalent_key<Q, K, V>(k: &Q) -> impl Fn(&(K, V)) -> bool + '_ +where + Q: ?Sized + Equivalent<K>, +{ + move |x| k.equivalent(&x.0) +} + +/// Ensures that a single closure type across uses of this which, in turn prevents multiple +/// instances of any functions like RawTable::reserve from being generated +#[cfg_attr(feature = "inline-more", inline)] +fn equivalent<Q, K>(k: &Q) -> impl Fn(&K) -> bool + '_ +where + Q: ?Sized + Equivalent<K>, +{ + move |x| k.equivalent(x) +} + +#[cfg(not(feature = "nightly"))] +#[cfg_attr(feature = "inline-more", inline)] +pub(crate) fn make_hash<Q, S>(hash_builder: &S, val: &Q) -> u64 +where + Q: Hash + ?Sized, + S: BuildHasher, +{ + use core::hash::Hasher; + let mut state = hash_builder.build_hasher(); + val.hash(&mut state); + state.finish() +} + +#[cfg(feature = "nightly")] +#[cfg_attr(feature = "inline-more", inline)] +pub(crate) fn make_hash<Q, S>(hash_builder: &S, val: &Q) -> u64 +where + Q: Hash + ?Sized, + S: BuildHasher, +{ + hash_builder.hash_one(val) +} + +#[cfg(feature = "ahash")] +impl<K, V> HashMap<K, V, DefaultHashBuilder> { + /// Creates an empty `HashMap`. + /// + /// The hash map is initially created with a capacity of 0, so it will not allocate until it + /// is first inserted into. + /// + /// # HashDoS resistance + /// + /// The `hash_builder` normally use a fixed key by default and that does + /// not allow the `HashMap` to be protected against attacks such as [`HashDoS`]. + /// Users who require HashDoS resistance should explicitly use + /// [`ahash::RandomState`] or [`std::collections::hash_map::RandomState`] + /// as the hasher when creating a [`HashMap`], for example with + /// [`with_hasher`](HashMap::with_hasher) method. + /// + /// [`HashDoS`]: https://en.wikipedia.org/wiki/Collision_attack + /// [`std::collections::hash_map::RandomState`]: https://doc.rust-lang.org/std/collections/hash_map/struct.RandomState.html + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// let mut map: HashMap<&str, i32> = HashMap::new(); + /// assert_eq!(map.len(), 0); + /// assert_eq!(map.capacity(), 0); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn new() -> Self { + Self::default() + } + + /// Creates an empty `HashMap` with the specified capacity. + /// + /// The hash map will be able to hold at least `capacity` elements without + /// reallocating. If `capacity` is 0, the hash map will not allocate. + /// + /// # HashDoS resistance + /// + /// The `hash_builder` normally use a fixed key by default and that does + /// not allow the `HashMap` to be protected against attacks such as [`HashDoS`]. + /// Users who require HashDoS resistance should explicitly use + /// [`ahash::RandomState`] or [`std::collections::hash_map::RandomState`] + /// as the hasher when creating a [`HashMap`], for example with + /// [`with_capacity_and_hasher`](HashMap::with_capacity_and_hasher) method. + /// + /// [`HashDoS`]: https://en.wikipedia.org/wiki/Collision_attack + /// [`std::collections::hash_map::RandomState`]: https://doc.rust-lang.org/std/collections/hash_map/struct.RandomState.html + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// let mut map: HashMap<&str, i32> = HashMap::with_capacity(10); + /// assert_eq!(map.len(), 0); + /// assert!(map.capacity() >= 10); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn with_capacity(capacity: usize) -> Self { + Self::with_capacity_and_hasher(capacity, DefaultHashBuilder::default()) + } +} + +#[cfg(feature = "ahash")] +impl<K, V, A: Allocator> HashMap<K, V, DefaultHashBuilder, A> { + /// Creates an empty `HashMap` using the given allocator. + /// + /// The hash map is initially created with a capacity of 0, so it will not allocate until it + /// is first inserted into. + /// + /// # HashDoS resistance + /// + /// The `hash_builder` normally use a fixed key by default and that does + /// not allow the `HashMap` to be protected against attacks such as [`HashDoS`]. + /// Users who require HashDoS resistance should explicitly use + /// [`ahash::RandomState`] or [`std::collections::hash_map::RandomState`] + /// as the hasher when creating a [`HashMap`], for example with + /// [`with_hasher_in`](HashMap::with_hasher_in) method. + /// + /// [`HashDoS`]: https://en.wikipedia.org/wiki/Collision_attack + /// [`std::collections::hash_map::RandomState`]: https://doc.rust-lang.org/std/collections/hash_map/struct.RandomState.html + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// use bumpalo::Bump; + /// + /// let bump = Bump::new(); + /// let mut map = HashMap::new_in(&bump); + /// + /// // The created HashMap holds none elements + /// assert_eq!(map.len(), 0); + /// + /// // The created HashMap also doesn't allocate memory + /// assert_eq!(map.capacity(), 0); + /// + /// // Now we insert element inside created HashMap + /// map.insert("One", 1); + /// // We can see that the HashMap holds 1 element + /// assert_eq!(map.len(), 1); + /// // And it also allocates some capacity + /// assert!(map.capacity() > 1); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn new_in(alloc: A) -> Self { + Self::with_hasher_in(DefaultHashBuilder::default(), alloc) + } + + /// Creates an empty `HashMap` with the specified capacity using the given allocator. + /// + /// The hash map will be able to hold at least `capacity` elements without + /// reallocating. If `capacity` is 0, the hash map will not allocate. + /// + /// # HashDoS resistance + /// + /// The `hash_builder` normally use a fixed key by default and that does + /// not allow the `HashMap` to be protected against attacks such as [`HashDoS`]. + /// Users who require HashDoS resistance should explicitly use + /// [`ahash::RandomState`] or [`std::collections::hash_map::RandomState`] + /// as the hasher when creating a [`HashMap`], for example with + /// [`with_capacity_and_hasher_in`](HashMap::with_capacity_and_hasher_in) method. + /// + /// [`HashDoS`]: https://en.wikipedia.org/wiki/Collision_attack + /// [`std::collections::hash_map::RandomState`]: https://doc.rust-lang.org/std/collections/hash_map/struct.RandomState.html + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// use bumpalo::Bump; + /// + /// let bump = Bump::new(); + /// let mut map = HashMap::with_capacity_in(5, &bump); + /// + /// // The created HashMap holds none elements + /// assert_eq!(map.len(), 0); + /// // But it can hold at least 5 elements without reallocating + /// let empty_map_capacity = map.capacity(); + /// assert!(empty_map_capacity >= 5); + /// + /// // Now we insert some 5 elements inside created HashMap + /// map.insert("One", 1); + /// map.insert("Two", 2); + /// map.insert("Three", 3); + /// map.insert("Four", 4); + /// map.insert("Five", 5); + /// + /// // We can see that the HashMap holds 5 elements + /// assert_eq!(map.len(), 5); + /// // But its capacity isn't changed + /// assert_eq!(map.capacity(), empty_map_capacity) + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn with_capacity_in(capacity: usize, alloc: A) -> Self { + Self::with_capacity_and_hasher_in(capacity, DefaultHashBuilder::default(), alloc) + } +} + +impl<K, V, S> HashMap<K, V, S> { + /// Creates an empty `HashMap` which will use the given hash builder to hash + /// keys. + /// + /// The hash map is initially created with a capacity of 0, so it will not + /// allocate until it is first inserted into. + /// + /// # HashDoS resistance + /// + /// The `hash_builder` normally use a fixed key by default and that does + /// not allow the `HashMap` to be protected against attacks such as [`HashDoS`]. + /// Users who require HashDoS resistance should explicitly use + /// [`ahash::RandomState`] or [`std::collections::hash_map::RandomState`] + /// as the hasher when creating a [`HashMap`]. + /// + /// The `hash_builder` passed should implement the [`BuildHasher`] trait for + /// the HashMap to be useful, see its documentation for details. + /// + /// [`HashDoS`]: https://en.wikipedia.org/wiki/Collision_attack + /// [`std::collections::hash_map::RandomState`]: https://doc.rust-lang.org/std/collections/hash_map/struct.RandomState.html + /// [`BuildHasher`]: https://doc.rust-lang.org/std/hash/trait.BuildHasher.html + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// use hashbrown::hash_map::DefaultHashBuilder; + /// + /// let s = DefaultHashBuilder::default(); + /// let mut map = HashMap::with_hasher(s); + /// assert_eq!(map.len(), 0); + /// assert_eq!(map.capacity(), 0); + /// + /// map.insert(1, 2); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub const fn with_hasher(hash_builder: S) -> Self { + Self { + hash_builder, + table: RawTable::new(), + } + } + + /// Creates an empty `HashMap` with the specified capacity, using `hash_builder` + /// to hash the keys. + /// + /// The hash map will be able to hold at least `capacity` elements without + /// reallocating. If `capacity` is 0, the hash map will not allocate. + /// + /// # HashDoS resistance + /// + /// The `hash_builder` normally use a fixed key by default and that does + /// not allow the `HashMap` to be protected against attacks such as [`HashDoS`]. + /// Users who require HashDoS resistance should explicitly use + /// [`ahash::RandomState`] or [`std::collections::hash_map::RandomState`] + /// as the hasher when creating a [`HashMap`]. + /// + /// The `hash_builder` passed should implement the [`BuildHasher`] trait for + /// the HashMap to be useful, see its documentation for details. + /// + /// [`HashDoS`]: https://en.wikipedia.org/wiki/Collision_attack + /// [`std::collections::hash_map::RandomState`]: https://doc.rust-lang.org/std/collections/hash_map/struct.RandomState.html + /// [`BuildHasher`]: https://doc.rust-lang.org/std/hash/trait.BuildHasher.html + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// use hashbrown::hash_map::DefaultHashBuilder; + /// + /// let s = DefaultHashBuilder::default(); + /// let mut map = HashMap::with_capacity_and_hasher(10, s); + /// assert_eq!(map.len(), 0); + /// assert!(map.capacity() >= 10); + /// + /// map.insert(1, 2); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn with_capacity_and_hasher(capacity: usize, hash_builder: S) -> Self { + Self { + hash_builder, + table: RawTable::with_capacity(capacity), + } + } +} + +impl<K, V, S, A: Allocator> HashMap<K, V, S, A> { + /// Returns a reference to the underlying allocator. + #[inline] + pub fn allocator(&self) -> &A { + self.table.allocator() + } + + /// Creates an empty `HashMap` which will use the given hash builder to hash + /// keys. It will be allocated with the given allocator. + /// + /// The hash map is initially created with a capacity of 0, so it will not allocate until it + /// is first inserted into. + /// + /// # HashDoS resistance + /// + /// The `hash_builder` normally use a fixed key by default and that does + /// not allow the `HashMap` to be protected against attacks such as [`HashDoS`]. + /// Users who require HashDoS resistance should explicitly use + /// [`ahash::RandomState`] or [`std::collections::hash_map::RandomState`] + /// as the hasher when creating a [`HashMap`]. + /// + /// [`HashDoS`]: https://en.wikipedia.org/wiki/Collision_attack + /// [`std::collections::hash_map::RandomState`]: https://doc.rust-lang.org/std/collections/hash_map/struct.RandomState.html + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// use hashbrown::hash_map::DefaultHashBuilder; + /// + /// let s = DefaultHashBuilder::default(); + /// let mut map = HashMap::with_hasher(s); + /// map.insert(1, 2); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub const fn with_hasher_in(hash_builder: S, alloc: A) -> Self { + Self { + hash_builder, + table: RawTable::new_in(alloc), + } + } + + /// Creates an empty `HashMap` with the specified capacity, using `hash_builder` + /// to hash the keys. It will be allocated with the given allocator. + /// + /// The hash map will be able to hold at least `capacity` elements without + /// reallocating. If `capacity` is 0, the hash map will not allocate. + /// + /// # HashDoS resistance + /// + /// The `hash_builder` normally use a fixed key by default and that does + /// not allow the `HashMap` to be protected against attacks such as [`HashDoS`]. + /// Users who require HashDoS resistance should explicitly use + /// [`ahash::RandomState`] or [`std::collections::hash_map::RandomState`] + /// as the hasher when creating a [`HashMap`]. + /// + /// [`HashDoS`]: https://en.wikipedia.org/wiki/Collision_attack + /// [`std::collections::hash_map::RandomState`]: https://doc.rust-lang.org/std/collections/hash_map/struct.RandomState.html + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// use hashbrown::hash_map::DefaultHashBuilder; + /// + /// let s = DefaultHashBuilder::default(); + /// let mut map = HashMap::with_capacity_and_hasher(10, s); + /// map.insert(1, 2); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn with_capacity_and_hasher_in(capacity: usize, hash_builder: S, alloc: A) -> Self { + Self { + hash_builder, + table: RawTable::with_capacity_in(capacity, alloc), + } + } + + /// Returns a reference to the map's [`BuildHasher`]. + /// + /// [`BuildHasher`]: https://doc.rust-lang.org/std/hash/trait.BuildHasher.html + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// use hashbrown::hash_map::DefaultHashBuilder; + /// + /// let hasher = DefaultHashBuilder::default(); + /// let map: HashMap<i32, i32> = HashMap::with_hasher(hasher); + /// let hasher: &DefaultHashBuilder = map.hasher(); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn hasher(&self) -> &S { + &self.hash_builder + } + + /// Returns the number of elements the map can hold without reallocating. + /// + /// This number is a lower bound; the `HashMap<K, V>` might be able to hold + /// more, but is guaranteed to be able to hold at least this many. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// let map: HashMap<i32, i32> = HashMap::with_capacity(100); + /// assert_eq!(map.len(), 0); + /// assert!(map.capacity() >= 100); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn capacity(&self) -> usize { + self.table.capacity() + } + + /// An iterator visiting all keys in arbitrary order. + /// The iterator element type is `&'a K`. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let mut map = HashMap::new(); + /// map.insert("a", 1); + /// map.insert("b", 2); + /// map.insert("c", 3); + /// assert_eq!(map.len(), 3); + /// let mut vec: Vec<&str> = Vec::new(); + /// + /// for key in map.keys() { + /// println!("{}", key); + /// vec.push(*key); + /// } + /// + /// // The `Keys` iterator produces keys in arbitrary order, so the + /// // keys must be sorted to test them against a sorted array. + /// vec.sort_unstable(); + /// assert_eq!(vec, ["a", "b", "c"]); + /// + /// assert_eq!(map.len(), 3); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn keys(&self) -> Keys<'_, K, V> { + Keys { inner: self.iter() } + } + + /// An iterator visiting all values in arbitrary order. + /// The iterator element type is `&'a V`. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let mut map = HashMap::new(); + /// map.insert("a", 1); + /// map.insert("b", 2); + /// map.insert("c", 3); + /// assert_eq!(map.len(), 3); + /// let mut vec: Vec<i32> = Vec::new(); + /// + /// for val in map.values() { + /// println!("{}", val); + /// vec.push(*val); + /// } + /// + /// // The `Values` iterator produces values in arbitrary order, so the + /// // values must be sorted to test them against a sorted array. + /// vec.sort_unstable(); + /// assert_eq!(vec, [1, 2, 3]); + /// + /// assert_eq!(map.len(), 3); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn values(&self) -> Values<'_, K, V> { + Values { inner: self.iter() } + } + + /// An iterator visiting all values mutably in arbitrary order. + /// The iterator element type is `&'a mut V`. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let mut map = HashMap::new(); + /// + /// map.insert("a", 1); + /// map.insert("b", 2); + /// map.insert("c", 3); + /// + /// for val in map.values_mut() { + /// *val = *val + 10; + /// } + /// + /// assert_eq!(map.len(), 3); + /// let mut vec: Vec<i32> = Vec::new(); + /// + /// for val in map.values() { + /// println!("{}", val); + /// vec.push(*val); + /// } + /// + /// // The `Values` iterator produces values in arbitrary order, so the + /// // values must be sorted to test them against a sorted array. + /// vec.sort_unstable(); + /// assert_eq!(vec, [11, 12, 13]); + /// + /// assert_eq!(map.len(), 3); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn values_mut(&mut self) -> ValuesMut<'_, K, V> { + ValuesMut { + inner: self.iter_mut(), + } + } + + /// An iterator visiting all key-value pairs in arbitrary order. + /// The iterator element type is `(&'a K, &'a V)`. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let mut map = HashMap::new(); + /// map.insert("a", 1); + /// map.insert("b", 2); + /// map.insert("c", 3); + /// assert_eq!(map.len(), 3); + /// let mut vec: Vec<(&str, i32)> = Vec::new(); + /// + /// for (key, val) in map.iter() { + /// println!("key: {} val: {}", key, val); + /// vec.push((*key, *val)); + /// } + /// + /// // The `Iter` iterator produces items in arbitrary order, so the + /// // items must be sorted to test them against a sorted array. + /// vec.sort_unstable(); + /// assert_eq!(vec, [("a", 1), ("b", 2), ("c", 3)]); + /// + /// assert_eq!(map.len(), 3); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn iter(&self) -> Iter<'_, K, V> { + // Here we tie the lifetime of self to the iter. + unsafe { + Iter { + inner: self.table.iter(), + marker: PhantomData, + } + } + } + + /// An iterator visiting all key-value pairs in arbitrary order, + /// with mutable references to the values. + /// The iterator element type is `(&'a K, &'a mut V)`. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let mut map = HashMap::new(); + /// map.insert("a", 1); + /// map.insert("b", 2); + /// map.insert("c", 3); + /// + /// // Update all values + /// for (_, val) in map.iter_mut() { + /// *val *= 2; + /// } + /// + /// assert_eq!(map.len(), 3); + /// let mut vec: Vec<(&str, i32)> = Vec::new(); + /// + /// for (key, val) in &map { + /// println!("key: {} val: {}", key, val); + /// vec.push((*key, *val)); + /// } + /// + /// // The `Iter` iterator produces items in arbitrary order, so the + /// // items must be sorted to test them against a sorted array. + /// vec.sort_unstable(); + /// assert_eq!(vec, [("a", 2), ("b", 4), ("c", 6)]); + /// + /// assert_eq!(map.len(), 3); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn iter_mut(&mut self) -> IterMut<'_, K, V> { + // Here we tie the lifetime of self to the iter. + unsafe { + IterMut { + inner: self.table.iter(), + marker: PhantomData, + } + } + } + + #[cfg(test)] + #[cfg_attr(feature = "inline-more", inline)] + fn raw_capacity(&self) -> usize { + self.table.buckets() + } + + /// Returns the number of elements in the map. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let mut a = HashMap::new(); + /// assert_eq!(a.len(), 0); + /// a.insert(1, "a"); + /// assert_eq!(a.len(), 1); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn len(&self) -> usize { + self.table.len() + } + + /// Returns `true` if the map contains no elements. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let mut a = HashMap::new(); + /// assert!(a.is_empty()); + /// a.insert(1, "a"); + /// assert!(!a.is_empty()); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn is_empty(&self) -> bool { + self.len() == 0 + } + + /// Clears the map, returning all key-value pairs as an iterator. Keeps the + /// allocated memory for reuse. + /// + /// If the returned iterator is dropped before being fully consumed, it + /// drops the remaining key-value pairs. The returned iterator keeps a + /// mutable borrow on the vector to optimize its implementation. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let mut a = HashMap::new(); + /// a.insert(1, "a"); + /// a.insert(2, "b"); + /// let capacity_before_drain = a.capacity(); + /// + /// for (k, v) in a.drain().take(1) { + /// assert!(k == 1 || k == 2); + /// assert!(v == "a" || v == "b"); + /// } + /// + /// // As we can see, the map is empty and contains no element. + /// assert!(a.is_empty() && a.len() == 0); + /// // But map capacity is equal to old one. + /// assert_eq!(a.capacity(), capacity_before_drain); + /// + /// let mut a = HashMap::new(); + /// a.insert(1, "a"); + /// a.insert(2, "b"); + /// + /// { // Iterator is dropped without being consumed. + /// let d = a.drain(); + /// } + /// + /// // But the map is empty even if we do not use Drain iterator. + /// assert!(a.is_empty()); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn drain(&mut self) -> Drain<'_, K, V, A> { + Drain { + inner: self.table.drain(), + } + } + + /// Retains only the elements specified by the predicate. Keeps the + /// allocated memory for reuse. + /// + /// In other words, remove all pairs `(k, v)` such that `f(&k, &mut v)` returns `false`. + /// The elements are visited in unsorted (and unspecified) order. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let mut map: HashMap<i32, i32> = (0..8).map(|x|(x, x*10)).collect(); + /// assert_eq!(map.len(), 8); + /// + /// map.retain(|&k, _| k % 2 == 0); + /// + /// // We can see, that the number of elements inside map is changed. + /// assert_eq!(map.len(), 4); + /// + /// let mut vec: Vec<(i32, i32)> = map.iter().map(|(&k, &v)| (k, v)).collect(); + /// vec.sort_unstable(); + /// assert_eq!(vec, [(0, 0), (2, 20), (4, 40), (6, 60)]); + /// ``` + pub fn retain<F>(&mut self, mut f: F) + where + F: FnMut(&K, &mut V) -> bool, + { + // Here we only use `iter` as a temporary, preventing use-after-free + unsafe { + for item in self.table.iter() { + let &mut (ref key, ref mut value) = item.as_mut(); + if !f(key, value) { + self.table.erase(item); + } + } + } + } + + /// Drains elements which are true under the given predicate, + /// and returns an iterator over the removed items. + /// + /// In other words, move all pairs `(k, v)` such that `f(&k, &mut v)` returns `true` out + /// into another iterator. + /// + /// Note that `extract_if` lets you mutate every value in the filter closure, regardless of + /// whether you choose to keep or remove it. + /// + /// If the returned `ExtractIf` is not exhausted, e.g. because it is dropped without iterating + /// or the iteration short-circuits, then the remaining elements will be retained. + /// Use [`retain()`] with a negated predicate if you do not need the returned iterator. + /// + /// Keeps the allocated memory for reuse. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let mut map: HashMap<i32, i32> = (0..8).map(|x| (x, x)).collect(); + /// + /// let drained: HashMap<i32, i32> = map.extract_if(|k, _v| k % 2 == 0).collect(); + /// + /// let mut evens = drained.keys().cloned().collect::<Vec<_>>(); + /// let mut odds = map.keys().cloned().collect::<Vec<_>>(); + /// evens.sort(); + /// odds.sort(); + /// + /// assert_eq!(evens, vec![0, 2, 4, 6]); + /// assert_eq!(odds, vec![1, 3, 5, 7]); + /// + /// let mut map: HashMap<i32, i32> = (0..8).map(|x| (x, x)).collect(); + /// + /// { // Iterator is dropped without being consumed. + /// let d = map.extract_if(|k, _v| k % 2 != 0); + /// } + /// + /// // ExtractIf was not exhausted, therefore no elements were drained. + /// assert_eq!(map.len(), 8); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn extract_if<F>(&mut self, f: F) -> ExtractIf<'_, K, V, F, A> + where + F: FnMut(&K, &mut V) -> bool, + { + ExtractIf { + f, + inner: ExtractIfInner { + iter: unsafe { self.table.iter() }, + table: &mut self.table, + }, + } + } + + /// Clears the map, removing all key-value pairs. Keeps the allocated memory + /// for reuse. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let mut a = HashMap::new(); + /// a.insert(1, "a"); + /// let capacity_before_clear = a.capacity(); + /// + /// a.clear(); + /// + /// // Map is empty. + /// assert!(a.is_empty()); + /// // But map capacity is equal to old one. + /// assert_eq!(a.capacity(), capacity_before_clear); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn clear(&mut self) { + self.table.clear(); + } + + /// Creates a consuming iterator visiting all the keys in arbitrary order. + /// The map cannot be used after calling this. + /// The iterator element type is `K`. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let mut map = HashMap::new(); + /// map.insert("a", 1); + /// map.insert("b", 2); + /// map.insert("c", 3); + /// + /// let mut vec: Vec<&str> = map.into_keys().collect(); + /// + /// // The `IntoKeys` iterator produces keys in arbitrary order, so the + /// // keys must be sorted to test them against a sorted array. + /// vec.sort_unstable(); + /// assert_eq!(vec, ["a", "b", "c"]); + /// ``` + #[inline] + pub fn into_keys(self) -> IntoKeys<K, V, A> { + IntoKeys { + inner: self.into_iter(), + } + } + + /// Creates a consuming iterator visiting all the values in arbitrary order. + /// The map cannot be used after calling this. + /// The iterator element type is `V`. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let mut map = HashMap::new(); + /// map.insert("a", 1); + /// map.insert("b", 2); + /// map.insert("c", 3); + /// + /// let mut vec: Vec<i32> = map.into_values().collect(); + /// + /// // The `IntoValues` iterator produces values in arbitrary order, so + /// // the values must be sorted to test them against a sorted array. + /// vec.sort_unstable(); + /// assert_eq!(vec, [1, 2, 3]); + /// ``` + #[inline] + pub fn into_values(self) -> IntoValues<K, V, A> { + IntoValues { + inner: self.into_iter(), + } + } +} + +impl<K, V, S, A> HashMap<K, V, S, A> +where + K: Eq + Hash, + S: BuildHasher, + A: Allocator, +{ + /// Reserves capacity for at least `additional` more elements to be inserted + /// in the `HashMap`. The collection may reserve more space to avoid + /// frequent reallocations. + /// + /// # Panics + /// + /// Panics if the new capacity exceeds [`isize::MAX`] bytes and [`abort`] the program + /// in case of allocation error. Use [`try_reserve`](HashMap::try_reserve) instead + /// if you want to handle memory allocation failure. + /// + /// [`isize::MAX`]: https://doc.rust-lang.org/std/primitive.isize.html + /// [`abort`]: https://doc.rust-lang.org/alloc/alloc/fn.handle_alloc_error.html + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// let mut map: HashMap<&str, i32> = HashMap::new(); + /// // Map is empty and doesn't allocate memory + /// assert_eq!(map.capacity(), 0); + /// + /// map.reserve(10); + /// + /// // And now map can hold at least 10 elements + /// assert!(map.capacity() >= 10); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn reserve(&mut self, additional: usize) { + self.table + .reserve(additional, make_hasher::<_, V, S>(&self.hash_builder)); + } + + /// Tries to reserve capacity for at least `additional` more elements to be inserted + /// in the given `HashMap<K,V>`. The collection may reserve more space to avoid + /// frequent reallocations. + /// + /// # Errors + /// + /// If the capacity overflows, or the allocator reports a failure, then an error + /// is returned. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let mut map: HashMap<&str, isize> = HashMap::new(); + /// // Map is empty and doesn't allocate memory + /// assert_eq!(map.capacity(), 0); + /// + /// map.try_reserve(10).expect("why is the test harness OOMing on 10 bytes?"); + /// + /// // And now map can hold at least 10 elements + /// assert!(map.capacity() >= 10); + /// ``` + /// If the capacity overflows, or the allocator reports a failure, then an error + /// is returned: + /// ``` + /// # fn test() { + /// use hashbrown::HashMap; + /// use hashbrown::TryReserveError; + /// let mut map: HashMap<i32, i32> = HashMap::new(); + /// + /// match map.try_reserve(usize::MAX) { + /// Err(error) => match error { + /// TryReserveError::CapacityOverflow => {} + /// _ => panic!("TryReserveError::AllocError ?"), + /// }, + /// _ => panic!(), + /// } + /// # } + /// # fn main() { + /// # #[cfg(not(miri))] + /// # test() + /// # } + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn try_reserve(&mut self, additional: usize) -> Result<(), TryReserveError> { + self.table + .try_reserve(additional, make_hasher::<_, V, S>(&self.hash_builder)) + } + + /// Shrinks the capacity of the map as much as possible. It will drop + /// down as much as possible while maintaining the internal rules + /// and possibly leaving some space in accordance with the resize policy. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let mut map: HashMap<i32, i32> = HashMap::with_capacity(100); + /// map.insert(1, 2); + /// map.insert(3, 4); + /// assert!(map.capacity() >= 100); + /// map.shrink_to_fit(); + /// assert!(map.capacity() >= 2); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn shrink_to_fit(&mut self) { + self.table + .shrink_to(0, make_hasher::<_, V, S>(&self.hash_builder)); + } + + /// Shrinks the capacity of the map with a lower limit. It will drop + /// down no lower than the supplied limit while maintaining the internal rules + /// and possibly leaving some space in accordance with the resize policy. + /// + /// This function does nothing if the current capacity is smaller than the + /// supplied minimum capacity. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let mut map: HashMap<i32, i32> = HashMap::with_capacity(100); + /// map.insert(1, 2); + /// map.insert(3, 4); + /// assert!(map.capacity() >= 100); + /// map.shrink_to(10); + /// assert!(map.capacity() >= 10); + /// map.shrink_to(0); + /// assert!(map.capacity() >= 2); + /// map.shrink_to(10); + /// assert!(map.capacity() >= 2); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn shrink_to(&mut self, min_capacity: usize) { + self.table + .shrink_to(min_capacity, make_hasher::<_, V, S>(&self.hash_builder)); + } + + /// Gets the given key's corresponding entry in the map for in-place manipulation. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let mut letters = HashMap::new(); + /// + /// for ch in "a short treatise on fungi".chars() { + /// let counter = letters.entry(ch).or_insert(0); + /// *counter += 1; + /// } + /// + /// assert_eq!(letters[&'s'], 2); + /// assert_eq!(letters[&'t'], 3); + /// assert_eq!(letters[&'u'], 1); + /// assert_eq!(letters.get(&'y'), None); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn entry(&mut self, key: K) -> Entry<'_, K, V, S, A> { + let hash = make_hash::<K, S>(&self.hash_builder, &key); + if let Some(elem) = self.table.find(hash, equivalent_key(&key)) { + Entry::Occupied(OccupiedEntry { + hash, + key: Some(key), + elem, + table: self, + }) + } else { + Entry::Vacant(VacantEntry { + hash, + key, + table: self, + }) + } + } + + /// Gets the given key's corresponding entry by reference in the map for in-place manipulation. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let mut words: HashMap<String, usize> = HashMap::new(); + /// let source = ["poneyland", "horseyland", "poneyland", "poneyland"]; + /// for (i, &s) in source.iter().enumerate() { + /// let counter = words.entry_ref(s).or_insert(0); + /// *counter += 1; + /// } + /// + /// assert_eq!(words["poneyland"], 3); + /// assert_eq!(words["horseyland"], 1); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn entry_ref<'a, 'b, Q: ?Sized>(&'a mut self, key: &'b Q) -> EntryRef<'a, 'b, K, Q, V, S, A> + where + Q: Hash + Equivalent<K>, + { + let hash = make_hash::<Q, S>(&self.hash_builder, key); + if let Some(elem) = self.table.find(hash, equivalent_key(key)) { + EntryRef::Occupied(OccupiedEntryRef { + hash, + key: Some(KeyOrRef::Borrowed(key)), + elem, + table: self, + }) + } else { + EntryRef::Vacant(VacantEntryRef { + hash, + key: KeyOrRef::Borrowed(key), + table: self, + }) + } + } + + /// Returns a reference to the value corresponding to the key. + /// + /// The key may be any borrowed form of the map's key type, but + /// [`Hash`] and [`Eq`] on the borrowed form *must* match those for + /// the key type. + /// + /// [`Eq`]: https://doc.rust-lang.org/std/cmp/trait.Eq.html + /// [`Hash`]: https://doc.rust-lang.org/std/hash/trait.Hash.html + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let mut map = HashMap::new(); + /// map.insert(1, "a"); + /// assert_eq!(map.get(&1), Some(&"a")); + /// assert_eq!(map.get(&2), None); + /// ``` + #[inline] + pub fn get<Q: ?Sized>(&self, k: &Q) -> Option<&V> + where + Q: Hash + Equivalent<K>, + { + // Avoid `Option::map` because it bloats LLVM IR. + match self.get_inner(k) { + Some((_, v)) => Some(v), + None => None, + } + } + + /// Returns the key-value pair corresponding to the supplied key. + /// + /// The supplied key may be any borrowed form of the map's key type, but + /// [`Hash`] and [`Eq`] on the borrowed form *must* match those for + /// the key type. + /// + /// [`Eq`]: https://doc.rust-lang.org/std/cmp/trait.Eq.html + /// [`Hash`]: https://doc.rust-lang.org/std/hash/trait.Hash.html + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let mut map = HashMap::new(); + /// map.insert(1, "a"); + /// assert_eq!(map.get_key_value(&1), Some((&1, &"a"))); + /// assert_eq!(map.get_key_value(&2), None); + /// ``` + #[inline] + pub fn get_key_value<Q: ?Sized>(&self, k: &Q) -> Option<(&K, &V)> + where + Q: Hash + Equivalent<K>, + { + // Avoid `Option::map` because it bloats LLVM IR. + match self.get_inner(k) { + Some((key, value)) => Some((key, value)), + None => None, + } + } + + #[inline] + fn get_inner<Q: ?Sized>(&self, k: &Q) -> Option<&(K, V)> + where + Q: Hash + Equivalent<K>, + { + if self.table.is_empty() { + None + } else { + let hash = make_hash::<Q, S>(&self.hash_builder, k); + self.table.get(hash, equivalent_key(k)) + } + } + + /// Returns the key-value pair corresponding to the supplied key, with a mutable reference to value. + /// + /// The supplied key may be any borrowed form of the map's key type, but + /// [`Hash`] and [`Eq`] on the borrowed form *must* match those for + /// the key type. + /// + /// [`Eq`]: https://doc.rust-lang.org/std/cmp/trait.Eq.html + /// [`Hash`]: https://doc.rust-lang.org/std/hash/trait.Hash.html + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let mut map = HashMap::new(); + /// map.insert(1, "a"); + /// let (k, v) = map.get_key_value_mut(&1).unwrap(); + /// assert_eq!(k, &1); + /// assert_eq!(v, &mut "a"); + /// *v = "b"; + /// assert_eq!(map.get_key_value_mut(&1), Some((&1, &mut "b"))); + /// assert_eq!(map.get_key_value_mut(&2), None); + /// ``` + #[inline] + pub fn get_key_value_mut<Q: ?Sized>(&mut self, k: &Q) -> Option<(&K, &mut V)> + where + Q: Hash + Equivalent<K>, + { + // Avoid `Option::map` because it bloats LLVM IR. + match self.get_inner_mut(k) { + Some(&mut (ref key, ref mut value)) => Some((key, value)), + None => None, + } + } + + /// Returns `true` if the map contains a value for the specified key. + /// + /// The key may be any borrowed form of the map's key type, but + /// [`Hash`] and [`Eq`] on the borrowed form *must* match those for + /// the key type. + /// + /// [`Eq`]: https://doc.rust-lang.org/std/cmp/trait.Eq.html + /// [`Hash`]: https://doc.rust-lang.org/std/hash/trait.Hash.html + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let mut map = HashMap::new(); + /// map.insert(1, "a"); + /// assert_eq!(map.contains_key(&1), true); + /// assert_eq!(map.contains_key(&2), false); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn contains_key<Q: ?Sized>(&self, k: &Q) -> bool + where + Q: Hash + Equivalent<K>, + { + self.get_inner(k).is_some() + } + + /// Returns a mutable reference to the value corresponding to the key. + /// + /// The key may be any borrowed form of the map's key type, but + /// [`Hash`] and [`Eq`] on the borrowed form *must* match those for + /// the key type. + /// + /// [`Eq`]: https://doc.rust-lang.org/std/cmp/trait.Eq.html + /// [`Hash`]: https://doc.rust-lang.org/std/hash/trait.Hash.html + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let mut map = HashMap::new(); + /// map.insert(1, "a"); + /// if let Some(x) = map.get_mut(&1) { + /// *x = "b"; + /// } + /// assert_eq!(map[&1], "b"); + /// + /// assert_eq!(map.get_mut(&2), None); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn get_mut<Q: ?Sized>(&mut self, k: &Q) -> Option<&mut V> + where + Q: Hash + Equivalent<K>, + { + // Avoid `Option::map` because it bloats LLVM IR. + match self.get_inner_mut(k) { + Some(&mut (_, ref mut v)) => Some(v), + None => None, + } + } + + #[inline] + fn get_inner_mut<Q: ?Sized>(&mut self, k: &Q) -> Option<&mut (K, V)> + where + Q: Hash + Equivalent<K>, + { + if self.table.is_empty() { + None + } else { + let hash = make_hash::<Q, S>(&self.hash_builder, k); + self.table.get_mut(hash, equivalent_key(k)) + } + } + + /// Attempts to get mutable references to `N` values in the map at once. + /// + /// Returns an array of length `N` with the results of each query. For soundness, at most one + /// mutable reference will be returned to any value. `None` will be returned if any of the + /// keys are duplicates or missing. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let mut libraries = HashMap::new(); + /// libraries.insert("Bodleian Library".to_string(), 1602); + /// libraries.insert("Athenæum".to_string(), 1807); + /// libraries.insert("Herzogin-Anna-Amalia-Bibliothek".to_string(), 1691); + /// libraries.insert("Library of Congress".to_string(), 1800); + /// + /// let got = libraries.get_many_mut([ + /// "Athenæum", + /// "Library of Congress", + /// ]); + /// assert_eq!( + /// got, + /// Some([ + /// &mut 1807, + /// &mut 1800, + /// ]), + /// ); + /// + /// // Missing keys result in None + /// let got = libraries.get_many_mut([ + /// "Athenæum", + /// "New York Public Library", + /// ]); + /// assert_eq!(got, None); + /// + /// // Duplicate keys result in None + /// let got = libraries.get_many_mut([ + /// "Athenæum", + /// "Athenæum", + /// ]); + /// assert_eq!(got, None); + /// ``` + pub fn get_many_mut<Q: ?Sized, const N: usize>(&mut self, ks: [&Q; N]) -> Option<[&'_ mut V; N]> + where + Q: Hash + Equivalent<K>, + { + self.get_many_mut_inner(ks).map(|res| res.map(|(_, v)| v)) + } + + /// Attempts to get mutable references to `N` values in the map at once, without validating that + /// the values are unique. + /// + /// Returns an array of length `N` with the results of each query. `None` will be returned if + /// any of the keys are missing. + /// + /// For a safe alternative see [`get_many_mut`](`HashMap::get_many_mut`). + /// + /// # Safety + /// + /// Calling this method with overlapping keys is *[undefined behavior]* even if the resulting + /// references are not used. + /// + /// [undefined behavior]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let mut libraries = HashMap::new(); + /// libraries.insert("Bodleian Library".to_string(), 1602); + /// libraries.insert("Athenæum".to_string(), 1807); + /// libraries.insert("Herzogin-Anna-Amalia-Bibliothek".to_string(), 1691); + /// libraries.insert("Library of Congress".to_string(), 1800); + /// + /// let got = libraries.get_many_mut([ + /// "Athenæum", + /// "Library of Congress", + /// ]); + /// assert_eq!( + /// got, + /// Some([ + /// &mut 1807, + /// &mut 1800, + /// ]), + /// ); + /// + /// // Missing keys result in None + /// let got = libraries.get_many_mut([ + /// "Athenæum", + /// "New York Public Library", + /// ]); + /// assert_eq!(got, None); + /// ``` + pub unsafe fn get_many_unchecked_mut<Q: ?Sized, const N: usize>( + &mut self, + ks: [&Q; N], + ) -> Option<[&'_ mut V; N]> + where + Q: Hash + Equivalent<K>, + { + self.get_many_unchecked_mut_inner(ks) + .map(|res| res.map(|(_, v)| v)) + } + + /// Attempts to get mutable references to `N` values in the map at once, with immutable + /// references to the corresponding keys. + /// + /// Returns an array of length `N` with the results of each query. For soundness, at most one + /// mutable reference will be returned to any value. `None` will be returned if any of the keys + /// are duplicates or missing. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let mut libraries = HashMap::new(); + /// libraries.insert("Bodleian Library".to_string(), 1602); + /// libraries.insert("Athenæum".to_string(), 1807); + /// libraries.insert("Herzogin-Anna-Amalia-Bibliothek".to_string(), 1691); + /// libraries.insert("Library of Congress".to_string(), 1800); + /// + /// let got = libraries.get_many_key_value_mut([ + /// "Bodleian Library", + /// "Herzogin-Anna-Amalia-Bibliothek", + /// ]); + /// assert_eq!( + /// got, + /// Some([ + /// (&"Bodleian Library".to_string(), &mut 1602), + /// (&"Herzogin-Anna-Amalia-Bibliothek".to_string(), &mut 1691), + /// ]), + /// ); + /// // Missing keys result in None + /// let got = libraries.get_many_key_value_mut([ + /// "Bodleian Library", + /// "Gewandhaus", + /// ]); + /// assert_eq!(got, None); + /// + /// // Duplicate keys result in None + /// let got = libraries.get_many_key_value_mut([ + /// "Bodleian Library", + /// "Herzogin-Anna-Amalia-Bibliothek", + /// "Herzogin-Anna-Amalia-Bibliothek", + /// ]); + /// assert_eq!(got, None); + /// ``` + pub fn get_many_key_value_mut<Q: ?Sized, const N: usize>( + &mut self, + ks: [&Q; N], + ) -> Option<[(&'_ K, &'_ mut V); N]> + where + Q: Hash + Equivalent<K>, + { + self.get_many_mut_inner(ks) + .map(|res| res.map(|(k, v)| (&*k, v))) + } + + /// Attempts to get mutable references to `N` values in the map at once, with immutable + /// references to the corresponding keys, without validating that the values are unique. + /// + /// Returns an array of length `N` with the results of each query. `None` will be returned if + /// any of the keys are missing. + /// + /// For a safe alternative see [`get_many_key_value_mut`](`HashMap::get_many_key_value_mut`). + /// + /// # Safety + /// + /// Calling this method with overlapping keys is *[undefined behavior]* even if the resulting + /// references are not used. + /// + /// [undefined behavior]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let mut libraries = HashMap::new(); + /// libraries.insert("Bodleian Library".to_string(), 1602); + /// libraries.insert("Athenæum".to_string(), 1807); + /// libraries.insert("Herzogin-Anna-Amalia-Bibliothek".to_string(), 1691); + /// libraries.insert("Library of Congress".to_string(), 1800); + /// + /// let got = libraries.get_many_key_value_mut([ + /// "Bodleian Library", + /// "Herzogin-Anna-Amalia-Bibliothek", + /// ]); + /// assert_eq!( + /// got, + /// Some([ + /// (&"Bodleian Library".to_string(), &mut 1602), + /// (&"Herzogin-Anna-Amalia-Bibliothek".to_string(), &mut 1691), + /// ]), + /// ); + /// // Missing keys result in None + /// let got = libraries.get_many_key_value_mut([ + /// "Bodleian Library", + /// "Gewandhaus", + /// ]); + /// assert_eq!(got, None); + /// ``` + pub unsafe fn get_many_key_value_unchecked_mut<Q: ?Sized, const N: usize>( + &mut self, + ks: [&Q; N], + ) -> Option<[(&'_ K, &'_ mut V); N]> + where + Q: Hash + Equivalent<K>, + { + self.get_many_unchecked_mut_inner(ks) + .map(|res| res.map(|(k, v)| (&*k, v))) + } + + fn get_many_mut_inner<Q: ?Sized, const N: usize>( + &mut self, + ks: [&Q; N], + ) -> Option<[&'_ mut (K, V); N]> + where + Q: Hash + Equivalent<K>, + { + let hashes = self.build_hashes_inner(ks); + self.table + .get_many_mut(hashes, |i, (k, _)| ks[i].equivalent(k)) + } + + unsafe fn get_many_unchecked_mut_inner<Q: ?Sized, const N: usize>( + &mut self, + ks: [&Q; N], + ) -> Option<[&'_ mut (K, V); N]> + where + Q: Hash + Equivalent<K>, + { + let hashes = self.build_hashes_inner(ks); + self.table + .get_many_unchecked_mut(hashes, |i, (k, _)| ks[i].equivalent(k)) + } + + fn build_hashes_inner<Q: ?Sized, const N: usize>(&self, ks: [&Q; N]) -> [u64; N] + where + Q: Hash + Equivalent<K>, + { + let mut hashes = [0_u64; N]; + for i in 0..N { + hashes[i] = make_hash::<Q, S>(&self.hash_builder, ks[i]); + } + hashes + } + + /// Inserts a key-value pair into the map. + /// + /// If the map did not have this key present, [`None`] is returned. + /// + /// If the map did have this key present, the value is updated, and the old + /// value is returned. The key is not updated, though; this matters for + /// types that can be `==` without being identical. See the [`std::collections`] + /// [module-level documentation] for more. + /// + /// [`None`]: https://doc.rust-lang.org/std/option/enum.Option.html#variant.None + /// [`std::collections`]: https://doc.rust-lang.org/std/collections/index.html + /// [module-level documentation]: https://doc.rust-lang.org/std/collections/index.html#insert-and-complex-keys + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let mut map = HashMap::new(); + /// assert_eq!(map.insert(37, "a"), None); + /// assert_eq!(map.is_empty(), false); + /// + /// map.insert(37, "b"); + /// assert_eq!(map.insert(37, "c"), Some("b")); + /// assert_eq!(map[&37], "c"); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn insert(&mut self, k: K, v: V) -> Option<V> { + let hash = make_hash::<K, S>(&self.hash_builder, &k); + let hasher = make_hasher::<_, V, S>(&self.hash_builder); + match self + .table + .find_or_find_insert_slot(hash, equivalent_key(&k), hasher) + { + Ok(bucket) => Some(mem::replace(unsafe { &mut bucket.as_mut().1 }, v)), + Err(slot) => { + unsafe { + self.table.insert_in_slot(hash, slot, (k, v)); + } + None + } + } + } + + /// Insert a key-value pair into the map without checking + /// if the key already exists in the map. + /// + /// Returns a reference to the key and value just inserted. + /// + /// This operation is safe if a key does not exist in the map. + /// + /// However, if a key exists in the map already, the behavior is unspecified: + /// this operation may panic, loop forever, or any following operation with the map + /// may panic, loop forever or return arbitrary result. + /// + /// That said, this operation (and following operations) are guaranteed to + /// not violate memory safety. + /// + /// This operation is faster than regular insert, because it does not perform + /// lookup before insertion. + /// + /// This operation is useful during initial population of the map. + /// For example, when constructing a map from another map, we know + /// that keys are unique. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let mut map1 = HashMap::new(); + /// assert_eq!(map1.insert(1, "a"), None); + /// assert_eq!(map1.insert(2, "b"), None); + /// assert_eq!(map1.insert(3, "c"), None); + /// assert_eq!(map1.len(), 3); + /// + /// let mut map2 = HashMap::new(); + /// + /// for (key, value) in map1.into_iter() { + /// map2.insert_unique_unchecked(key, value); + /// } + /// + /// let (key, value) = map2.insert_unique_unchecked(4, "d"); + /// assert_eq!(key, &4); + /// assert_eq!(value, &mut "d"); + /// *value = "e"; + /// + /// assert_eq!(map2[&1], "a"); + /// assert_eq!(map2[&2], "b"); + /// assert_eq!(map2[&3], "c"); + /// assert_eq!(map2[&4], "e"); + /// assert_eq!(map2.len(), 4); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn insert_unique_unchecked(&mut self, k: K, v: V) -> (&K, &mut V) { + let hash = make_hash::<K, S>(&self.hash_builder, &k); + let bucket = self + .table + .insert(hash, (k, v), make_hasher::<_, V, S>(&self.hash_builder)); + let (k_ref, v_ref) = unsafe { bucket.as_mut() }; + (k_ref, v_ref) + } + + /// Tries to insert a key-value pair into the map, and returns + /// a mutable reference to the value in the entry. + /// + /// # Errors + /// + /// If the map already had this key present, nothing is updated, and + /// an error containing the occupied entry and the value is returned. + /// + /// # Examples + /// + /// Basic usage: + /// + /// ``` + /// use hashbrown::HashMap; + /// use hashbrown::hash_map::OccupiedError; + /// + /// let mut map = HashMap::new(); + /// assert_eq!(map.try_insert(37, "a").unwrap(), &"a"); + /// + /// match map.try_insert(37, "b") { + /// Err(OccupiedError { entry, value }) => { + /// assert_eq!(entry.key(), &37); + /// assert_eq!(entry.get(), &"a"); + /// assert_eq!(value, "b"); + /// } + /// _ => panic!() + /// } + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn try_insert( + &mut self, + key: K, + value: V, + ) -> Result<&mut V, OccupiedError<'_, K, V, S, A>> { + match self.entry(key) { + Entry::Occupied(entry) => Err(OccupiedError { entry, value }), + Entry::Vacant(entry) => Ok(entry.insert(value)), + } + } + + /// Removes a key from the map, returning the value at the key if the key + /// was previously in the map. Keeps the allocated memory for reuse. + /// + /// The key may be any borrowed form of the map's key type, but + /// [`Hash`] and [`Eq`] on the borrowed form *must* match those for + /// the key type. + /// + /// [`Eq`]: https://doc.rust-lang.org/std/cmp/trait.Eq.html + /// [`Hash`]: https://doc.rust-lang.org/std/hash/trait.Hash.html + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let mut map = HashMap::new(); + /// // The map is empty + /// assert!(map.is_empty() && map.capacity() == 0); + /// + /// map.insert(1, "a"); + /// + /// assert_eq!(map.remove(&1), Some("a")); + /// assert_eq!(map.remove(&1), None); + /// + /// // Now map holds none elements + /// assert!(map.is_empty()); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn remove<Q: ?Sized>(&mut self, k: &Q) -> Option<V> + where + Q: Hash + Equivalent<K>, + { + // Avoid `Option::map` because it bloats LLVM IR. + match self.remove_entry(k) { + Some((_, v)) => Some(v), + None => None, + } + } + + /// Removes a key from the map, returning the stored key and value if the + /// key was previously in the map. Keeps the allocated memory for reuse. + /// + /// The key may be any borrowed form of the map's key type, but + /// [`Hash`] and [`Eq`] on the borrowed form *must* match those for + /// the key type. + /// + /// [`Eq`]: https://doc.rust-lang.org/std/cmp/trait.Eq.html + /// [`Hash`]: https://doc.rust-lang.org/std/hash/trait.Hash.html + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let mut map = HashMap::new(); + /// // The map is empty + /// assert!(map.is_empty() && map.capacity() == 0); + /// + /// map.insert(1, "a"); + /// + /// assert_eq!(map.remove_entry(&1), Some((1, "a"))); + /// assert_eq!(map.remove(&1), None); + /// + /// // Now map hold none elements + /// assert!(map.is_empty()); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn remove_entry<Q: ?Sized>(&mut self, k: &Q) -> Option<(K, V)> + where + Q: Hash + Equivalent<K>, + { + let hash = make_hash::<Q, S>(&self.hash_builder, k); + self.table.remove_entry(hash, equivalent_key(k)) + } +} + +impl<K, V, S, A: Allocator> HashMap<K, V, S, A> { + /// Creates a raw entry builder for the HashMap. + /// + /// Raw entries provide the lowest level of control for searching and + /// manipulating a map. They must be manually initialized with a hash and + /// then manually searched. After this, insertions into a vacant entry + /// still require an owned key to be provided. + /// + /// Raw entries are useful for such exotic situations as: + /// + /// * Hash memoization + /// * Deferring the creation of an owned key until it is known to be required + /// * Using a search key that doesn't work with the Borrow trait + /// * Using custom comparison logic without newtype wrappers + /// + /// Because raw entries provide much more low-level control, it's much easier + /// to put the HashMap into an inconsistent state which, while memory-safe, + /// will cause the map to produce seemingly random results. Higher-level and + /// more foolproof APIs like `entry` should be preferred when possible. + /// + /// In particular, the hash used to initialized the raw entry must still be + /// consistent with the hash of the key that is ultimately stored in the entry. + /// This is because implementations of HashMap may need to recompute hashes + /// when resizing, at which point only the keys are available. + /// + /// Raw entries give mutable access to the keys. This must not be used + /// to modify how the key would compare or hash, as the map will not re-evaluate + /// where the key should go, meaning the keys may become "lost" if their + /// location does not reflect their state. For instance, if you change a key + /// so that the map now contains keys which compare equal, search may start + /// acting erratically, with two keys randomly masking each other. Implementations + /// are free to assume this doesn't happen (within the limits of memory-safety). + /// + /// # Examples + /// + /// ``` + /// use core::hash::{BuildHasher, Hash}; + /// use hashbrown::hash_map::{HashMap, RawEntryMut}; + /// + /// let mut map = HashMap::new(); + /// map.extend([("a", 100), ("b", 200), ("c", 300)]); + /// + /// fn compute_hash<K: Hash + ?Sized, S: BuildHasher>(hash_builder: &S, key: &K) -> u64 { + /// use core::hash::Hasher; + /// let mut state = hash_builder.build_hasher(); + /// key.hash(&mut state); + /// state.finish() + /// } + /// + /// // Existing key (insert and update) + /// match map.raw_entry_mut().from_key(&"a") { + /// RawEntryMut::Vacant(_) => unreachable!(), + /// RawEntryMut::Occupied(mut view) => { + /// assert_eq!(view.get(), &100); + /// let v = view.get_mut(); + /// let new_v = (*v) * 10; + /// *v = new_v; + /// assert_eq!(view.insert(1111), 1000); + /// } + /// } + /// + /// assert_eq!(map[&"a"], 1111); + /// assert_eq!(map.len(), 3); + /// + /// // Existing key (take) + /// let hash = compute_hash(map.hasher(), &"c"); + /// match map.raw_entry_mut().from_key_hashed_nocheck(hash, &"c") { + /// RawEntryMut::Vacant(_) => unreachable!(), + /// RawEntryMut::Occupied(view) => { + /// assert_eq!(view.remove_entry(), ("c", 300)); + /// } + /// } + /// assert_eq!(map.raw_entry().from_key(&"c"), None); + /// assert_eq!(map.len(), 2); + /// + /// // Nonexistent key (insert and update) + /// let key = "d"; + /// let hash = compute_hash(map.hasher(), &key); + /// match map.raw_entry_mut().from_hash(hash, |q| *q == key) { + /// RawEntryMut::Occupied(_) => unreachable!(), + /// RawEntryMut::Vacant(view) => { + /// let (k, value) = view.insert("d", 4000); + /// assert_eq!((*k, *value), ("d", 4000)); + /// *value = 40000; + /// } + /// } + /// assert_eq!(map[&"d"], 40000); + /// assert_eq!(map.len(), 3); + /// + /// match map.raw_entry_mut().from_hash(hash, |q| *q == key) { + /// RawEntryMut::Vacant(_) => unreachable!(), + /// RawEntryMut::Occupied(view) => { + /// assert_eq!(view.remove_entry(), ("d", 40000)); + /// } + /// } + /// assert_eq!(map.get(&"d"), None); + /// assert_eq!(map.len(), 2); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn raw_entry_mut(&mut self) -> RawEntryBuilderMut<'_, K, V, S, A> { + RawEntryBuilderMut { map: self } + } + + /// Creates a raw immutable entry builder for the HashMap. + /// + /// Raw entries provide the lowest level of control for searching and + /// manipulating a map. They must be manually initialized with a hash and + /// then manually searched. + /// + /// This is useful for + /// * Hash memoization + /// * Using a search key that doesn't work with the Borrow trait + /// * Using custom comparison logic without newtype wrappers + /// + /// Unless you are in such a situation, higher-level and more foolproof APIs like + /// `get` should be preferred. + /// + /// Immutable raw entries have very limited use; you might instead want `raw_entry_mut`. + /// + /// # Examples + /// + /// ``` + /// use core::hash::{BuildHasher, Hash}; + /// use hashbrown::HashMap; + /// + /// let mut map = HashMap::new(); + /// map.extend([("a", 100), ("b", 200), ("c", 300)]); + /// + /// fn compute_hash<K: Hash + ?Sized, S: BuildHasher>(hash_builder: &S, key: &K) -> u64 { + /// use core::hash::Hasher; + /// let mut state = hash_builder.build_hasher(); + /// key.hash(&mut state); + /// state.finish() + /// } + /// + /// for k in ["a", "b", "c", "d", "e", "f"] { + /// let hash = compute_hash(map.hasher(), k); + /// let v = map.get(&k).cloned(); + /// let kv = v.as_ref().map(|v| (&k, v)); + /// + /// println!("Key: {} and value: {:?}", k, v); + /// + /// assert_eq!(map.raw_entry().from_key(&k), kv); + /// assert_eq!(map.raw_entry().from_hash(hash, |q| *q == k), kv); + /// assert_eq!(map.raw_entry().from_key_hashed_nocheck(hash, &k), kv); + /// } + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn raw_entry(&self) -> RawEntryBuilder<'_, K, V, S, A> { + RawEntryBuilder { map: self } + } + + /// Returns a reference to the [`RawTable`] used underneath [`HashMap`]. + /// This function is only available if the `raw` feature of the crate is enabled. + /// + /// See [`raw_table_mut`] for more. + /// + /// [`raw_table_mut`]: Self::raw_table_mut + #[cfg(feature = "raw")] + #[cfg_attr(feature = "inline-more", inline)] + pub fn raw_table(&self) -> &RawTable<(K, V), A> { + &self.table + } + + /// Returns a mutable reference to the [`RawTable`] used underneath [`HashMap`]. + /// This function is only available if the `raw` feature of the crate is enabled. + /// + /// # Note + /// + /// Calling this function is safe, but using the raw hash table API may require + /// unsafe functions or blocks. + /// + /// `RawTable` API gives the lowest level of control under the map that can be useful + /// for extending the HashMap's API, but may lead to *[undefined behavior]*. + /// + /// [`HashMap`]: struct.HashMap.html + /// [`RawTable`]: crate::raw::RawTable + /// [undefined behavior]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html + /// + /// # Examples + /// + /// ``` + /// use core::hash::{BuildHasher, Hash}; + /// use hashbrown::HashMap; + /// + /// let mut map = HashMap::new(); + /// map.extend([("a", 10), ("b", 20), ("c", 30)]); + /// assert_eq!(map.len(), 3); + /// + /// // Let's imagine that we have a value and a hash of the key, but not the key itself. + /// // However, if you want to remove the value from the map by hash and value, and you + /// // know exactly that the value is unique, then you can create a function like this: + /// fn remove_by_hash<K, V, S, F>( + /// map: &mut HashMap<K, V, S>, + /// hash: u64, + /// is_match: F, + /// ) -> Option<(K, V)> + /// where + /// F: Fn(&(K, V)) -> bool, + /// { + /// let raw_table = map.raw_table_mut(); + /// match raw_table.find(hash, is_match) { + /// Some(bucket) => Some(unsafe { raw_table.remove(bucket).0 }), + /// None => None, + /// } + /// } + /// + /// fn compute_hash<K: Hash + ?Sized, S: BuildHasher>(hash_builder: &S, key: &K) -> u64 { + /// use core::hash::Hasher; + /// let mut state = hash_builder.build_hasher(); + /// key.hash(&mut state); + /// state.finish() + /// } + /// + /// let hash = compute_hash(map.hasher(), "a"); + /// assert_eq!(remove_by_hash(&mut map, hash, |(_, v)| *v == 10), Some(("a", 10))); + /// assert_eq!(map.get(&"a"), None); + /// assert_eq!(map.len(), 2); + /// ``` + #[cfg(feature = "raw")] + #[cfg_attr(feature = "inline-more", inline)] + pub fn raw_table_mut(&mut self) -> &mut RawTable<(K, V), A> { + &mut self.table + } +} + +impl<K, V, S, A> PartialEq for HashMap<K, V, S, A> +where + K: Eq + Hash, + V: PartialEq, + S: BuildHasher, + A: Allocator, +{ + fn eq(&self, other: &Self) -> bool { + if self.len() != other.len() { + return false; + } + + self.iter() + .all(|(key, value)| other.get(key).map_or(false, |v| *value == *v)) + } +} + +impl<K, V, S, A> Eq for HashMap<K, V, S, A> +where + K: Eq + Hash, + V: Eq, + S: BuildHasher, + A: Allocator, +{ +} + +impl<K, V, S, A> Debug for HashMap<K, V, S, A> +where + K: Debug, + V: Debug, + A: Allocator, +{ + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + f.debug_map().entries(self.iter()).finish() + } +} + +impl<K, V, S, A> Default for HashMap<K, V, S, A> +where + S: Default, + A: Default + Allocator, +{ + /// Creates an empty `HashMap<K, V, S, A>`, with the `Default` value for the hasher and allocator. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// use std::collections::hash_map::RandomState; + /// + /// // You can specify all types of HashMap, including hasher and allocator. + /// // Created map is empty and don't allocate memory + /// let map: HashMap<u32, String> = Default::default(); + /// assert_eq!(map.capacity(), 0); + /// let map: HashMap<u32, String, RandomState> = HashMap::default(); + /// assert_eq!(map.capacity(), 0); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + fn default() -> Self { + Self::with_hasher_in(Default::default(), Default::default()) + } +} + +impl<K, Q: ?Sized, V, S, A> Index<&Q> for HashMap<K, V, S, A> +where + K: Eq + Hash, + Q: Hash + Equivalent<K>, + S: BuildHasher, + A: Allocator, +{ + type Output = V; + + /// Returns a reference to the value corresponding to the supplied key. + /// + /// # Panics + /// + /// Panics if the key is not present in the `HashMap`. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let map: HashMap<_, _> = [("a", "One"), ("b", "Two")].into(); + /// + /// assert_eq!(map[&"a"], "One"); + /// assert_eq!(map[&"b"], "Two"); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + fn index(&self, key: &Q) -> &V { + self.get(key).expect("no entry found for key") + } +} + +// The default hasher is used to match the std implementation signature +#[cfg(feature = "ahash")] +impl<K, V, A, const N: usize> From<[(K, V); N]> for HashMap<K, V, DefaultHashBuilder, A> +where + K: Eq + Hash, + A: Default + Allocator, +{ + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let map1 = HashMap::from([(1, 2), (3, 4)]); + /// let map2: HashMap<_, _> = [(1, 2), (3, 4)].into(); + /// assert_eq!(map1, map2); + /// ``` + fn from(arr: [(K, V); N]) -> Self { + arr.into_iter().collect() + } +} + +/// An iterator over the entries of a `HashMap` in arbitrary order. +/// The iterator element type is `(&'a K, &'a V)`. +/// +/// This `struct` is created by the [`iter`] method on [`HashMap`]. See its +/// documentation for more. +/// +/// [`iter`]: struct.HashMap.html#method.iter +/// [`HashMap`]: struct.HashMap.html +/// +/// # Examples +/// +/// ``` +/// use hashbrown::HashMap; +/// +/// let map: HashMap<_, _> = [(1, "a"), (2, "b"), (3, "c")].into(); +/// +/// let mut iter = map.iter(); +/// let mut vec = vec![iter.next(), iter.next(), iter.next()]; +/// +/// // The `Iter` iterator produces items in arbitrary order, so the +/// // items must be sorted to test them against a sorted array. +/// vec.sort_unstable(); +/// assert_eq!(vec, [Some((&1, &"a")), Some((&2, &"b")), Some((&3, &"c"))]); +/// +/// // It is fused iterator +/// assert_eq!(iter.next(), None); +/// assert_eq!(iter.next(), None); +/// ``` +pub struct Iter<'a, K, V> { + inner: RawIter<(K, V)>, + marker: PhantomData<(&'a K, &'a V)>, +} + +// FIXME(#26925) Remove in favor of `#[derive(Clone)]` +impl<K, V> Clone for Iter<'_, K, V> { + #[cfg_attr(feature = "inline-more", inline)] + fn clone(&self) -> Self { + Iter { + inner: self.inner.clone(), + marker: PhantomData, + } + } +} + +impl<K: Debug, V: Debug> fmt::Debug for Iter<'_, K, V> { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + f.debug_list().entries(self.clone()).finish() + } +} + +/// A mutable iterator over the entries of a `HashMap` in arbitrary order. +/// The iterator element type is `(&'a K, &'a mut V)`. +/// +/// This `struct` is created by the [`iter_mut`] method on [`HashMap`]. See its +/// documentation for more. +/// +/// [`iter_mut`]: struct.HashMap.html#method.iter_mut +/// [`HashMap`]: struct.HashMap.html +/// +/// # Examples +/// +/// ``` +/// use hashbrown::HashMap; +/// +/// let mut map: HashMap<_, _> = [(1, "One".to_owned()), (2, "Two".into())].into(); +/// +/// let mut iter = map.iter_mut(); +/// iter.next().map(|(_, v)| v.push_str(" Mississippi")); +/// iter.next().map(|(_, v)| v.push_str(" Mississippi")); +/// +/// // It is fused iterator +/// assert_eq!(iter.next(), None); +/// assert_eq!(iter.next(), None); +/// +/// assert_eq!(map.get(&1).unwrap(), &"One Mississippi".to_owned()); +/// assert_eq!(map.get(&2).unwrap(), &"Two Mississippi".to_owned()); +/// ``` +pub struct IterMut<'a, K, V> { + inner: RawIter<(K, V)>, + // To ensure invariance with respect to V + marker: PhantomData<(&'a K, &'a mut V)>, +} + +// We override the default Send impl which has K: Sync instead of K: Send. Both +// are correct, but this one is more general since it allows keys which +// implement Send but not Sync. +unsafe impl<K: Send, V: Send> Send for IterMut<'_, K, V> {} + +impl<K, V> IterMut<'_, K, V> { + /// Returns a iterator of references over the remaining items. + #[cfg_attr(feature = "inline-more", inline)] + pub(super) fn iter(&self) -> Iter<'_, K, V> { + Iter { + inner: self.inner.clone(), + marker: PhantomData, + } + } +} + +/// An owning iterator over the entries of a `HashMap` in arbitrary order. +/// The iterator element type is `(K, V)`. +/// +/// This `struct` is created by the [`into_iter`] method on [`HashMap`] +/// (provided by the [`IntoIterator`] trait). See its documentation for more. +/// The map cannot be used after calling that method. +/// +/// [`into_iter`]: struct.HashMap.html#method.into_iter +/// [`HashMap`]: struct.HashMap.html +/// [`IntoIterator`]: https://doc.rust-lang.org/core/iter/trait.IntoIterator.html +/// +/// # Examples +/// +/// ``` +/// use hashbrown::HashMap; +/// +/// let map: HashMap<_, _> = [(1, "a"), (2, "b"), (3, "c")].into(); +/// +/// let mut iter = map.into_iter(); +/// let mut vec = vec![iter.next(), iter.next(), iter.next()]; +/// +/// // The `IntoIter` iterator produces items in arbitrary order, so the +/// // items must be sorted to test them against a sorted array. +/// vec.sort_unstable(); +/// assert_eq!(vec, [Some((1, "a")), Some((2, "b")), Some((3, "c"))]); +/// +/// // It is fused iterator +/// assert_eq!(iter.next(), None); +/// assert_eq!(iter.next(), None); +/// ``` +pub struct IntoIter<K, V, A: Allocator = Global> { + inner: RawIntoIter<(K, V), A>, +} + +impl<K, V, A: Allocator> IntoIter<K, V, A> { + /// Returns a iterator of references over the remaining items. + #[cfg_attr(feature = "inline-more", inline)] + pub(super) fn iter(&self) -> Iter<'_, K, V> { + Iter { + inner: self.inner.iter(), + marker: PhantomData, + } + } +} + +/// An owning iterator over the keys of a `HashMap` in arbitrary order. +/// The iterator element type is `K`. +/// +/// This `struct` is created by the [`into_keys`] method on [`HashMap`]. +/// See its documentation for more. +/// The map cannot be used after calling that method. +/// +/// [`into_keys`]: struct.HashMap.html#method.into_keys +/// [`HashMap`]: struct.HashMap.html +/// +/// # Examples +/// +/// ``` +/// use hashbrown::HashMap; +/// +/// let map: HashMap<_, _> = [(1, "a"), (2, "b"), (3, "c")].into(); +/// +/// let mut keys = map.into_keys(); +/// let mut vec = vec![keys.next(), keys.next(), keys.next()]; +/// +/// // The `IntoKeys` iterator produces keys in arbitrary order, so the +/// // keys must be sorted to test them against a sorted array. +/// vec.sort_unstable(); +/// assert_eq!(vec, [Some(1), Some(2), Some(3)]); +/// +/// // It is fused iterator +/// assert_eq!(keys.next(), None); +/// assert_eq!(keys.next(), None); +/// ``` +pub struct IntoKeys<K, V, A: Allocator = Global> { + inner: IntoIter<K, V, A>, +} + +impl<K, V, A: Allocator> Iterator for IntoKeys<K, V, A> { + type Item = K; + + #[inline] + fn next(&mut self) -> Option<K> { + self.inner.next().map(|(k, _)| k) + } + #[inline] + fn size_hint(&self) -> (usize, Option<usize>) { + self.inner.size_hint() + } +} + +impl<K, V, A: Allocator> ExactSizeIterator for IntoKeys<K, V, A> { + #[inline] + fn len(&self) -> usize { + self.inner.len() + } +} + +impl<K, V, A: Allocator> FusedIterator for IntoKeys<K, V, A> {} + +impl<K: Debug, V: Debug, A: Allocator> fmt::Debug for IntoKeys<K, V, A> { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + f.debug_list() + .entries(self.inner.iter().map(|(k, _)| k)) + .finish() + } +} + +/// An owning iterator over the values of a `HashMap` in arbitrary order. +/// The iterator element type is `V`. +/// +/// This `struct` is created by the [`into_values`] method on [`HashMap`]. +/// See its documentation for more. The map cannot be used after calling that method. +/// +/// [`into_values`]: struct.HashMap.html#method.into_values +/// [`HashMap`]: struct.HashMap.html +/// +/// # Examples +/// +/// ``` +/// use hashbrown::HashMap; +/// +/// let map: HashMap<_, _> = [(1, "a"), (2, "b"), (3, "c")].into(); +/// +/// let mut values = map.into_values(); +/// let mut vec = vec![values.next(), values.next(), values.next()]; +/// +/// // The `IntoValues` iterator produces values in arbitrary order, so +/// // the values must be sorted to test them against a sorted array. +/// vec.sort_unstable(); +/// assert_eq!(vec, [Some("a"), Some("b"), Some("c")]); +/// +/// // It is fused iterator +/// assert_eq!(values.next(), None); +/// assert_eq!(values.next(), None); +/// ``` +pub struct IntoValues<K, V, A: Allocator = Global> { + inner: IntoIter<K, V, A>, +} + +impl<K, V, A: Allocator> Iterator for IntoValues<K, V, A> { + type Item = V; + + #[inline] + fn next(&mut self) -> Option<V> { + self.inner.next().map(|(_, v)| v) + } + #[inline] + fn size_hint(&self) -> (usize, Option<usize>) { + self.inner.size_hint() + } +} + +impl<K, V, A: Allocator> ExactSizeIterator for IntoValues<K, V, A> { + #[inline] + fn len(&self) -> usize { + self.inner.len() + } +} + +impl<K, V, A: Allocator> FusedIterator for IntoValues<K, V, A> {} + +impl<K, V: Debug, A: Allocator> fmt::Debug for IntoValues<K, V, A> { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + f.debug_list() + .entries(self.inner.iter().map(|(_, v)| v)) + .finish() + } +} + +/// An iterator over the keys of a `HashMap` in arbitrary order. +/// The iterator element type is `&'a K`. +/// +/// This `struct` is created by the [`keys`] method on [`HashMap`]. See its +/// documentation for more. +/// +/// [`keys`]: struct.HashMap.html#method.keys +/// [`HashMap`]: struct.HashMap.html +/// +/// # Examples +/// +/// ``` +/// use hashbrown::HashMap; +/// +/// let map: HashMap<_, _> = [(1, "a"), (2, "b"), (3, "c")].into(); +/// +/// let mut keys = map.keys(); +/// let mut vec = vec![keys.next(), keys.next(), keys.next()]; +/// +/// // The `Keys` iterator produces keys in arbitrary order, so the +/// // keys must be sorted to test them against a sorted array. +/// vec.sort_unstable(); +/// assert_eq!(vec, [Some(&1), Some(&2), Some(&3)]); +/// +/// // It is fused iterator +/// assert_eq!(keys.next(), None); +/// assert_eq!(keys.next(), None); +/// ``` +pub struct Keys<'a, K, V> { + inner: Iter<'a, K, V>, +} + +// FIXME(#26925) Remove in favor of `#[derive(Clone)]` +impl<K, V> Clone for Keys<'_, K, V> { + #[cfg_attr(feature = "inline-more", inline)] + fn clone(&self) -> Self { + Keys { + inner: self.inner.clone(), + } + } +} + +impl<K: Debug, V> fmt::Debug for Keys<'_, K, V> { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + f.debug_list().entries(self.clone()).finish() + } +} + +/// An iterator over the values of a `HashMap` in arbitrary order. +/// The iterator element type is `&'a V`. +/// +/// This `struct` is created by the [`values`] method on [`HashMap`]. See its +/// documentation for more. +/// +/// [`values`]: struct.HashMap.html#method.values +/// [`HashMap`]: struct.HashMap.html +/// +/// # Examples +/// +/// ``` +/// use hashbrown::HashMap; +/// +/// let map: HashMap<_, _> = [(1, "a"), (2, "b"), (3, "c")].into(); +/// +/// let mut values = map.values(); +/// let mut vec = vec![values.next(), values.next(), values.next()]; +/// +/// // The `Values` iterator produces values in arbitrary order, so the +/// // values must be sorted to test them against a sorted array. +/// vec.sort_unstable(); +/// assert_eq!(vec, [Some(&"a"), Some(&"b"), Some(&"c")]); +/// +/// // It is fused iterator +/// assert_eq!(values.next(), None); +/// assert_eq!(values.next(), None); +/// ``` +pub struct Values<'a, K, V> { + inner: Iter<'a, K, V>, +} + +// FIXME(#26925) Remove in favor of `#[derive(Clone)]` +impl<K, V> Clone for Values<'_, K, V> { + #[cfg_attr(feature = "inline-more", inline)] + fn clone(&self) -> Self { + Values { + inner: self.inner.clone(), + } + } +} + +impl<K, V: Debug> fmt::Debug for Values<'_, K, V> { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + f.debug_list().entries(self.clone()).finish() + } +} + +/// A draining iterator over the entries of a `HashMap` in arbitrary +/// order. The iterator element type is `(K, V)`. +/// +/// This `struct` is created by the [`drain`] method on [`HashMap`]. See its +/// documentation for more. +/// +/// [`drain`]: struct.HashMap.html#method.drain +/// [`HashMap`]: struct.HashMap.html +/// +/// # Examples +/// +/// ``` +/// use hashbrown::HashMap; +/// +/// let mut map: HashMap<_, _> = [(1, "a"), (2, "b"), (3, "c")].into(); +/// +/// let mut drain_iter = map.drain(); +/// let mut vec = vec![drain_iter.next(), drain_iter.next(), drain_iter.next()]; +/// +/// // The `Drain` iterator produces items in arbitrary order, so the +/// // items must be sorted to test them against a sorted array. +/// vec.sort_unstable(); +/// assert_eq!(vec, [Some((1, "a")), Some((2, "b")), Some((3, "c"))]); +/// +/// // It is fused iterator +/// assert_eq!(drain_iter.next(), None); +/// assert_eq!(drain_iter.next(), None); +/// ``` +pub struct Drain<'a, K, V, A: Allocator = Global> { + inner: RawDrain<'a, (K, V), A>, +} + +impl<K, V, A: Allocator> Drain<'_, K, V, A> { + /// Returns a iterator of references over the remaining items. + #[cfg_attr(feature = "inline-more", inline)] + pub(super) fn iter(&self) -> Iter<'_, K, V> { + Iter { + inner: self.inner.iter(), + marker: PhantomData, + } + } +} + +/// A draining iterator over entries of a `HashMap` which don't satisfy the predicate +/// `f(&k, &mut v)` in arbitrary order. The iterator element type is `(K, V)`. +/// +/// This `struct` is created by the [`extract_if`] method on [`HashMap`]. See its +/// documentation for more. +/// +/// [`extract_if`]: struct.HashMap.html#method.extract_if +/// [`HashMap`]: struct.HashMap.html +/// +/// # Examples +/// +/// ``` +/// use hashbrown::HashMap; +/// +/// let mut map: HashMap<i32, &str> = [(1, "a"), (2, "b"), (3, "c")].into(); +/// +/// let mut extract_if = map.extract_if(|k, _v| k % 2 != 0); +/// let mut vec = vec![extract_if.next(), extract_if.next()]; +/// +/// // The `ExtractIf` iterator produces items in arbitrary order, so the +/// // items must be sorted to test them against a sorted array. +/// vec.sort_unstable(); +/// assert_eq!(vec, [Some((1, "a")),Some((3, "c"))]); +/// +/// // It is fused iterator +/// assert_eq!(extract_if.next(), None); +/// assert_eq!(extract_if.next(), None); +/// drop(extract_if); +/// +/// assert_eq!(map.len(), 1); +/// ``` +#[must_use = "Iterators are lazy unless consumed"] +pub struct ExtractIf<'a, K, V, F, A: Allocator = Global> +where + F: FnMut(&K, &mut V) -> bool, +{ + f: F, + inner: ExtractIfInner<'a, K, V, A>, +} + +impl<K, V, F, A> Iterator for ExtractIf<'_, K, V, F, A> +where + F: FnMut(&K, &mut V) -> bool, + A: Allocator, +{ + type Item = (K, V); + + #[cfg_attr(feature = "inline-more", inline)] + fn next(&mut self) -> Option<Self::Item> { + self.inner.next(&mut self.f) + } + + #[inline] + fn size_hint(&self) -> (usize, Option<usize>) { + (0, self.inner.iter.size_hint().1) + } +} + +impl<K, V, F> FusedIterator for ExtractIf<'_, K, V, F> where F: FnMut(&K, &mut V) -> bool {} + +/// Portions of `ExtractIf` shared with `set::ExtractIf` +pub(super) struct ExtractIfInner<'a, K, V, A: Allocator> { + pub iter: RawIter<(K, V)>, + pub table: &'a mut RawTable<(K, V), A>, +} + +impl<K, V, A: Allocator> ExtractIfInner<'_, K, V, A> { + #[cfg_attr(feature = "inline-more", inline)] + pub(super) fn next<F>(&mut self, f: &mut F) -> Option<(K, V)> + where + F: FnMut(&K, &mut V) -> bool, + { + unsafe { + for item in &mut self.iter { + let &mut (ref key, ref mut value) = item.as_mut(); + if f(key, value) { + return Some(self.table.remove(item).0); + } + } + } + None + } +} + +/// A mutable iterator over the values of a `HashMap` in arbitrary order. +/// The iterator element type is `&'a mut V`. +/// +/// This `struct` is created by the [`values_mut`] method on [`HashMap`]. See its +/// documentation for more. +/// +/// [`values_mut`]: struct.HashMap.html#method.values_mut +/// [`HashMap`]: struct.HashMap.html +/// +/// # Examples +/// +/// ``` +/// use hashbrown::HashMap; +/// +/// let mut map: HashMap<_, _> = [(1, "One".to_owned()), (2, "Two".into())].into(); +/// +/// let mut values = map.values_mut(); +/// values.next().map(|v| v.push_str(" Mississippi")); +/// values.next().map(|v| v.push_str(" Mississippi")); +/// +/// // It is fused iterator +/// assert_eq!(values.next(), None); +/// assert_eq!(values.next(), None); +/// +/// assert_eq!(map.get(&1).unwrap(), &"One Mississippi".to_owned()); +/// assert_eq!(map.get(&2).unwrap(), &"Two Mississippi".to_owned()); +/// ``` +pub struct ValuesMut<'a, K, V> { + inner: IterMut<'a, K, V>, +} + +/// A builder for computing where in a [`HashMap`] a key-value pair would be stored. +/// +/// See the [`HashMap::raw_entry_mut`] docs for usage examples. +/// +/// [`HashMap::raw_entry_mut`]: struct.HashMap.html#method.raw_entry_mut +/// +/// # Examples +/// +/// ``` +/// use hashbrown::hash_map::{RawEntryBuilderMut, RawEntryMut::Vacant, RawEntryMut::Occupied}; +/// use hashbrown::HashMap; +/// use core::hash::{BuildHasher, Hash}; +/// +/// let mut map = HashMap::new(); +/// map.extend([(1, 11), (2, 12), (3, 13), (4, 14), (5, 15), (6, 16)]); +/// assert_eq!(map.len(), 6); +/// +/// fn compute_hash<K: Hash + ?Sized, S: BuildHasher>(hash_builder: &S, key: &K) -> u64 { +/// use core::hash::Hasher; +/// let mut state = hash_builder.build_hasher(); +/// key.hash(&mut state); +/// state.finish() +/// } +/// +/// let builder: RawEntryBuilderMut<_, _, _> = map.raw_entry_mut(); +/// +/// // Existing key +/// match builder.from_key(&6) { +/// Vacant(_) => unreachable!(), +/// Occupied(view) => assert_eq!(view.get(), &16), +/// } +/// +/// for key in 0..12 { +/// let hash = compute_hash(map.hasher(), &key); +/// let value = map.get(&key).cloned(); +/// let key_value = value.as_ref().map(|v| (&key, v)); +/// +/// println!("Key: {} and value: {:?}", key, value); +/// +/// match map.raw_entry_mut().from_key(&key) { +/// Occupied(mut o) => assert_eq!(Some(o.get_key_value()), key_value), +/// Vacant(_) => assert_eq!(value, None), +/// } +/// match map.raw_entry_mut().from_key_hashed_nocheck(hash, &key) { +/// Occupied(mut o) => assert_eq!(Some(o.get_key_value()), key_value), +/// Vacant(_) => assert_eq!(value, None), +/// } +/// match map.raw_entry_mut().from_hash(hash, |q| *q == key) { +/// Occupied(mut o) => assert_eq!(Some(o.get_key_value()), key_value), +/// Vacant(_) => assert_eq!(value, None), +/// } +/// } +/// +/// assert_eq!(map.len(), 6); +/// ``` +pub struct RawEntryBuilderMut<'a, K, V, S, A: Allocator = Global> { + map: &'a mut HashMap<K, V, S, A>, +} + +/// A view into a single entry in a map, which may either be vacant or occupied. +/// +/// This is a lower-level version of [`Entry`]. +/// +/// This `enum` is constructed through the [`raw_entry_mut`] method on [`HashMap`], +/// then calling one of the methods of that [`RawEntryBuilderMut`]. +/// +/// [`HashMap`]: struct.HashMap.html +/// [`Entry`]: enum.Entry.html +/// [`raw_entry_mut`]: struct.HashMap.html#method.raw_entry_mut +/// [`RawEntryBuilderMut`]: struct.RawEntryBuilderMut.html +/// +/// # Examples +/// +/// ``` +/// use core::hash::{BuildHasher, Hash}; +/// use hashbrown::hash_map::{HashMap, RawEntryMut, RawOccupiedEntryMut}; +/// +/// let mut map = HashMap::new(); +/// map.extend([('a', 1), ('b', 2), ('c', 3)]); +/// assert_eq!(map.len(), 3); +/// +/// fn compute_hash<K: Hash + ?Sized, S: BuildHasher>(hash_builder: &S, key: &K) -> u64 { +/// use core::hash::Hasher; +/// let mut state = hash_builder.build_hasher(); +/// key.hash(&mut state); +/// state.finish() +/// } +/// +/// // Existing key (insert) +/// let raw: RawEntryMut<_, _, _> = map.raw_entry_mut().from_key(&'a'); +/// let _raw_o: RawOccupiedEntryMut<_, _, _> = raw.insert('a', 10); +/// assert_eq!(map.len(), 3); +/// +/// // Nonexistent key (insert) +/// map.raw_entry_mut().from_key(&'d').insert('d', 40); +/// assert_eq!(map.len(), 4); +/// +/// // Existing key (or_insert) +/// let hash = compute_hash(map.hasher(), &'b'); +/// let kv = map +/// .raw_entry_mut() +/// .from_key_hashed_nocheck(hash, &'b') +/// .or_insert('b', 20); +/// assert_eq!(kv, (&mut 'b', &mut 2)); +/// *kv.1 = 20; +/// assert_eq!(map.len(), 4); +/// +/// // Nonexistent key (or_insert) +/// let hash = compute_hash(map.hasher(), &'e'); +/// let kv = map +/// .raw_entry_mut() +/// .from_key_hashed_nocheck(hash, &'e') +/// .or_insert('e', 50); +/// assert_eq!(kv, (&mut 'e', &mut 50)); +/// assert_eq!(map.len(), 5); +/// +/// // Existing key (or_insert_with) +/// let hash = compute_hash(map.hasher(), &'c'); +/// let kv = map +/// .raw_entry_mut() +/// .from_hash(hash, |q| q == &'c') +/// .or_insert_with(|| ('c', 30)); +/// assert_eq!(kv, (&mut 'c', &mut 3)); +/// *kv.1 = 30; +/// assert_eq!(map.len(), 5); +/// +/// // Nonexistent key (or_insert_with) +/// let hash = compute_hash(map.hasher(), &'f'); +/// let kv = map +/// .raw_entry_mut() +/// .from_hash(hash, |q| q == &'f') +/// .or_insert_with(|| ('f', 60)); +/// assert_eq!(kv, (&mut 'f', &mut 60)); +/// assert_eq!(map.len(), 6); +/// +/// println!("Our HashMap: {:?}", map); +/// +/// let mut vec: Vec<_> = map.iter().map(|(&k, &v)| (k, v)).collect(); +/// // The `Iter` iterator produces items in arbitrary order, so the +/// // items must be sorted to test them against a sorted array. +/// vec.sort_unstable(); +/// assert_eq!(vec, [('a', 10), ('b', 20), ('c', 30), ('d', 40), ('e', 50), ('f', 60)]); +/// ``` +pub enum RawEntryMut<'a, K, V, S, A: Allocator = Global> { + /// An occupied entry. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::{hash_map::RawEntryMut, HashMap}; + /// let mut map: HashMap<_, _> = [("a", 100), ("b", 200)].into(); + /// + /// match map.raw_entry_mut().from_key(&"a") { + /// RawEntryMut::Vacant(_) => unreachable!(), + /// RawEntryMut::Occupied(_) => { } + /// } + /// ``` + Occupied(RawOccupiedEntryMut<'a, K, V, S, A>), + /// A vacant entry. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::{hash_map::RawEntryMut, HashMap}; + /// let mut map: HashMap<&str, i32> = HashMap::new(); + /// + /// match map.raw_entry_mut().from_key("a") { + /// RawEntryMut::Occupied(_) => unreachable!(), + /// RawEntryMut::Vacant(_) => { } + /// } + /// ``` + Vacant(RawVacantEntryMut<'a, K, V, S, A>), +} + +/// A view into an occupied entry in a `HashMap`. +/// It is part of the [`RawEntryMut`] enum. +/// +/// [`RawEntryMut`]: enum.RawEntryMut.html +/// +/// # Examples +/// +/// ``` +/// use core::hash::{BuildHasher, Hash}; +/// use hashbrown::hash_map::{HashMap, RawEntryMut, RawOccupiedEntryMut}; +/// +/// let mut map = HashMap::new(); +/// map.extend([("a", 10), ("b", 20), ("c", 30)]); +/// +/// fn compute_hash<K: Hash + ?Sized, S: BuildHasher>(hash_builder: &S, key: &K) -> u64 { +/// use core::hash::Hasher; +/// let mut state = hash_builder.build_hasher(); +/// key.hash(&mut state); +/// state.finish() +/// } +/// +/// let _raw_o: RawOccupiedEntryMut<_, _, _> = map.raw_entry_mut().from_key(&"a").insert("a", 100); +/// assert_eq!(map.len(), 3); +/// +/// // Existing key (insert and update) +/// match map.raw_entry_mut().from_key(&"a") { +/// RawEntryMut::Vacant(_) => unreachable!(), +/// RawEntryMut::Occupied(mut view) => { +/// assert_eq!(view.get(), &100); +/// let v = view.get_mut(); +/// let new_v = (*v) * 10; +/// *v = new_v; +/// assert_eq!(view.insert(1111), 1000); +/// } +/// } +/// +/// assert_eq!(map[&"a"], 1111); +/// assert_eq!(map.len(), 3); +/// +/// // Existing key (take) +/// let hash = compute_hash(map.hasher(), &"c"); +/// match map.raw_entry_mut().from_key_hashed_nocheck(hash, &"c") { +/// RawEntryMut::Vacant(_) => unreachable!(), +/// RawEntryMut::Occupied(view) => { +/// assert_eq!(view.remove_entry(), ("c", 30)); +/// } +/// } +/// assert_eq!(map.raw_entry().from_key(&"c"), None); +/// assert_eq!(map.len(), 2); +/// +/// let hash = compute_hash(map.hasher(), &"b"); +/// match map.raw_entry_mut().from_hash(hash, |q| *q == "b") { +/// RawEntryMut::Vacant(_) => unreachable!(), +/// RawEntryMut::Occupied(view) => { +/// assert_eq!(view.remove_entry(), ("b", 20)); +/// } +/// } +/// assert_eq!(map.get(&"b"), None); +/// assert_eq!(map.len(), 1); +/// ``` +pub struct RawOccupiedEntryMut<'a, K, V, S, A: Allocator = Global> { + elem: Bucket<(K, V)>, + table: &'a mut RawTable<(K, V), A>, + hash_builder: &'a S, +} + +unsafe impl<K, V, S, A> Send for RawOccupiedEntryMut<'_, K, V, S, A> +where + K: Send, + V: Send, + S: Send, + A: Send + Allocator, +{ +} +unsafe impl<K, V, S, A> Sync for RawOccupiedEntryMut<'_, K, V, S, A> +where + K: Sync, + V: Sync, + S: Sync, + A: Sync + Allocator, +{ +} + +/// A view into a vacant entry in a `HashMap`. +/// It is part of the [`RawEntryMut`] enum. +/// +/// [`RawEntryMut`]: enum.RawEntryMut.html +/// +/// # Examples +/// +/// ``` +/// use core::hash::{BuildHasher, Hash}; +/// use hashbrown::hash_map::{HashMap, RawEntryMut, RawVacantEntryMut}; +/// +/// let mut map = HashMap::<&str, i32>::new(); +/// +/// fn compute_hash<K: Hash + ?Sized, S: BuildHasher>(hash_builder: &S, key: &K) -> u64 { +/// use core::hash::Hasher; +/// let mut state = hash_builder.build_hasher(); +/// key.hash(&mut state); +/// state.finish() +/// } +/// +/// let raw_v: RawVacantEntryMut<_, _, _> = match map.raw_entry_mut().from_key(&"a") { +/// RawEntryMut::Vacant(view) => view, +/// RawEntryMut::Occupied(_) => unreachable!(), +/// }; +/// raw_v.insert("a", 10); +/// assert!(map[&"a"] == 10 && map.len() == 1); +/// +/// // Nonexistent key (insert and update) +/// let hash = compute_hash(map.hasher(), &"b"); +/// match map.raw_entry_mut().from_key_hashed_nocheck(hash, &"b") { +/// RawEntryMut::Occupied(_) => unreachable!(), +/// RawEntryMut::Vacant(view) => { +/// let (k, value) = view.insert("b", 2); +/// assert_eq!((*k, *value), ("b", 2)); +/// *value = 20; +/// } +/// } +/// assert!(map[&"b"] == 20 && map.len() == 2); +/// +/// let hash = compute_hash(map.hasher(), &"c"); +/// match map.raw_entry_mut().from_hash(hash, |q| *q == "c") { +/// RawEntryMut::Occupied(_) => unreachable!(), +/// RawEntryMut::Vacant(view) => { +/// assert_eq!(view.insert("c", 30), (&mut "c", &mut 30)); +/// } +/// } +/// assert!(map[&"c"] == 30 && map.len() == 3); +/// ``` +pub struct RawVacantEntryMut<'a, K, V, S, A: Allocator = Global> { + table: &'a mut RawTable<(K, V), A>, + hash_builder: &'a S, +} + +/// A builder for computing where in a [`HashMap`] a key-value pair would be stored. +/// +/// See the [`HashMap::raw_entry`] docs for usage examples. +/// +/// [`HashMap::raw_entry`]: struct.HashMap.html#method.raw_entry +/// +/// # Examples +/// +/// ``` +/// use hashbrown::hash_map::{HashMap, RawEntryBuilder}; +/// use core::hash::{BuildHasher, Hash}; +/// +/// let mut map = HashMap::new(); +/// map.extend([(1, 10), (2, 20), (3, 30)]); +/// +/// fn compute_hash<K: Hash + ?Sized, S: BuildHasher>(hash_builder: &S, key: &K) -> u64 { +/// use core::hash::Hasher; +/// let mut state = hash_builder.build_hasher(); +/// key.hash(&mut state); +/// state.finish() +/// } +/// +/// for k in 0..6 { +/// let hash = compute_hash(map.hasher(), &k); +/// let v = map.get(&k).cloned(); +/// let kv = v.as_ref().map(|v| (&k, v)); +/// +/// println!("Key: {} and value: {:?}", k, v); +/// let builder: RawEntryBuilder<_, _, _> = map.raw_entry(); +/// assert_eq!(builder.from_key(&k), kv); +/// assert_eq!(map.raw_entry().from_hash(hash, |q| *q == k), kv); +/// assert_eq!(map.raw_entry().from_key_hashed_nocheck(hash, &k), kv); +/// } +/// ``` +pub struct RawEntryBuilder<'a, K, V, S, A: Allocator = Global> { + map: &'a HashMap<K, V, S, A>, +} + +impl<'a, K, V, S, A: Allocator> RawEntryBuilderMut<'a, K, V, S, A> { + /// Creates a `RawEntryMut` from the given key. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::hash_map::{HashMap, RawEntryMut}; + /// + /// let mut map: HashMap<&str, u32> = HashMap::new(); + /// let key = "a"; + /// let entry: RawEntryMut<&str, u32, _> = map.raw_entry_mut().from_key(&key); + /// entry.insert(key, 100); + /// assert_eq!(map[&"a"], 100); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + #[allow(clippy::wrong_self_convention)] + pub fn from_key<Q: ?Sized>(self, k: &Q) -> RawEntryMut<'a, K, V, S, A> + where + S: BuildHasher, + Q: Hash + Equivalent<K>, + { + let hash = make_hash::<Q, S>(&self.map.hash_builder, k); + self.from_key_hashed_nocheck(hash, k) + } + + /// Creates a `RawEntryMut` from the given key and its hash. + /// + /// # Examples + /// + /// ``` + /// use core::hash::{BuildHasher, Hash}; + /// use hashbrown::hash_map::{HashMap, RawEntryMut}; + /// + /// fn compute_hash<K: Hash + ?Sized, S: BuildHasher>(hash_builder: &S, key: &K) -> u64 { + /// use core::hash::Hasher; + /// let mut state = hash_builder.build_hasher(); + /// key.hash(&mut state); + /// state.finish() + /// } + /// + /// let mut map: HashMap<&str, u32> = HashMap::new(); + /// let key = "a"; + /// let hash = compute_hash(map.hasher(), &key); + /// let entry: RawEntryMut<&str, u32, _> = map.raw_entry_mut().from_key_hashed_nocheck(hash, &key); + /// entry.insert(key, 100); + /// assert_eq!(map[&"a"], 100); + /// ``` + #[inline] + #[allow(clippy::wrong_self_convention)] + pub fn from_key_hashed_nocheck<Q: ?Sized>(self, hash: u64, k: &Q) -> RawEntryMut<'a, K, V, S, A> + where + Q: Equivalent<K>, + { + self.from_hash(hash, equivalent(k)) + } +} + +impl<'a, K, V, S, A: Allocator> RawEntryBuilderMut<'a, K, V, S, A> { + /// Creates a `RawEntryMut` from the given hash and matching function. + /// + /// # Examples + /// + /// ``` + /// use core::hash::{BuildHasher, Hash}; + /// use hashbrown::hash_map::{HashMap, RawEntryMut}; + /// + /// fn compute_hash<K: Hash + ?Sized, S: BuildHasher>(hash_builder: &S, key: &K) -> u64 { + /// use core::hash::Hasher; + /// let mut state = hash_builder.build_hasher(); + /// key.hash(&mut state); + /// state.finish() + /// } + /// + /// let mut map: HashMap<&str, u32> = HashMap::new(); + /// let key = "a"; + /// let hash = compute_hash(map.hasher(), &key); + /// let entry: RawEntryMut<&str, u32, _> = map.raw_entry_mut().from_hash(hash, |k| k == &key); + /// entry.insert(key, 100); + /// assert_eq!(map[&"a"], 100); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + #[allow(clippy::wrong_self_convention)] + pub fn from_hash<F>(self, hash: u64, is_match: F) -> RawEntryMut<'a, K, V, S, A> + where + for<'b> F: FnMut(&'b K) -> bool, + { + self.search(hash, is_match) + } + + #[cfg_attr(feature = "inline-more", inline)] + fn search<F>(self, hash: u64, mut is_match: F) -> RawEntryMut<'a, K, V, S, A> + where + for<'b> F: FnMut(&'b K) -> bool, + { + match self.map.table.find(hash, |(k, _)| is_match(k)) { + Some(elem) => RawEntryMut::Occupied(RawOccupiedEntryMut { + elem, + table: &mut self.map.table, + hash_builder: &self.map.hash_builder, + }), + None => RawEntryMut::Vacant(RawVacantEntryMut { + table: &mut self.map.table, + hash_builder: &self.map.hash_builder, + }), + } + } +} + +impl<'a, K, V, S, A: Allocator> RawEntryBuilder<'a, K, V, S, A> { + /// Access an immutable entry by key. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let map: HashMap<&str, u32> = [("a", 100), ("b", 200)].into(); + /// let key = "a"; + /// assert_eq!(map.raw_entry().from_key(&key), Some((&"a", &100))); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + #[allow(clippy::wrong_self_convention)] + pub fn from_key<Q: ?Sized>(self, k: &Q) -> Option<(&'a K, &'a V)> + where + S: BuildHasher, + Q: Hash + Equivalent<K>, + { + let hash = make_hash::<Q, S>(&self.map.hash_builder, k); + self.from_key_hashed_nocheck(hash, k) + } + + /// Access an immutable entry by a key and its hash. + /// + /// # Examples + /// + /// ``` + /// use core::hash::{BuildHasher, Hash}; + /// use hashbrown::HashMap; + /// + /// fn compute_hash<K: Hash + ?Sized, S: BuildHasher>(hash_builder: &S, key: &K) -> u64 { + /// use core::hash::Hasher; + /// let mut state = hash_builder.build_hasher(); + /// key.hash(&mut state); + /// state.finish() + /// } + /// + /// let map: HashMap<&str, u32> = [("a", 100), ("b", 200)].into(); + /// let key = "a"; + /// let hash = compute_hash(map.hasher(), &key); + /// assert_eq!(map.raw_entry().from_key_hashed_nocheck(hash, &key), Some((&"a", &100))); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + #[allow(clippy::wrong_self_convention)] + pub fn from_key_hashed_nocheck<Q: ?Sized>(self, hash: u64, k: &Q) -> Option<(&'a K, &'a V)> + where + Q: Equivalent<K>, + { + self.from_hash(hash, equivalent(k)) + } + + #[cfg_attr(feature = "inline-more", inline)] + fn search<F>(self, hash: u64, mut is_match: F) -> Option<(&'a K, &'a V)> + where + F: FnMut(&K) -> bool, + { + match self.map.table.get(hash, |(k, _)| is_match(k)) { + Some((key, value)) => Some((key, value)), + None => None, + } + } + + /// Access an immutable entry by hash and matching function. + /// + /// # Examples + /// + /// ``` + /// use core::hash::{BuildHasher, Hash}; + /// use hashbrown::HashMap; + /// + /// fn compute_hash<K: Hash + ?Sized, S: BuildHasher>(hash_builder: &S, key: &K) -> u64 { + /// use core::hash::Hasher; + /// let mut state = hash_builder.build_hasher(); + /// key.hash(&mut state); + /// state.finish() + /// } + /// + /// let map: HashMap<&str, u32> = [("a", 100), ("b", 200)].into(); + /// let key = "a"; + /// let hash = compute_hash(map.hasher(), &key); + /// assert_eq!(map.raw_entry().from_hash(hash, |k| k == &key), Some((&"a", &100))); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + #[allow(clippy::wrong_self_convention)] + pub fn from_hash<F>(self, hash: u64, is_match: F) -> Option<(&'a K, &'a V)> + where + F: FnMut(&K) -> bool, + { + self.search(hash, is_match) + } +} + +impl<'a, K, V, S, A: Allocator> RawEntryMut<'a, K, V, S, A> { + /// Sets the value of the entry, and returns a RawOccupiedEntryMut. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let mut map: HashMap<&str, u32> = HashMap::new(); + /// let entry = map.raw_entry_mut().from_key("horseyland").insert("horseyland", 37); + /// + /// assert_eq!(entry.remove_entry(), ("horseyland", 37)); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn insert(self, key: K, value: V) -> RawOccupiedEntryMut<'a, K, V, S, A> + where + K: Hash, + S: BuildHasher, + { + match self { + RawEntryMut::Occupied(mut entry) => { + entry.insert(value); + entry + } + RawEntryMut::Vacant(entry) => entry.insert_entry(key, value), + } + } + + /// Ensures a value is in the entry by inserting the default if empty, and returns + /// mutable references to the key and value in the entry. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let mut map: HashMap<&str, u32> = HashMap::new(); + /// + /// map.raw_entry_mut().from_key("poneyland").or_insert("poneyland", 3); + /// assert_eq!(map["poneyland"], 3); + /// + /// *map.raw_entry_mut().from_key("poneyland").or_insert("poneyland", 10).1 *= 2; + /// assert_eq!(map["poneyland"], 6); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn or_insert(self, default_key: K, default_val: V) -> (&'a mut K, &'a mut V) + where + K: Hash, + S: BuildHasher, + { + match self { + RawEntryMut::Occupied(entry) => entry.into_key_value(), + RawEntryMut::Vacant(entry) => entry.insert(default_key, default_val), + } + } + + /// Ensures a value is in the entry by inserting the result of the default function if empty, + /// and returns mutable references to the key and value in the entry. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let mut map: HashMap<&str, String> = HashMap::new(); + /// + /// map.raw_entry_mut().from_key("poneyland").or_insert_with(|| { + /// ("poneyland", "hoho".to_string()) + /// }); + /// + /// assert_eq!(map["poneyland"], "hoho".to_string()); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn or_insert_with<F>(self, default: F) -> (&'a mut K, &'a mut V) + where + F: FnOnce() -> (K, V), + K: Hash, + S: BuildHasher, + { + match self { + RawEntryMut::Occupied(entry) => entry.into_key_value(), + RawEntryMut::Vacant(entry) => { + let (k, v) = default(); + entry.insert(k, v) + } + } + } + + /// Provides in-place mutable access to an occupied entry before any + /// potential inserts into the map. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let mut map: HashMap<&str, u32> = HashMap::new(); + /// + /// map.raw_entry_mut() + /// .from_key("poneyland") + /// .and_modify(|_k, v| { *v += 1 }) + /// .or_insert("poneyland", 42); + /// assert_eq!(map["poneyland"], 42); + /// + /// map.raw_entry_mut() + /// .from_key("poneyland") + /// .and_modify(|_k, v| { *v += 1 }) + /// .or_insert("poneyland", 0); + /// assert_eq!(map["poneyland"], 43); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn and_modify<F>(self, f: F) -> Self + where + F: FnOnce(&mut K, &mut V), + { + match self { + RawEntryMut::Occupied(mut entry) => { + { + let (k, v) = entry.get_key_value_mut(); + f(k, v); + } + RawEntryMut::Occupied(entry) + } + RawEntryMut::Vacant(entry) => RawEntryMut::Vacant(entry), + } + } + + /// Provides shared access to the key and owned access to the value of + /// an occupied entry and allows to replace or remove it based on the + /// value of the returned option. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// use hashbrown::hash_map::RawEntryMut; + /// + /// let mut map: HashMap<&str, u32> = HashMap::new(); + /// + /// let entry = map + /// .raw_entry_mut() + /// .from_key("poneyland") + /// .and_replace_entry_with(|_k, _v| panic!()); + /// + /// match entry { + /// RawEntryMut::Vacant(_) => {}, + /// RawEntryMut::Occupied(_) => panic!(), + /// } + /// + /// map.insert("poneyland", 42); + /// + /// let entry = map + /// .raw_entry_mut() + /// .from_key("poneyland") + /// .and_replace_entry_with(|k, v| { + /// assert_eq!(k, &"poneyland"); + /// assert_eq!(v, 42); + /// Some(v + 1) + /// }); + /// + /// match entry { + /// RawEntryMut::Occupied(e) => { + /// assert_eq!(e.key(), &"poneyland"); + /// assert_eq!(e.get(), &43); + /// }, + /// RawEntryMut::Vacant(_) => panic!(), + /// } + /// + /// assert_eq!(map["poneyland"], 43); + /// + /// let entry = map + /// .raw_entry_mut() + /// .from_key("poneyland") + /// .and_replace_entry_with(|_k, _v| None); + /// + /// match entry { + /// RawEntryMut::Vacant(_) => {}, + /// RawEntryMut::Occupied(_) => panic!(), + /// } + /// + /// assert!(!map.contains_key("poneyland")); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn and_replace_entry_with<F>(self, f: F) -> Self + where + F: FnOnce(&K, V) -> Option<V>, + { + match self { + RawEntryMut::Occupied(entry) => entry.replace_entry_with(f), + RawEntryMut::Vacant(_) => self, + } + } +} + +impl<'a, K, V, S, A: Allocator> RawOccupiedEntryMut<'a, K, V, S, A> { + /// Gets a reference to the key in the entry. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::hash_map::{HashMap, RawEntryMut}; + /// + /// let mut map: HashMap<&str, u32> = [("a", 100), ("b", 200)].into(); + /// + /// match map.raw_entry_mut().from_key(&"a") { + /// RawEntryMut::Vacant(_) => panic!(), + /// RawEntryMut::Occupied(o) => assert_eq!(o.key(), &"a") + /// } + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn key(&self) -> &K { + unsafe { &self.elem.as_ref().0 } + } + + /// Gets a mutable reference to the key in the entry. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::hash_map::{HashMap, RawEntryMut}; + /// use std::rc::Rc; + /// + /// let key_one = Rc::new("a"); + /// let key_two = Rc::new("a"); + /// + /// let mut map: HashMap<Rc<&str>, u32> = HashMap::new(); + /// map.insert(key_one.clone(), 10); + /// + /// assert_eq!(map[&key_one], 10); + /// assert!(Rc::strong_count(&key_one) == 2 && Rc::strong_count(&key_two) == 1); + /// + /// match map.raw_entry_mut().from_key(&key_one) { + /// RawEntryMut::Vacant(_) => panic!(), + /// RawEntryMut::Occupied(mut o) => { + /// *o.key_mut() = key_two.clone(); + /// } + /// } + /// assert_eq!(map[&key_two], 10); + /// assert!(Rc::strong_count(&key_one) == 1 && Rc::strong_count(&key_two) == 2); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn key_mut(&mut self) -> &mut K { + unsafe { &mut self.elem.as_mut().0 } + } + + /// Converts the entry into a mutable reference to the key in the entry + /// with a lifetime bound to the map itself. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::hash_map::{HashMap, RawEntryMut}; + /// use std::rc::Rc; + /// + /// let key_one = Rc::new("a"); + /// let key_two = Rc::new("a"); + /// + /// let mut map: HashMap<Rc<&str>, u32> = HashMap::new(); + /// map.insert(key_one.clone(), 10); + /// + /// assert_eq!(map[&key_one], 10); + /// assert!(Rc::strong_count(&key_one) == 2 && Rc::strong_count(&key_two) == 1); + /// + /// let inside_key: &mut Rc<&str>; + /// + /// match map.raw_entry_mut().from_key(&key_one) { + /// RawEntryMut::Vacant(_) => panic!(), + /// RawEntryMut::Occupied(o) => inside_key = o.into_key(), + /// } + /// *inside_key = key_two.clone(); + /// + /// assert_eq!(map[&key_two], 10); + /// assert!(Rc::strong_count(&key_one) == 1 && Rc::strong_count(&key_two) == 2); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn into_key(self) -> &'a mut K { + unsafe { &mut self.elem.as_mut().0 } + } + + /// Gets a reference to the value in the entry. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::hash_map::{HashMap, RawEntryMut}; + /// + /// let mut map: HashMap<&str, u32> = [("a", 100), ("b", 200)].into(); + /// + /// match map.raw_entry_mut().from_key(&"a") { + /// RawEntryMut::Vacant(_) => panic!(), + /// RawEntryMut::Occupied(o) => assert_eq!(o.get(), &100), + /// } + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn get(&self) -> &V { + unsafe { &self.elem.as_ref().1 } + } + + /// Converts the OccupiedEntry into a mutable reference to the value in the entry + /// with a lifetime bound to the map itself. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::hash_map::{HashMap, RawEntryMut}; + /// + /// let mut map: HashMap<&str, u32> = [("a", 100), ("b", 200)].into(); + /// + /// let value: &mut u32; + /// + /// match map.raw_entry_mut().from_key(&"a") { + /// RawEntryMut::Vacant(_) => panic!(), + /// RawEntryMut::Occupied(o) => value = o.into_mut(), + /// } + /// *value += 900; + /// + /// assert_eq!(map[&"a"], 1000); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn into_mut(self) -> &'a mut V { + unsafe { &mut self.elem.as_mut().1 } + } + + /// Gets a mutable reference to the value in the entry. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::hash_map::{HashMap, RawEntryMut}; + /// + /// let mut map: HashMap<&str, u32> = [("a", 100), ("b", 200)].into(); + /// + /// match map.raw_entry_mut().from_key(&"a") { + /// RawEntryMut::Vacant(_) => panic!(), + /// RawEntryMut::Occupied(mut o) => *o.get_mut() += 900, + /// } + /// + /// assert_eq!(map[&"a"], 1000); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn get_mut(&mut self) -> &mut V { + unsafe { &mut self.elem.as_mut().1 } + } + + /// Gets a reference to the key and value in the entry. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::hash_map::{HashMap, RawEntryMut}; + /// + /// let mut map: HashMap<&str, u32> = [("a", 100), ("b", 200)].into(); + /// + /// match map.raw_entry_mut().from_key(&"a") { + /// RawEntryMut::Vacant(_) => panic!(), + /// RawEntryMut::Occupied(o) => assert_eq!(o.get_key_value(), (&"a", &100)), + /// } + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn get_key_value(&self) -> (&K, &V) { + unsafe { + let (key, value) = self.elem.as_ref(); + (key, value) + } + } + + /// Gets a mutable reference to the key and value in the entry. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::hash_map::{HashMap, RawEntryMut}; + /// use std::rc::Rc; + /// + /// let key_one = Rc::new("a"); + /// let key_two = Rc::new("a"); + /// + /// let mut map: HashMap<Rc<&str>, u32> = HashMap::new(); + /// map.insert(key_one.clone(), 10); + /// + /// assert_eq!(map[&key_one], 10); + /// assert!(Rc::strong_count(&key_one) == 2 && Rc::strong_count(&key_two) == 1); + /// + /// match map.raw_entry_mut().from_key(&key_one) { + /// RawEntryMut::Vacant(_) => panic!(), + /// RawEntryMut::Occupied(mut o) => { + /// let (inside_key, inside_value) = o.get_key_value_mut(); + /// *inside_key = key_two.clone(); + /// *inside_value = 100; + /// } + /// } + /// assert_eq!(map[&key_two], 100); + /// assert!(Rc::strong_count(&key_one) == 1 && Rc::strong_count(&key_two) == 2); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn get_key_value_mut(&mut self) -> (&mut K, &mut V) { + unsafe { + let &mut (ref mut key, ref mut value) = self.elem.as_mut(); + (key, value) + } + } + + /// Converts the OccupiedEntry into a mutable reference to the key and value in the entry + /// with a lifetime bound to the map itself. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::hash_map::{HashMap, RawEntryMut}; + /// use std::rc::Rc; + /// + /// let key_one = Rc::new("a"); + /// let key_two = Rc::new("a"); + /// + /// let mut map: HashMap<Rc<&str>, u32> = HashMap::new(); + /// map.insert(key_one.clone(), 10); + /// + /// assert_eq!(map[&key_one], 10); + /// assert!(Rc::strong_count(&key_one) == 2 && Rc::strong_count(&key_two) == 1); + /// + /// let inside_key: &mut Rc<&str>; + /// let inside_value: &mut u32; + /// match map.raw_entry_mut().from_key(&key_one) { + /// RawEntryMut::Vacant(_) => panic!(), + /// RawEntryMut::Occupied(o) => { + /// let tuple = o.into_key_value(); + /// inside_key = tuple.0; + /// inside_value = tuple.1; + /// } + /// } + /// *inside_key = key_two.clone(); + /// *inside_value = 100; + /// assert_eq!(map[&key_two], 100); + /// assert!(Rc::strong_count(&key_one) == 1 && Rc::strong_count(&key_two) == 2); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn into_key_value(self) -> (&'a mut K, &'a mut V) { + unsafe { + let &mut (ref mut key, ref mut value) = self.elem.as_mut(); + (key, value) + } + } + + /// Sets the value of the entry, and returns the entry's old value. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::hash_map::{HashMap, RawEntryMut}; + /// + /// let mut map: HashMap<&str, u32> = [("a", 100), ("b", 200)].into(); + /// + /// match map.raw_entry_mut().from_key(&"a") { + /// RawEntryMut::Vacant(_) => panic!(), + /// RawEntryMut::Occupied(mut o) => assert_eq!(o.insert(1000), 100), + /// } + /// + /// assert_eq!(map[&"a"], 1000); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn insert(&mut self, value: V) -> V { + mem::replace(self.get_mut(), value) + } + + /// Sets the value of the entry, and returns the entry's old value. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::hash_map::{HashMap, RawEntryMut}; + /// use std::rc::Rc; + /// + /// let key_one = Rc::new("a"); + /// let key_two = Rc::new("a"); + /// + /// let mut map: HashMap<Rc<&str>, u32> = HashMap::new(); + /// map.insert(key_one.clone(), 10); + /// + /// assert_eq!(map[&key_one], 10); + /// assert!(Rc::strong_count(&key_one) == 2 && Rc::strong_count(&key_two) == 1); + /// + /// match map.raw_entry_mut().from_key(&key_one) { + /// RawEntryMut::Vacant(_) => panic!(), + /// RawEntryMut::Occupied(mut o) => { + /// let old_key = o.insert_key(key_two.clone()); + /// assert!(Rc::ptr_eq(&old_key, &key_one)); + /// } + /// } + /// assert_eq!(map[&key_two], 10); + /// assert!(Rc::strong_count(&key_one) == 1 && Rc::strong_count(&key_two) == 2); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn insert_key(&mut self, key: K) -> K { + mem::replace(self.key_mut(), key) + } + + /// Takes the value out of the entry, and returns it. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::hash_map::{HashMap, RawEntryMut}; + /// + /// let mut map: HashMap<&str, u32> = [("a", 100), ("b", 200)].into(); + /// + /// match map.raw_entry_mut().from_key(&"a") { + /// RawEntryMut::Vacant(_) => panic!(), + /// RawEntryMut::Occupied(o) => assert_eq!(o.remove(), 100), + /// } + /// assert_eq!(map.get(&"a"), None); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn remove(self) -> V { + self.remove_entry().1 + } + + /// Take the ownership of the key and value from the map. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::hash_map::{HashMap, RawEntryMut}; + /// + /// let mut map: HashMap<&str, u32> = [("a", 100), ("b", 200)].into(); + /// + /// match map.raw_entry_mut().from_key(&"a") { + /// RawEntryMut::Vacant(_) => panic!(), + /// RawEntryMut::Occupied(o) => assert_eq!(o.remove_entry(), ("a", 100)), + /// } + /// assert_eq!(map.get(&"a"), None); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn remove_entry(self) -> (K, V) { + unsafe { self.table.remove(self.elem).0 } + } + + /// Provides shared access to the key and owned access to the value of + /// the entry and allows to replace or remove it based on the + /// value of the returned option. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::hash_map::{HashMap, RawEntryMut}; + /// + /// let mut map: HashMap<&str, u32> = [("a", 100), ("b", 200)].into(); + /// + /// let raw_entry = match map.raw_entry_mut().from_key(&"a") { + /// RawEntryMut::Vacant(_) => panic!(), + /// RawEntryMut::Occupied(o) => o.replace_entry_with(|k, v| { + /// assert_eq!(k, &"a"); + /// assert_eq!(v, 100); + /// Some(v + 900) + /// }), + /// }; + /// let raw_entry = match raw_entry { + /// RawEntryMut::Vacant(_) => panic!(), + /// RawEntryMut::Occupied(o) => o.replace_entry_with(|k, v| { + /// assert_eq!(k, &"a"); + /// assert_eq!(v, 1000); + /// None + /// }), + /// }; + /// match raw_entry { + /// RawEntryMut::Vacant(_) => { }, + /// RawEntryMut::Occupied(_) => panic!(), + /// }; + /// assert_eq!(map.get(&"a"), None); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn replace_entry_with<F>(self, f: F) -> RawEntryMut<'a, K, V, S, A> + where + F: FnOnce(&K, V) -> Option<V>, + { + unsafe { + let still_occupied = self + .table + .replace_bucket_with(self.elem.clone(), |(key, value)| { + f(&key, value).map(|new_value| (key, new_value)) + }); + + if still_occupied { + RawEntryMut::Occupied(self) + } else { + RawEntryMut::Vacant(RawVacantEntryMut { + table: self.table, + hash_builder: self.hash_builder, + }) + } + } + } +} + +impl<'a, K, V, S, A: Allocator> RawVacantEntryMut<'a, K, V, S, A> { + /// Sets the value of the entry with the VacantEntry's key, + /// and returns a mutable reference to it. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::hash_map::{HashMap, RawEntryMut}; + /// + /// let mut map: HashMap<&str, u32> = [("a", 100), ("b", 200)].into(); + /// + /// match map.raw_entry_mut().from_key(&"c") { + /// RawEntryMut::Occupied(_) => panic!(), + /// RawEntryMut::Vacant(v) => assert_eq!(v.insert("c", 300), (&mut "c", &mut 300)), + /// } + /// + /// assert_eq!(map[&"c"], 300); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn insert(self, key: K, value: V) -> (&'a mut K, &'a mut V) + where + K: Hash, + S: BuildHasher, + { + let hash = make_hash::<K, S>(self.hash_builder, &key); + self.insert_hashed_nocheck(hash, key, value) + } + + /// Sets the value of the entry with the VacantEntry's key, + /// and returns a mutable reference to it. + /// + /// # Examples + /// + /// ``` + /// use core::hash::{BuildHasher, Hash}; + /// use hashbrown::hash_map::{HashMap, RawEntryMut}; + /// + /// fn compute_hash<K: Hash + ?Sized, S: BuildHasher>(hash_builder: &S, key: &K) -> u64 { + /// use core::hash::Hasher; + /// let mut state = hash_builder.build_hasher(); + /// key.hash(&mut state); + /// state.finish() + /// } + /// + /// let mut map: HashMap<&str, u32> = [("a", 100), ("b", 200)].into(); + /// let key = "c"; + /// let hash = compute_hash(map.hasher(), &key); + /// + /// match map.raw_entry_mut().from_key_hashed_nocheck(hash, &key) { + /// RawEntryMut::Occupied(_) => panic!(), + /// RawEntryMut::Vacant(v) => assert_eq!( + /// v.insert_hashed_nocheck(hash, key, 300), + /// (&mut "c", &mut 300) + /// ), + /// } + /// + /// assert_eq!(map[&"c"], 300); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + #[allow(clippy::shadow_unrelated)] + pub fn insert_hashed_nocheck(self, hash: u64, key: K, value: V) -> (&'a mut K, &'a mut V) + where + K: Hash, + S: BuildHasher, + { + let &mut (ref mut k, ref mut v) = self.table.insert_entry( + hash, + (key, value), + make_hasher::<_, V, S>(self.hash_builder), + ); + (k, v) + } + + /// Set the value of an entry with a custom hasher function. + /// + /// # Examples + /// + /// ``` + /// use core::hash::{BuildHasher, Hash}; + /// use hashbrown::hash_map::{HashMap, RawEntryMut}; + /// + /// fn make_hasher<K, S>(hash_builder: &S) -> impl Fn(&K) -> u64 + '_ + /// where + /// K: Hash + ?Sized, + /// S: BuildHasher, + /// { + /// move |key: &K| { + /// use core::hash::Hasher; + /// let mut state = hash_builder.build_hasher(); + /// key.hash(&mut state); + /// state.finish() + /// } + /// } + /// + /// let mut map: HashMap<&str, u32> = HashMap::new(); + /// let key = "a"; + /// let hash_builder = map.hasher().clone(); + /// let hash = make_hasher(&hash_builder)(&key); + /// + /// match map.raw_entry_mut().from_hash(hash, |q| q == &key) { + /// RawEntryMut::Occupied(_) => panic!(), + /// RawEntryMut::Vacant(v) => assert_eq!( + /// v.insert_with_hasher(hash, key, 100, make_hasher(&hash_builder)), + /// (&mut "a", &mut 100) + /// ), + /// } + /// map.extend([("b", 200), ("c", 300), ("d", 400), ("e", 500), ("f", 600)]); + /// assert_eq!(map[&"a"], 100); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn insert_with_hasher<H>( + self, + hash: u64, + key: K, + value: V, + hasher: H, + ) -> (&'a mut K, &'a mut V) + where + H: Fn(&K) -> u64, + { + let &mut (ref mut k, ref mut v) = self + .table + .insert_entry(hash, (key, value), |x| hasher(&x.0)); + (k, v) + } + + #[cfg_attr(feature = "inline-more", inline)] + fn insert_entry(self, key: K, value: V) -> RawOccupiedEntryMut<'a, K, V, S, A> + where + K: Hash, + S: BuildHasher, + { + let hash = make_hash::<K, S>(self.hash_builder, &key); + let elem = self.table.insert( + hash, + (key, value), + make_hasher::<_, V, S>(self.hash_builder), + ); + RawOccupiedEntryMut { + elem, + table: self.table, + hash_builder: self.hash_builder, + } + } +} + +impl<K, V, S, A: Allocator> Debug for RawEntryBuilderMut<'_, K, V, S, A> { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + f.debug_struct("RawEntryBuilder").finish() + } +} + +impl<K: Debug, V: Debug, S, A: Allocator> Debug for RawEntryMut<'_, K, V, S, A> { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + match *self { + RawEntryMut::Vacant(ref v) => f.debug_tuple("RawEntry").field(v).finish(), + RawEntryMut::Occupied(ref o) => f.debug_tuple("RawEntry").field(o).finish(), + } + } +} + +impl<K: Debug, V: Debug, S, A: Allocator> Debug for RawOccupiedEntryMut<'_, K, V, S, A> { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + f.debug_struct("RawOccupiedEntryMut") + .field("key", self.key()) + .field("value", self.get()) + .finish() + } +} + +impl<K, V, S, A: Allocator> Debug for RawVacantEntryMut<'_, K, V, S, A> { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + f.debug_struct("RawVacantEntryMut").finish() + } +} + +impl<K, V, S, A: Allocator> Debug for RawEntryBuilder<'_, K, V, S, A> { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + f.debug_struct("RawEntryBuilder").finish() + } +} + +/// A view into a single entry in a map, which may either be vacant or occupied. +/// +/// This `enum` is constructed from the [`entry`] method on [`HashMap`]. +/// +/// [`HashMap`]: struct.HashMap.html +/// [`entry`]: struct.HashMap.html#method.entry +/// +/// # Examples +/// +/// ``` +/// use hashbrown::hash_map::{Entry, HashMap, OccupiedEntry}; +/// +/// let mut map = HashMap::new(); +/// map.extend([("a", 10), ("b", 20), ("c", 30)]); +/// assert_eq!(map.len(), 3); +/// +/// // Existing key (insert) +/// let entry: Entry<_, _, _> = map.entry("a"); +/// let _raw_o: OccupiedEntry<_, _, _> = entry.insert(1); +/// assert_eq!(map.len(), 3); +/// // Nonexistent key (insert) +/// map.entry("d").insert(4); +/// +/// // Existing key (or_insert) +/// let v = map.entry("b").or_insert(2); +/// assert_eq!(std::mem::replace(v, 2), 20); +/// // Nonexistent key (or_insert) +/// map.entry("e").or_insert(5); +/// +/// // Existing key (or_insert_with) +/// let v = map.entry("c").or_insert_with(|| 3); +/// assert_eq!(std::mem::replace(v, 3), 30); +/// // Nonexistent key (or_insert_with) +/// map.entry("f").or_insert_with(|| 6); +/// +/// println!("Our HashMap: {:?}", map); +/// +/// let mut vec: Vec<_> = map.iter().map(|(&k, &v)| (k, v)).collect(); +/// // The `Iter` iterator produces items in arbitrary order, so the +/// // items must be sorted to test them against a sorted array. +/// vec.sort_unstable(); +/// assert_eq!(vec, [("a", 1), ("b", 2), ("c", 3), ("d", 4), ("e", 5), ("f", 6)]); +/// ``` +pub enum Entry<'a, K, V, S, A = Global> +where + A: Allocator, +{ + /// An occupied entry. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::hash_map::{Entry, HashMap}; + /// let mut map: HashMap<_, _> = [("a", 100), ("b", 200)].into(); + /// + /// match map.entry("a") { + /// Entry::Vacant(_) => unreachable!(), + /// Entry::Occupied(_) => { } + /// } + /// ``` + Occupied(OccupiedEntry<'a, K, V, S, A>), + + /// A vacant entry. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::hash_map::{Entry, HashMap}; + /// let mut map: HashMap<&str, i32> = HashMap::new(); + /// + /// match map.entry("a") { + /// Entry::Occupied(_) => unreachable!(), + /// Entry::Vacant(_) => { } + /// } + /// ``` + Vacant(VacantEntry<'a, K, V, S, A>), +} + +impl<K: Debug, V: Debug, S, A: Allocator> Debug for Entry<'_, K, V, S, A> { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + match *self { + Entry::Vacant(ref v) => f.debug_tuple("Entry").field(v).finish(), + Entry::Occupied(ref o) => f.debug_tuple("Entry").field(o).finish(), + } + } +} + +/// A view into an occupied entry in a `HashMap`. +/// It is part of the [`Entry`] enum. +/// +/// [`Entry`]: enum.Entry.html +/// +/// # Examples +/// +/// ``` +/// use hashbrown::hash_map::{Entry, HashMap, OccupiedEntry}; +/// +/// let mut map = HashMap::new(); +/// map.extend([("a", 10), ("b", 20), ("c", 30)]); +/// +/// let _entry_o: OccupiedEntry<_, _, _> = map.entry("a").insert(100); +/// assert_eq!(map.len(), 3); +/// +/// // Existing key (insert and update) +/// match map.entry("a") { +/// Entry::Vacant(_) => unreachable!(), +/// Entry::Occupied(mut view) => { +/// assert_eq!(view.get(), &100); +/// let v = view.get_mut(); +/// *v *= 10; +/// assert_eq!(view.insert(1111), 1000); +/// } +/// } +/// +/// assert_eq!(map[&"a"], 1111); +/// assert_eq!(map.len(), 3); +/// +/// // Existing key (take) +/// match map.entry("c") { +/// Entry::Vacant(_) => unreachable!(), +/// Entry::Occupied(view) => { +/// assert_eq!(view.remove_entry(), ("c", 30)); +/// } +/// } +/// assert_eq!(map.get(&"c"), None); +/// assert_eq!(map.len(), 2); +/// ``` +pub struct OccupiedEntry<'a, K, V, S = DefaultHashBuilder, A: Allocator = Global> { + hash: u64, + key: Option<K>, + elem: Bucket<(K, V)>, + table: &'a mut HashMap<K, V, S, A>, +} + +unsafe impl<K, V, S, A> Send for OccupiedEntry<'_, K, V, S, A> +where + K: Send, + V: Send, + S: Send, + A: Send + Allocator, +{ +} +unsafe impl<K, V, S, A> Sync for OccupiedEntry<'_, K, V, S, A> +where + K: Sync, + V: Sync, + S: Sync, + A: Sync + Allocator, +{ +} + +impl<K: Debug, V: Debug, S, A: Allocator> Debug for OccupiedEntry<'_, K, V, S, A> { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + f.debug_struct("OccupiedEntry") + .field("key", self.key()) + .field("value", self.get()) + .finish() + } +} + +/// A view into a vacant entry in a `HashMap`. +/// It is part of the [`Entry`] enum. +/// +/// [`Entry`]: enum.Entry.html +/// +/// # Examples +/// +/// ``` +/// use hashbrown::hash_map::{Entry, HashMap, VacantEntry}; +/// +/// let mut map = HashMap::<&str, i32>::new(); +/// +/// let entry_v: VacantEntry<_, _, _> = match map.entry("a") { +/// Entry::Vacant(view) => view, +/// Entry::Occupied(_) => unreachable!(), +/// }; +/// entry_v.insert(10); +/// assert!(map[&"a"] == 10 && map.len() == 1); +/// +/// // Nonexistent key (insert and update) +/// match map.entry("b") { +/// Entry::Occupied(_) => unreachable!(), +/// Entry::Vacant(view) => { +/// let value = view.insert(2); +/// assert_eq!(*value, 2); +/// *value = 20; +/// } +/// } +/// assert!(map[&"b"] == 20 && map.len() == 2); +/// ``` +pub struct VacantEntry<'a, K, V, S = DefaultHashBuilder, A: Allocator = Global> { + hash: u64, + key: K, + table: &'a mut HashMap<K, V, S, A>, +} + +impl<K: Debug, V, S, A: Allocator> Debug for VacantEntry<'_, K, V, S, A> { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + f.debug_tuple("VacantEntry").field(self.key()).finish() + } +} + +/// A view into a single entry in a map, which may either be vacant or occupied, +/// with any borrowed form of the map's key type. +/// +/// +/// This `enum` is constructed from the [`entry_ref`] method on [`HashMap`]. +/// +/// [`Hash`] and [`Eq`] on the borrowed form of the map's key type *must* match those +/// for the key type. It also require that key may be constructed from the borrowed +/// form through the [`From`] trait. +/// +/// [`HashMap`]: struct.HashMap.html +/// [`entry_ref`]: struct.HashMap.html#method.entry_ref +/// [`Eq`]: https://doc.rust-lang.org/std/cmp/trait.Eq.html +/// [`Hash`]: https://doc.rust-lang.org/std/hash/trait.Hash.html +/// [`From`]: https://doc.rust-lang.org/std/convert/trait.From.html +/// +/// # Examples +/// +/// ``` +/// use hashbrown::hash_map::{EntryRef, HashMap, OccupiedEntryRef}; +/// +/// let mut map = HashMap::new(); +/// map.extend([("a".to_owned(), 10), ("b".into(), 20), ("c".into(), 30)]); +/// assert_eq!(map.len(), 3); +/// +/// // Existing key (insert) +/// let key = String::from("a"); +/// let entry: EntryRef<_, _, _, _> = map.entry_ref(&key); +/// let _raw_o: OccupiedEntryRef<_, _, _, _> = entry.insert(1); +/// assert_eq!(map.len(), 3); +/// // Nonexistent key (insert) +/// map.entry_ref("d").insert(4); +/// +/// // Existing key (or_insert) +/// let v = map.entry_ref("b").or_insert(2); +/// assert_eq!(std::mem::replace(v, 2), 20); +/// // Nonexistent key (or_insert) +/// map.entry_ref("e").or_insert(5); +/// +/// // Existing key (or_insert_with) +/// let v = map.entry_ref("c").or_insert_with(|| 3); +/// assert_eq!(std::mem::replace(v, 3), 30); +/// // Nonexistent key (or_insert_with) +/// map.entry_ref("f").or_insert_with(|| 6); +/// +/// println!("Our HashMap: {:?}", map); +/// +/// for (key, value) in ["a", "b", "c", "d", "e", "f"].into_iter().zip(1..=6) { +/// assert_eq!(map[key], value) +/// } +/// assert_eq!(map.len(), 6); +/// ``` +pub enum EntryRef<'a, 'b, K, Q: ?Sized, V, S, A = Global> +where + A: Allocator, +{ + /// An occupied entry. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::hash_map::{EntryRef, HashMap}; + /// let mut map: HashMap<_, _> = [("a".to_owned(), 100), ("b".into(), 200)].into(); + /// + /// match map.entry_ref("a") { + /// EntryRef::Vacant(_) => unreachable!(), + /// EntryRef::Occupied(_) => { } + /// } + /// ``` + Occupied(OccupiedEntryRef<'a, 'b, K, Q, V, S, A>), + + /// A vacant entry. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::hash_map::{EntryRef, HashMap}; + /// let mut map: HashMap<String, i32> = HashMap::new(); + /// + /// match map.entry_ref("a") { + /// EntryRef::Occupied(_) => unreachable!(), + /// EntryRef::Vacant(_) => { } + /// } + /// ``` + Vacant(VacantEntryRef<'a, 'b, K, Q, V, S, A>), +} + +impl<K: Borrow<Q>, Q: ?Sized + Debug, V: Debug, S, A: Allocator> Debug + for EntryRef<'_, '_, K, Q, V, S, A> +{ + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + match *self { + EntryRef::Vacant(ref v) => f.debug_tuple("EntryRef").field(v).finish(), + EntryRef::Occupied(ref o) => f.debug_tuple("EntryRef").field(o).finish(), + } + } +} + +enum KeyOrRef<'a, K, Q: ?Sized> { + Borrowed(&'a Q), + Owned(K), +} + +impl<'a, K, Q: ?Sized> KeyOrRef<'a, K, Q> { + fn into_owned(self) -> K + where + K: From<&'a Q>, + { + match self { + Self::Borrowed(borrowed) => borrowed.into(), + Self::Owned(owned) => owned, + } + } +} + +impl<'a, K: Borrow<Q>, Q: ?Sized> AsRef<Q> for KeyOrRef<'a, K, Q> { + fn as_ref(&self) -> &Q { + match self { + Self::Borrowed(borrowed) => borrowed, + Self::Owned(owned) => owned.borrow(), + } + } +} + +/// A view into an occupied entry in a `HashMap`. +/// It is part of the [`EntryRef`] enum. +/// +/// [`EntryRef`]: enum.EntryRef.html +/// +/// # Examples +/// +/// ``` +/// use hashbrown::hash_map::{EntryRef, HashMap, OccupiedEntryRef}; +/// +/// let mut map = HashMap::new(); +/// map.extend([("a".to_owned(), 10), ("b".into(), 20), ("c".into(), 30)]); +/// +/// let key = String::from("a"); +/// let _entry_o: OccupiedEntryRef<_, _, _, _> = map.entry_ref(&key).insert(100); +/// assert_eq!(map.len(), 3); +/// +/// // Existing key (insert and update) +/// match map.entry_ref("a") { +/// EntryRef::Vacant(_) => unreachable!(), +/// EntryRef::Occupied(mut view) => { +/// assert_eq!(view.get(), &100); +/// let v = view.get_mut(); +/// *v *= 10; +/// assert_eq!(view.insert(1111), 1000); +/// } +/// } +/// +/// assert_eq!(map["a"], 1111); +/// assert_eq!(map.len(), 3); +/// +/// // Existing key (take) +/// match map.entry_ref("c") { +/// EntryRef::Vacant(_) => unreachable!(), +/// EntryRef::Occupied(view) => { +/// assert_eq!(view.remove_entry(), ("c".to_owned(), 30)); +/// } +/// } +/// assert_eq!(map.get("c"), None); +/// assert_eq!(map.len(), 2); +/// ``` +pub struct OccupiedEntryRef<'a, 'b, K, Q: ?Sized, V, S, A: Allocator = Global> { + hash: u64, + key: Option<KeyOrRef<'b, K, Q>>, + elem: Bucket<(K, V)>, + table: &'a mut HashMap<K, V, S, A>, +} + +unsafe impl<'a, 'b, K, Q, V, S, A> Send for OccupiedEntryRef<'a, 'b, K, Q, V, S, A> +where + K: Send, + Q: Sync + ?Sized, + V: Send, + S: Send, + A: Send + Allocator, +{ +} +unsafe impl<'a, 'b, K, Q, V, S, A> Sync for OccupiedEntryRef<'a, 'b, K, Q, V, S, A> +where + K: Sync, + Q: Sync + ?Sized, + V: Sync, + S: Sync, + A: Sync + Allocator, +{ +} + +impl<K: Borrow<Q>, Q: ?Sized + Debug, V: Debug, S, A: Allocator> Debug + for OccupiedEntryRef<'_, '_, K, Q, V, S, A> +{ + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + f.debug_struct("OccupiedEntryRef") + .field("key", &self.key().borrow()) + .field("value", &self.get()) + .finish() + } +} + +/// A view into a vacant entry in a `HashMap`. +/// It is part of the [`EntryRef`] enum. +/// +/// [`EntryRef`]: enum.EntryRef.html +/// +/// # Examples +/// +/// ``` +/// use hashbrown::hash_map::{EntryRef, HashMap, VacantEntryRef}; +/// +/// let mut map = HashMap::<String, i32>::new(); +/// +/// let entry_v: VacantEntryRef<_, _, _, _> = match map.entry_ref("a") { +/// EntryRef::Vacant(view) => view, +/// EntryRef::Occupied(_) => unreachable!(), +/// }; +/// entry_v.insert(10); +/// assert!(map["a"] == 10 && map.len() == 1); +/// +/// // Nonexistent key (insert and update) +/// match map.entry_ref("b") { +/// EntryRef::Occupied(_) => unreachable!(), +/// EntryRef::Vacant(view) => { +/// let value = view.insert(2); +/// assert_eq!(*value, 2); +/// *value = 20; +/// } +/// } +/// assert!(map["b"] == 20 && map.len() == 2); +/// ``` +pub struct VacantEntryRef<'a, 'b, K, Q: ?Sized, V, S, A: Allocator = Global> { + hash: u64, + key: KeyOrRef<'b, K, Q>, + table: &'a mut HashMap<K, V, S, A>, +} + +impl<K: Borrow<Q>, Q: ?Sized + Debug, V, S, A: Allocator> Debug + for VacantEntryRef<'_, '_, K, Q, V, S, A> +{ + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + f.debug_tuple("VacantEntryRef").field(&self.key()).finish() + } +} + +/// The error returned by [`try_insert`](HashMap::try_insert) when the key already exists. +/// +/// Contains the occupied entry, and the value that was not inserted. +/// +/// # Examples +/// +/// ``` +/// use hashbrown::hash_map::{HashMap, OccupiedError}; +/// +/// let mut map: HashMap<_, _> = [("a", 10), ("b", 20)].into(); +/// +/// // try_insert method returns mutable reference to the value if keys are vacant, +/// // but if the map did have key present, nothing is updated, and the provided +/// // value is returned inside `Err(_)` variant +/// match map.try_insert("a", 100) { +/// Err(OccupiedError { mut entry, value }) => { +/// assert_eq!(entry.key(), &"a"); +/// assert_eq!(value, 100); +/// assert_eq!(entry.insert(100), 10) +/// } +/// _ => unreachable!(), +/// } +/// assert_eq!(map[&"a"], 100); +/// ``` +pub struct OccupiedError<'a, K, V, S, A: Allocator = Global> { + /// The entry in the map that was already occupied. + pub entry: OccupiedEntry<'a, K, V, S, A>, + /// The value which was not inserted, because the entry was already occupied. + pub value: V, +} + +impl<K: Debug, V: Debug, S, A: Allocator> Debug for OccupiedError<'_, K, V, S, A> { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + f.debug_struct("OccupiedError") + .field("key", self.entry.key()) + .field("old_value", self.entry.get()) + .field("new_value", &self.value) + .finish() + } +} + +impl<'a, K: Debug, V: Debug, S, A: Allocator> fmt::Display for OccupiedError<'a, K, V, S, A> { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + write!( + f, + "failed to insert {:?}, key {:?} already exists with value {:?}", + self.value, + self.entry.key(), + self.entry.get(), + ) + } +} + +impl<'a, K, V, S, A: Allocator> IntoIterator for &'a HashMap<K, V, S, A> { + type Item = (&'a K, &'a V); + type IntoIter = Iter<'a, K, V>; + + /// Creates an iterator over the entries of a `HashMap` in arbitrary order. + /// The iterator element type is `(&'a K, &'a V)`. + /// + /// Return the same `Iter` struct as by the [`iter`] method on [`HashMap`]. + /// + /// [`iter`]: struct.HashMap.html#method.iter + /// [`HashMap`]: struct.HashMap.html + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// let map_one: HashMap<_, _> = [(1, "a"), (2, "b"), (3, "c")].into(); + /// let mut map_two = HashMap::new(); + /// + /// for (key, value) in &map_one { + /// println!("Key: {}, Value: {}", key, value); + /// map_two.insert_unique_unchecked(*key, *value); + /// } + /// + /// assert_eq!(map_one, map_two); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + fn into_iter(self) -> Iter<'a, K, V> { + self.iter() + } +} + +impl<'a, K, V, S, A: Allocator> IntoIterator for &'a mut HashMap<K, V, S, A> { + type Item = (&'a K, &'a mut V); + type IntoIter = IterMut<'a, K, V>; + + /// Creates an iterator over the entries of a `HashMap` in arbitrary order + /// with mutable references to the values. The iterator element type is + /// `(&'a K, &'a mut V)`. + /// + /// Return the same `IterMut` struct as by the [`iter_mut`] method on + /// [`HashMap`]. + /// + /// [`iter_mut`]: struct.HashMap.html#method.iter_mut + /// [`HashMap`]: struct.HashMap.html + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// let mut map: HashMap<_, _> = [("a", 1), ("b", 2), ("c", 3)].into(); + /// + /// for (key, value) in &mut map { + /// println!("Key: {}, Value: {}", key, value); + /// *value *= 2; + /// } + /// + /// let mut vec = map.iter().collect::<Vec<_>>(); + /// // The `Iter` iterator produces items in arbitrary order, so the + /// // items must be sorted to test them against a sorted array. + /// vec.sort_unstable(); + /// assert_eq!(vec, [(&"a", &2), (&"b", &4), (&"c", &6)]); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + fn into_iter(self) -> IterMut<'a, K, V> { + self.iter_mut() + } +} + +impl<K, V, S, A: Allocator> IntoIterator for HashMap<K, V, S, A> { + type Item = (K, V); + type IntoIter = IntoIter<K, V, A>; + + /// Creates a consuming iterator, that is, one that moves each key-value + /// pair out of the map in arbitrary order. The map cannot be used after + /// calling this. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let map: HashMap<_, _> = [("a", 1), ("b", 2), ("c", 3)].into(); + /// + /// // Not possible with .iter() + /// let mut vec: Vec<(&str, i32)> = map.into_iter().collect(); + /// // The `IntoIter` iterator produces items in arbitrary order, so + /// // the items must be sorted to test them against a sorted array. + /// vec.sort_unstable(); + /// assert_eq!(vec, [("a", 1), ("b", 2), ("c", 3)]); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + fn into_iter(self) -> IntoIter<K, V, A> { + IntoIter { + inner: self.table.into_iter(), + } + } +} + +impl<'a, K, V> Iterator for Iter<'a, K, V> { + type Item = (&'a K, &'a V); + + #[cfg_attr(feature = "inline-more", inline)] + fn next(&mut self) -> Option<(&'a K, &'a V)> { + // Avoid `Option::map` because it bloats LLVM IR. + match self.inner.next() { + Some(x) => unsafe { + let r = x.as_ref(); + Some((&r.0, &r.1)) + }, + None => None, + } + } + #[cfg_attr(feature = "inline-more", inline)] + fn size_hint(&self) -> (usize, Option<usize>) { + self.inner.size_hint() + } +} +impl<K, V> ExactSizeIterator for Iter<'_, K, V> { + #[cfg_attr(feature = "inline-more", inline)] + fn len(&self) -> usize { + self.inner.len() + } +} + +impl<K, V> FusedIterator for Iter<'_, K, V> {} + +impl<'a, K, V> Iterator for IterMut<'a, K, V> { + type Item = (&'a K, &'a mut V); + + #[cfg_attr(feature = "inline-more", inline)] + fn next(&mut self) -> Option<(&'a K, &'a mut V)> { + // Avoid `Option::map` because it bloats LLVM IR. + match self.inner.next() { + Some(x) => unsafe { + let r = x.as_mut(); + Some((&r.0, &mut r.1)) + }, + None => None, + } + } + #[cfg_attr(feature = "inline-more", inline)] + fn size_hint(&self) -> (usize, Option<usize>) { + self.inner.size_hint() + } +} +impl<K, V> ExactSizeIterator for IterMut<'_, K, V> { + #[cfg_attr(feature = "inline-more", inline)] + fn len(&self) -> usize { + self.inner.len() + } +} +impl<K, V> FusedIterator for IterMut<'_, K, V> {} + +impl<K, V> fmt::Debug for IterMut<'_, K, V> +where + K: fmt::Debug, + V: fmt::Debug, +{ + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + f.debug_list().entries(self.iter()).finish() + } +} + +impl<K, V, A: Allocator> Iterator for IntoIter<K, V, A> { + type Item = (K, V); + + #[cfg_attr(feature = "inline-more", inline)] + fn next(&mut self) -> Option<(K, V)> { + self.inner.next() + } + #[cfg_attr(feature = "inline-more", inline)] + fn size_hint(&self) -> (usize, Option<usize>) { + self.inner.size_hint() + } +} +impl<K, V, A: Allocator> ExactSizeIterator for IntoIter<K, V, A> { + #[cfg_attr(feature = "inline-more", inline)] + fn len(&self) -> usize { + self.inner.len() + } +} +impl<K, V, A: Allocator> FusedIterator for IntoIter<K, V, A> {} + +impl<K: Debug, V: Debug, A: Allocator> fmt::Debug for IntoIter<K, V, A> { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + f.debug_list().entries(self.iter()).finish() + } +} + +impl<'a, K, V> Iterator for Keys<'a, K, V> { + type Item = &'a K; + + #[cfg_attr(feature = "inline-more", inline)] + fn next(&mut self) -> Option<&'a K> { + // Avoid `Option::map` because it bloats LLVM IR. + match self.inner.next() { + Some((k, _)) => Some(k), + None => None, + } + } + #[cfg_attr(feature = "inline-more", inline)] + fn size_hint(&self) -> (usize, Option<usize>) { + self.inner.size_hint() + } +} +impl<K, V> ExactSizeIterator for Keys<'_, K, V> { + #[cfg_attr(feature = "inline-more", inline)] + fn len(&self) -> usize { + self.inner.len() + } +} +impl<K, V> FusedIterator for Keys<'_, K, V> {} + +impl<'a, K, V> Iterator for Values<'a, K, V> { + type Item = &'a V; + + #[cfg_attr(feature = "inline-more", inline)] + fn next(&mut self) -> Option<&'a V> { + // Avoid `Option::map` because it bloats LLVM IR. + match self.inner.next() { + Some((_, v)) => Some(v), + None => None, + } + } + #[cfg_attr(feature = "inline-more", inline)] + fn size_hint(&self) -> (usize, Option<usize>) { + self.inner.size_hint() + } +} +impl<K, V> ExactSizeIterator for Values<'_, K, V> { + #[cfg_attr(feature = "inline-more", inline)] + fn len(&self) -> usize { + self.inner.len() + } +} +impl<K, V> FusedIterator for Values<'_, K, V> {} + +impl<'a, K, V> Iterator for ValuesMut<'a, K, V> { + type Item = &'a mut V; + + #[cfg_attr(feature = "inline-more", inline)] + fn next(&mut self) -> Option<&'a mut V> { + // Avoid `Option::map` because it bloats LLVM IR. + match self.inner.next() { + Some((_, v)) => Some(v), + None => None, + } + } + #[cfg_attr(feature = "inline-more", inline)] + fn size_hint(&self) -> (usize, Option<usize>) { + self.inner.size_hint() + } +} +impl<K, V> ExactSizeIterator for ValuesMut<'_, K, V> { + #[cfg_attr(feature = "inline-more", inline)] + fn len(&self) -> usize { + self.inner.len() + } +} +impl<K, V> FusedIterator for ValuesMut<'_, K, V> {} + +impl<K, V: Debug> fmt::Debug for ValuesMut<'_, K, V> { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + f.debug_list() + .entries(self.inner.iter().map(|(_, val)| val)) + .finish() + } +} + +impl<'a, K, V, A: Allocator> Iterator for Drain<'a, K, V, A> { + type Item = (K, V); + + #[cfg_attr(feature = "inline-more", inline)] + fn next(&mut self) -> Option<(K, V)> { + self.inner.next() + } + #[cfg_attr(feature = "inline-more", inline)] + fn size_hint(&self) -> (usize, Option<usize>) { + self.inner.size_hint() + } +} +impl<K, V, A: Allocator> ExactSizeIterator for Drain<'_, K, V, A> { + #[cfg_attr(feature = "inline-more", inline)] + fn len(&self) -> usize { + self.inner.len() + } +} +impl<K, V, A: Allocator> FusedIterator for Drain<'_, K, V, A> {} + +impl<K, V, A> fmt::Debug for Drain<'_, K, V, A> +where + K: fmt::Debug, + V: fmt::Debug, + A: Allocator, +{ + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + f.debug_list().entries(self.iter()).finish() + } +} + +impl<'a, K, V, S, A: Allocator> Entry<'a, K, V, S, A> { + /// Sets the value of the entry, and returns an OccupiedEntry. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let mut map: HashMap<&str, u32> = HashMap::new(); + /// let entry = map.entry("horseyland").insert(37); + /// + /// assert_eq!(entry.key(), &"horseyland"); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn insert(self, value: V) -> OccupiedEntry<'a, K, V, S, A> + where + K: Hash, + S: BuildHasher, + { + match self { + Entry::Occupied(mut entry) => { + entry.insert(value); + entry + } + Entry::Vacant(entry) => entry.insert_entry(value), + } + } + + /// Ensures a value is in the entry by inserting the default if empty, and returns + /// a mutable reference to the value in the entry. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let mut map: HashMap<&str, u32> = HashMap::new(); + /// + /// // nonexistent key + /// map.entry("poneyland").or_insert(3); + /// assert_eq!(map["poneyland"], 3); + /// + /// // existing key + /// *map.entry("poneyland").or_insert(10) *= 2; + /// assert_eq!(map["poneyland"], 6); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn or_insert(self, default: V) -> &'a mut V + where + K: Hash, + S: BuildHasher, + { + match self { + Entry::Occupied(entry) => entry.into_mut(), + Entry::Vacant(entry) => entry.insert(default), + } + } + + /// Ensures a value is in the entry by inserting the result of the default function if empty, + /// and returns a mutable reference to the value in the entry. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let mut map: HashMap<&str, u32> = HashMap::new(); + /// + /// // nonexistent key + /// map.entry("poneyland").or_insert_with(|| 3); + /// assert_eq!(map["poneyland"], 3); + /// + /// // existing key + /// *map.entry("poneyland").or_insert_with(|| 10) *= 2; + /// assert_eq!(map["poneyland"], 6); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn or_insert_with<F: FnOnce() -> V>(self, default: F) -> &'a mut V + where + K: Hash, + S: BuildHasher, + { + match self { + Entry::Occupied(entry) => entry.into_mut(), + Entry::Vacant(entry) => entry.insert(default()), + } + } + + /// Ensures a value is in the entry by inserting, if empty, the result of the default function. + /// This method allows for generating key-derived values for insertion by providing the default + /// function a reference to the key that was moved during the `.entry(key)` method call. + /// + /// The reference to the moved key is provided so that cloning or copying the key is + /// unnecessary, unlike with `.or_insert_with(|| ... )`. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let mut map: HashMap<&str, usize> = HashMap::new(); + /// + /// // nonexistent key + /// map.entry("poneyland").or_insert_with_key(|key| key.chars().count()); + /// assert_eq!(map["poneyland"], 9); + /// + /// // existing key + /// *map.entry("poneyland").or_insert_with_key(|key| key.chars().count() * 10) *= 2; + /// assert_eq!(map["poneyland"], 18); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn or_insert_with_key<F: FnOnce(&K) -> V>(self, default: F) -> &'a mut V + where + K: Hash, + S: BuildHasher, + { + match self { + Entry::Occupied(entry) => entry.into_mut(), + Entry::Vacant(entry) => { + let value = default(entry.key()); + entry.insert(value) + } + } + } + + /// Returns a reference to this entry's key. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let mut map: HashMap<&str, u32> = HashMap::new(); + /// map.entry("poneyland").or_insert(3); + /// // existing key + /// assert_eq!(map.entry("poneyland").key(), &"poneyland"); + /// // nonexistent key + /// assert_eq!(map.entry("horseland").key(), &"horseland"); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn key(&self) -> &K { + match *self { + Entry::Occupied(ref entry) => entry.key(), + Entry::Vacant(ref entry) => entry.key(), + } + } + + /// Provides in-place mutable access to an occupied entry before any + /// potential inserts into the map. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let mut map: HashMap<&str, u32> = HashMap::new(); + /// + /// map.entry("poneyland") + /// .and_modify(|e| { *e += 1 }) + /// .or_insert(42); + /// assert_eq!(map["poneyland"], 42); + /// + /// map.entry("poneyland") + /// .and_modify(|e| { *e += 1 }) + /// .or_insert(42); + /// assert_eq!(map["poneyland"], 43); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn and_modify<F>(self, f: F) -> Self + where + F: FnOnce(&mut V), + { + match self { + Entry::Occupied(mut entry) => { + f(entry.get_mut()); + Entry::Occupied(entry) + } + Entry::Vacant(entry) => Entry::Vacant(entry), + } + } + + /// Provides shared access to the key and owned access to the value of + /// an occupied entry and allows to replace or remove it based on the + /// value of the returned option. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// use hashbrown::hash_map::Entry; + /// + /// let mut map: HashMap<&str, u32> = HashMap::new(); + /// + /// let entry = map + /// .entry("poneyland") + /// .and_replace_entry_with(|_k, _v| panic!()); + /// + /// match entry { + /// Entry::Vacant(e) => { + /// assert_eq!(e.key(), &"poneyland"); + /// } + /// Entry::Occupied(_) => panic!(), + /// } + /// + /// map.insert("poneyland", 42); + /// + /// let entry = map + /// .entry("poneyland") + /// .and_replace_entry_with(|k, v| { + /// assert_eq!(k, &"poneyland"); + /// assert_eq!(v, 42); + /// Some(v + 1) + /// }); + /// + /// match entry { + /// Entry::Occupied(e) => { + /// assert_eq!(e.key(), &"poneyland"); + /// assert_eq!(e.get(), &43); + /// } + /// Entry::Vacant(_) => panic!(), + /// } + /// + /// assert_eq!(map["poneyland"], 43); + /// + /// let entry = map + /// .entry("poneyland") + /// .and_replace_entry_with(|_k, _v| None); + /// + /// match entry { + /// Entry::Vacant(e) => assert_eq!(e.key(), &"poneyland"), + /// Entry::Occupied(_) => panic!(), + /// } + /// + /// assert!(!map.contains_key("poneyland")); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn and_replace_entry_with<F>(self, f: F) -> Self + where + F: FnOnce(&K, V) -> Option<V>, + { + match self { + Entry::Occupied(entry) => entry.replace_entry_with(f), + Entry::Vacant(_) => self, + } + } +} + +impl<'a, K, V: Default, S, A: Allocator> Entry<'a, K, V, S, A> { + /// Ensures a value is in the entry by inserting the default value if empty, + /// and returns a mutable reference to the value in the entry. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let mut map: HashMap<&str, Option<u32>> = HashMap::new(); + /// + /// // nonexistent key + /// map.entry("poneyland").or_default(); + /// assert_eq!(map["poneyland"], None); + /// + /// map.insert("horseland", Some(3)); + /// + /// // existing key + /// assert_eq!(map.entry("horseland").or_default(), &mut Some(3)); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn or_default(self) -> &'a mut V + where + K: Hash, + S: BuildHasher, + { + match self { + Entry::Occupied(entry) => entry.into_mut(), + Entry::Vacant(entry) => entry.insert(Default::default()), + } + } +} + +impl<'a, K, V, S, A: Allocator> OccupiedEntry<'a, K, V, S, A> { + /// Gets a reference to the key in the entry. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::hash_map::{Entry, HashMap}; + /// + /// let mut map: HashMap<&str, u32> = HashMap::new(); + /// map.entry("poneyland").or_insert(12); + /// + /// match map.entry("poneyland") { + /// Entry::Vacant(_) => panic!(), + /// Entry::Occupied(entry) => assert_eq!(entry.key(), &"poneyland"), + /// } + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn key(&self) -> &K { + unsafe { &self.elem.as_ref().0 } + } + + /// Take the ownership of the key and value from the map. + /// Keeps the allocated memory for reuse. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// use hashbrown::hash_map::Entry; + /// + /// let mut map: HashMap<&str, u32> = HashMap::new(); + /// // The map is empty + /// assert!(map.is_empty() && map.capacity() == 0); + /// + /// map.entry("poneyland").or_insert(12); + /// + /// if let Entry::Occupied(o) = map.entry("poneyland") { + /// // We delete the entry from the map. + /// assert_eq!(o.remove_entry(), ("poneyland", 12)); + /// } + /// + /// assert_eq!(map.contains_key("poneyland"), false); + /// // Now map hold none elements + /// assert!(map.is_empty()); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn remove_entry(self) -> (K, V) { + unsafe { self.table.table.remove(self.elem).0 } + } + + /// Gets a reference to the value in the entry. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// use hashbrown::hash_map::Entry; + /// + /// let mut map: HashMap<&str, u32> = HashMap::new(); + /// map.entry("poneyland").or_insert(12); + /// + /// match map.entry("poneyland") { + /// Entry::Vacant(_) => panic!(), + /// Entry::Occupied(entry) => assert_eq!(entry.get(), &12), + /// } + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn get(&self) -> &V { + unsafe { &self.elem.as_ref().1 } + } + + /// Gets a mutable reference to the value in the entry. + /// + /// If you need a reference to the `OccupiedEntry` which may outlive the + /// destruction of the `Entry` value, see [`into_mut`]. + /// + /// [`into_mut`]: #method.into_mut + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// use hashbrown::hash_map::Entry; + /// + /// let mut map: HashMap<&str, u32> = HashMap::new(); + /// map.entry("poneyland").or_insert(12); + /// + /// assert_eq!(map["poneyland"], 12); + /// if let Entry::Occupied(mut o) = map.entry("poneyland") { + /// *o.get_mut() += 10; + /// assert_eq!(*o.get(), 22); + /// + /// // We can use the same Entry multiple times. + /// *o.get_mut() += 2; + /// } + /// + /// assert_eq!(map["poneyland"], 24); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn get_mut(&mut self) -> &mut V { + unsafe { &mut self.elem.as_mut().1 } + } + + /// Converts the OccupiedEntry into a mutable reference to the value in the entry + /// with a lifetime bound to the map itself. + /// + /// If you need multiple references to the `OccupiedEntry`, see [`get_mut`]. + /// + /// [`get_mut`]: #method.get_mut + /// + /// # Examples + /// + /// ``` + /// use hashbrown::hash_map::{Entry, HashMap}; + /// + /// let mut map: HashMap<&str, u32> = HashMap::new(); + /// map.entry("poneyland").or_insert(12); + /// + /// assert_eq!(map["poneyland"], 12); + /// + /// let value: &mut u32; + /// match map.entry("poneyland") { + /// Entry::Occupied(entry) => value = entry.into_mut(), + /// Entry::Vacant(_) => panic!(), + /// } + /// *value += 10; + /// + /// assert_eq!(map["poneyland"], 22); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn into_mut(self) -> &'a mut V { + unsafe { &mut self.elem.as_mut().1 } + } + + /// Sets the value of the entry, and returns the entry's old value. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// use hashbrown::hash_map::Entry; + /// + /// let mut map: HashMap<&str, u32> = HashMap::new(); + /// map.entry("poneyland").or_insert(12); + /// + /// if let Entry::Occupied(mut o) = map.entry("poneyland") { + /// assert_eq!(o.insert(15), 12); + /// } + /// + /// assert_eq!(map["poneyland"], 15); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn insert(&mut self, value: V) -> V { + mem::replace(self.get_mut(), value) + } + + /// Takes the value out of the entry, and returns it. + /// Keeps the allocated memory for reuse. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// use hashbrown::hash_map::Entry; + /// + /// let mut map: HashMap<&str, u32> = HashMap::new(); + /// // The map is empty + /// assert!(map.is_empty() && map.capacity() == 0); + /// + /// map.entry("poneyland").or_insert(12); + /// + /// if let Entry::Occupied(o) = map.entry("poneyland") { + /// assert_eq!(o.remove(), 12); + /// } + /// + /// assert_eq!(map.contains_key("poneyland"), false); + /// // Now map hold none elements + /// assert!(map.is_empty()); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn remove(self) -> V { + self.remove_entry().1 + } + + /// Replaces the entry, returning the old key and value. The new key in the hash map will be + /// the key used to create this entry. + /// + /// # Panics + /// + /// Will panic if this OccupiedEntry was created through [`Entry::insert`]. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::hash_map::{Entry, HashMap}; + /// use std::rc::Rc; + /// + /// let mut map: HashMap<Rc<String>, u32> = HashMap::new(); + /// let key_one = Rc::new("Stringthing".to_string()); + /// let key_two = Rc::new("Stringthing".to_string()); + /// + /// map.insert(key_one.clone(), 15); + /// assert!(Rc::strong_count(&key_one) == 2 && Rc::strong_count(&key_two) == 1); + /// + /// match map.entry(key_two.clone()) { + /// Entry::Occupied(entry) => { + /// let (old_key, old_value): (Rc<String>, u32) = entry.replace_entry(16); + /// assert!(Rc::ptr_eq(&key_one, &old_key) && old_value == 15); + /// } + /// Entry::Vacant(_) => panic!(), + /// } + /// + /// assert!(Rc::strong_count(&key_one) == 1 && Rc::strong_count(&key_two) == 2); + /// assert_eq!(map[&"Stringthing".to_owned()], 16); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn replace_entry(self, value: V) -> (K, V) { + let entry = unsafe { self.elem.as_mut() }; + + let old_key = mem::replace(&mut entry.0, self.key.unwrap()); + let old_value = mem::replace(&mut entry.1, value); + + (old_key, old_value) + } + + /// Replaces the key in the hash map with the key used to create this entry. + /// + /// # Panics + /// + /// Will panic if this OccupiedEntry was created through [`Entry::insert`]. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::hash_map::{Entry, HashMap}; + /// use std::rc::Rc; + /// + /// let mut map: HashMap<Rc<String>, usize> = HashMap::with_capacity(6); + /// let mut keys_one: Vec<Rc<String>> = Vec::with_capacity(6); + /// let mut keys_two: Vec<Rc<String>> = Vec::with_capacity(6); + /// + /// for (value, key) in ["a", "b", "c", "d", "e", "f"].into_iter().enumerate() { + /// let rc_key = Rc::new(key.to_owned()); + /// keys_one.push(rc_key.clone()); + /// map.insert(rc_key.clone(), value); + /// keys_two.push(Rc::new(key.to_owned())); + /// } + /// + /// assert!( + /// keys_one.iter().all(|key| Rc::strong_count(key) == 2) + /// && keys_two.iter().all(|key| Rc::strong_count(key) == 1) + /// ); + /// + /// reclaim_memory(&mut map, &keys_two); + /// + /// assert!( + /// keys_one.iter().all(|key| Rc::strong_count(key) == 1) + /// && keys_two.iter().all(|key| Rc::strong_count(key) == 2) + /// ); + /// + /// fn reclaim_memory(map: &mut HashMap<Rc<String>, usize>, keys: &[Rc<String>]) { + /// for key in keys { + /// if let Entry::Occupied(entry) = map.entry(key.clone()) { + /// // Replaces the entry's key with our version of it in `keys`. + /// entry.replace_key(); + /// } + /// } + /// } + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn replace_key(self) -> K { + let entry = unsafe { self.elem.as_mut() }; + mem::replace(&mut entry.0, self.key.unwrap()) + } + + /// Provides shared access to the key and owned access to the value of + /// the entry and allows to replace or remove it based on the + /// value of the returned option. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// use hashbrown::hash_map::Entry; + /// + /// let mut map: HashMap<&str, u32> = HashMap::new(); + /// map.insert("poneyland", 42); + /// + /// let entry = match map.entry("poneyland") { + /// Entry::Occupied(e) => { + /// e.replace_entry_with(|k, v| { + /// assert_eq!(k, &"poneyland"); + /// assert_eq!(v, 42); + /// Some(v + 1) + /// }) + /// } + /// Entry::Vacant(_) => panic!(), + /// }; + /// + /// match entry { + /// Entry::Occupied(e) => { + /// assert_eq!(e.key(), &"poneyland"); + /// assert_eq!(e.get(), &43); + /// } + /// Entry::Vacant(_) => panic!(), + /// } + /// + /// assert_eq!(map["poneyland"], 43); + /// + /// let entry = match map.entry("poneyland") { + /// Entry::Occupied(e) => e.replace_entry_with(|_k, _v| None), + /// Entry::Vacant(_) => panic!(), + /// }; + /// + /// match entry { + /// Entry::Vacant(e) => { + /// assert_eq!(e.key(), &"poneyland"); + /// } + /// Entry::Occupied(_) => panic!(), + /// } + /// + /// assert!(!map.contains_key("poneyland")); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn replace_entry_with<F>(self, f: F) -> Entry<'a, K, V, S, A> + where + F: FnOnce(&K, V) -> Option<V>, + { + unsafe { + let mut spare_key = None; + + self.table + .table + .replace_bucket_with(self.elem.clone(), |(key, value)| { + if let Some(new_value) = f(&key, value) { + Some((key, new_value)) + } else { + spare_key = Some(key); + None + } + }); + + if let Some(key) = spare_key { + Entry::Vacant(VacantEntry { + hash: self.hash, + key, + table: self.table, + }) + } else { + Entry::Occupied(self) + } + } + } +} + +impl<'a, K, V, S, A: Allocator> VacantEntry<'a, K, V, S, A> { + /// Gets a reference to the key that would be used when inserting a value + /// through the `VacantEntry`. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let mut map: HashMap<&str, u32> = HashMap::new(); + /// assert_eq!(map.entry("poneyland").key(), &"poneyland"); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn key(&self) -> &K { + &self.key + } + + /// Take ownership of the key. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::hash_map::{Entry, HashMap}; + /// + /// let mut map: HashMap<&str, u32> = HashMap::new(); + /// + /// match map.entry("poneyland") { + /// Entry::Occupied(_) => panic!(), + /// Entry::Vacant(v) => assert_eq!(v.into_key(), "poneyland"), + /// } + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn into_key(self) -> K { + self.key + } + + /// Sets the value of the entry with the VacantEntry's key, + /// and returns a mutable reference to it. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// use hashbrown::hash_map::Entry; + /// + /// let mut map: HashMap<&str, u32> = HashMap::new(); + /// + /// if let Entry::Vacant(o) = map.entry("poneyland") { + /// o.insert(37); + /// } + /// assert_eq!(map["poneyland"], 37); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn insert(self, value: V) -> &'a mut V + where + K: Hash, + S: BuildHasher, + { + let table = &mut self.table.table; + let entry = table.insert_entry( + self.hash, + (self.key, value), + make_hasher::<_, V, S>(&self.table.hash_builder), + ); + &mut entry.1 + } + + #[cfg_attr(feature = "inline-more", inline)] + pub(crate) fn insert_entry(self, value: V) -> OccupiedEntry<'a, K, V, S, A> + where + K: Hash, + S: BuildHasher, + { + let elem = self.table.table.insert( + self.hash, + (self.key, value), + make_hasher::<_, V, S>(&self.table.hash_builder), + ); + OccupiedEntry { + hash: self.hash, + key: None, + elem, + table: self.table, + } + } +} + +impl<'a, 'b, K, Q: ?Sized, V, S, A: Allocator> EntryRef<'a, 'b, K, Q, V, S, A> { + /// Sets the value of the entry, and returns an OccupiedEntryRef. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let mut map: HashMap<String, u32> = HashMap::new(); + /// let entry = map.entry_ref("horseyland").insert(37); + /// + /// assert_eq!(entry.key(), "horseyland"); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn insert(self, value: V) -> OccupiedEntryRef<'a, 'b, K, Q, V, S, A> + where + K: Hash + From<&'b Q>, + S: BuildHasher, + { + match self { + EntryRef::Occupied(mut entry) => { + entry.insert(value); + entry + } + EntryRef::Vacant(entry) => entry.insert_entry(value), + } + } + + /// Ensures a value is in the entry by inserting the default if empty, and returns + /// a mutable reference to the value in the entry. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let mut map: HashMap<String, u32> = HashMap::new(); + /// + /// // nonexistent key + /// map.entry_ref("poneyland").or_insert(3); + /// assert_eq!(map["poneyland"], 3); + /// + /// // existing key + /// *map.entry_ref("poneyland").or_insert(10) *= 2; + /// assert_eq!(map["poneyland"], 6); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn or_insert(self, default: V) -> &'a mut V + where + K: Hash + From<&'b Q>, + S: BuildHasher, + { + match self { + EntryRef::Occupied(entry) => entry.into_mut(), + EntryRef::Vacant(entry) => entry.insert(default), + } + } + + /// Ensures a value is in the entry by inserting the result of the default function if empty, + /// and returns a mutable reference to the value in the entry. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let mut map: HashMap<String, u32> = HashMap::new(); + /// + /// // nonexistent key + /// map.entry_ref("poneyland").or_insert_with(|| 3); + /// assert_eq!(map["poneyland"], 3); + /// + /// // existing key + /// *map.entry_ref("poneyland").or_insert_with(|| 10) *= 2; + /// assert_eq!(map["poneyland"], 6); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn or_insert_with<F: FnOnce() -> V>(self, default: F) -> &'a mut V + where + K: Hash + From<&'b Q>, + S: BuildHasher, + { + match self { + EntryRef::Occupied(entry) => entry.into_mut(), + EntryRef::Vacant(entry) => entry.insert(default()), + } + } + + /// Ensures a value is in the entry by inserting, if empty, the result of the default function. + /// This method allows for generating key-derived values for insertion by providing the default + /// function an access to the borrower form of the key. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let mut map: HashMap<String, usize> = HashMap::new(); + /// + /// // nonexistent key + /// map.entry_ref("poneyland").or_insert_with_key(|key| key.chars().count()); + /// assert_eq!(map["poneyland"], 9); + /// + /// // existing key + /// *map.entry_ref("poneyland").or_insert_with_key(|key| key.chars().count() * 10) *= 2; + /// assert_eq!(map["poneyland"], 18); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn or_insert_with_key<F: FnOnce(&Q) -> V>(self, default: F) -> &'a mut V + where + K: Hash + Borrow<Q> + From<&'b Q>, + S: BuildHasher, + { + match self { + EntryRef::Occupied(entry) => entry.into_mut(), + EntryRef::Vacant(entry) => { + let value = default(entry.key.as_ref()); + entry.insert(value) + } + } + } + + /// Returns a reference to this entry's key. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let mut map: HashMap<String, u32> = HashMap::new(); + /// map.entry_ref("poneyland").or_insert(3); + /// // existing key + /// assert_eq!(map.entry_ref("poneyland").key(), "poneyland"); + /// // nonexistent key + /// assert_eq!(map.entry_ref("horseland").key(), "horseland"); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn key(&self) -> &Q + where + K: Borrow<Q>, + { + match *self { + EntryRef::Occupied(ref entry) => entry.key().borrow(), + EntryRef::Vacant(ref entry) => entry.key(), + } + } + + /// Provides in-place mutable access to an occupied entry before any + /// potential inserts into the map. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let mut map: HashMap<String, u32> = HashMap::new(); + /// + /// map.entry_ref("poneyland") + /// .and_modify(|e| { *e += 1 }) + /// .or_insert(42); + /// assert_eq!(map["poneyland"], 42); + /// + /// map.entry_ref("poneyland") + /// .and_modify(|e| { *e += 1 }) + /// .or_insert(42); + /// assert_eq!(map["poneyland"], 43); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn and_modify<F>(self, f: F) -> Self + where + F: FnOnce(&mut V), + { + match self { + EntryRef::Occupied(mut entry) => { + f(entry.get_mut()); + EntryRef::Occupied(entry) + } + EntryRef::Vacant(entry) => EntryRef::Vacant(entry), + } + } + + /// Provides shared access to the key and owned access to the value of + /// an occupied entry and allows to replace or remove it based on the + /// value of the returned option. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// use hashbrown::hash_map::EntryRef; + /// + /// let mut map: HashMap<String, u32> = HashMap::new(); + /// + /// let entry = map + /// .entry_ref("poneyland") + /// .and_replace_entry_with(|_k, _v| panic!()); + /// + /// match entry { + /// EntryRef::Vacant(e) => { + /// assert_eq!(e.key(), "poneyland"); + /// } + /// EntryRef::Occupied(_) => panic!(), + /// } + /// + /// map.insert("poneyland".to_string(), 42); + /// + /// let entry = map + /// .entry_ref("poneyland") + /// .and_replace_entry_with(|k, v| { + /// assert_eq!(k, "poneyland"); + /// assert_eq!(v, 42); + /// Some(v + 1) + /// }); + /// + /// match entry { + /// EntryRef::Occupied(e) => { + /// assert_eq!(e.key(), "poneyland"); + /// assert_eq!(e.get(), &43); + /// } + /// EntryRef::Vacant(_) => panic!(), + /// } + /// + /// assert_eq!(map["poneyland"], 43); + /// + /// let entry = map + /// .entry_ref("poneyland") + /// .and_replace_entry_with(|_k, _v| None); + /// + /// match entry { + /// EntryRef::Vacant(e) => assert_eq!(e.key(), "poneyland"), + /// EntryRef::Occupied(_) => panic!(), + /// } + /// + /// assert!(!map.contains_key("poneyland")); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn and_replace_entry_with<F>(self, f: F) -> Self + where + F: FnOnce(&K, V) -> Option<V>, + { + match self { + EntryRef::Occupied(entry) => entry.replace_entry_with(f), + EntryRef::Vacant(_) => self, + } + } +} + +impl<'a, 'b, K, Q: ?Sized, V: Default, S, A: Allocator> EntryRef<'a, 'b, K, Q, V, S, A> { + /// Ensures a value is in the entry by inserting the default value if empty, + /// and returns a mutable reference to the value in the entry. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let mut map: HashMap<String, Option<u32>> = HashMap::new(); + /// + /// // nonexistent key + /// map.entry_ref("poneyland").or_default(); + /// assert_eq!(map["poneyland"], None); + /// + /// map.insert("horseland".to_string(), Some(3)); + /// + /// // existing key + /// assert_eq!(map.entry_ref("horseland").or_default(), &mut Some(3)); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn or_default(self) -> &'a mut V + where + K: Hash + From<&'b Q>, + S: BuildHasher, + { + match self { + EntryRef::Occupied(entry) => entry.into_mut(), + EntryRef::Vacant(entry) => entry.insert(Default::default()), + } + } +} + +impl<'a, 'b, K, Q: ?Sized, V, S, A: Allocator> OccupiedEntryRef<'a, 'b, K, Q, V, S, A> { + /// Gets a reference to the key in the entry. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::hash_map::{EntryRef, HashMap}; + /// + /// let mut map: HashMap<String, u32> = HashMap::new(); + /// map.entry_ref("poneyland").or_insert(12); + /// + /// match map.entry_ref("poneyland") { + /// EntryRef::Vacant(_) => panic!(), + /// EntryRef::Occupied(entry) => assert_eq!(entry.key(), "poneyland"), + /// } + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn key(&self) -> &K { + unsafe { &self.elem.as_ref().0 } + } + + /// Take the ownership of the key and value from the map. + /// Keeps the allocated memory for reuse. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// use hashbrown::hash_map::EntryRef; + /// + /// let mut map: HashMap<String, u32> = HashMap::new(); + /// // The map is empty + /// assert!(map.is_empty() && map.capacity() == 0); + /// + /// map.entry_ref("poneyland").or_insert(12); + /// + /// if let EntryRef::Occupied(o) = map.entry_ref("poneyland") { + /// // We delete the entry from the map. + /// assert_eq!(o.remove_entry(), ("poneyland".to_owned(), 12)); + /// } + /// + /// assert_eq!(map.contains_key("poneyland"), false); + /// // Now map hold none elements but capacity is equal to the old one + /// assert!(map.is_empty()); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn remove_entry(self) -> (K, V) { + unsafe { self.table.table.remove(self.elem).0 } + } + + /// Gets a reference to the value in the entry. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// use hashbrown::hash_map::EntryRef; + /// + /// let mut map: HashMap<String, u32> = HashMap::new(); + /// map.entry_ref("poneyland").or_insert(12); + /// + /// match map.entry_ref("poneyland") { + /// EntryRef::Vacant(_) => panic!(), + /// EntryRef::Occupied(entry) => assert_eq!(entry.get(), &12), + /// } + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn get(&self) -> &V { + unsafe { &self.elem.as_ref().1 } + } + + /// Gets a mutable reference to the value in the entry. + /// + /// If you need a reference to the `OccupiedEntryRef` which may outlive the + /// destruction of the `EntryRef` value, see [`into_mut`]. + /// + /// [`into_mut`]: #method.into_mut + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// use hashbrown::hash_map::EntryRef; + /// + /// let mut map: HashMap<String, u32> = HashMap::new(); + /// map.entry_ref("poneyland").or_insert(12); + /// + /// assert_eq!(map["poneyland"], 12); + /// if let EntryRef::Occupied(mut o) = map.entry_ref("poneyland") { + /// *o.get_mut() += 10; + /// assert_eq!(*o.get(), 22); + /// + /// // We can use the same Entry multiple times. + /// *o.get_mut() += 2; + /// } + /// + /// assert_eq!(map["poneyland"], 24); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn get_mut(&mut self) -> &mut V { + unsafe { &mut self.elem.as_mut().1 } + } + + /// Converts the OccupiedEntryRef into a mutable reference to the value in the entry + /// with a lifetime bound to the map itself. + /// + /// If you need multiple references to the `OccupiedEntryRef`, see [`get_mut`]. + /// + /// [`get_mut`]: #method.get_mut + /// + /// # Examples + /// + /// ``` + /// use hashbrown::hash_map::{EntryRef, HashMap}; + /// + /// let mut map: HashMap<String, u32> = HashMap::new(); + /// map.entry_ref("poneyland").or_insert(12); + /// + /// let value: &mut u32; + /// match map.entry_ref("poneyland") { + /// EntryRef::Occupied(entry) => value = entry.into_mut(), + /// EntryRef::Vacant(_) => panic!(), + /// } + /// *value += 10; + /// + /// assert_eq!(map["poneyland"], 22); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn into_mut(self) -> &'a mut V { + unsafe { &mut self.elem.as_mut().1 } + } + + /// Sets the value of the entry, and returns the entry's old value. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// use hashbrown::hash_map::EntryRef; + /// + /// let mut map: HashMap<String, u32> = HashMap::new(); + /// map.entry_ref("poneyland").or_insert(12); + /// + /// if let EntryRef::Occupied(mut o) = map.entry_ref("poneyland") { + /// assert_eq!(o.insert(15), 12); + /// } + /// + /// assert_eq!(map["poneyland"], 15); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn insert(&mut self, value: V) -> V { + mem::replace(self.get_mut(), value) + } + + /// Takes the value out of the entry, and returns it. + /// Keeps the allocated memory for reuse. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// use hashbrown::hash_map::EntryRef; + /// + /// let mut map: HashMap<String, u32> = HashMap::new(); + /// // The map is empty + /// assert!(map.is_empty() && map.capacity() == 0); + /// + /// map.entry_ref("poneyland").or_insert(12); + /// + /// if let EntryRef::Occupied(o) = map.entry_ref("poneyland") { + /// assert_eq!(o.remove(), 12); + /// } + /// + /// assert_eq!(map.contains_key("poneyland"), false); + /// // Now map hold none elements but capacity is equal to the old one + /// assert!(map.is_empty()); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn remove(self) -> V { + self.remove_entry().1 + } + + /// Replaces the entry, returning the old key and value. The new key in the hash map will be + /// the key used to create this entry. + /// + /// # Panics + /// + /// Will panic if this OccupiedEntryRef was created through [`EntryRef::insert`]. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::hash_map::{EntryRef, HashMap}; + /// use std::rc::Rc; + /// + /// let mut map: HashMap<Rc<str>, u32> = HashMap::new(); + /// let key: Rc<str> = Rc::from("Stringthing"); + /// + /// map.insert(key.clone(), 15); + /// assert_eq!(Rc::strong_count(&key), 2); + /// + /// match map.entry_ref("Stringthing") { + /// EntryRef::Occupied(entry) => { + /// let (old_key, old_value): (Rc<str>, u32) = entry.replace_entry(16); + /// assert!(Rc::ptr_eq(&key, &old_key) && old_value == 15); + /// } + /// EntryRef::Vacant(_) => panic!(), + /// } + /// + /// assert_eq!(Rc::strong_count(&key), 1); + /// assert_eq!(map["Stringthing"], 16); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn replace_entry(self, value: V) -> (K, V) + where + K: From<&'b Q>, + { + let entry = unsafe { self.elem.as_mut() }; + + let old_key = mem::replace(&mut entry.0, self.key.unwrap().into_owned()); + let old_value = mem::replace(&mut entry.1, value); + + (old_key, old_value) + } + + /// Replaces the key in the hash map with the key used to create this entry. + /// + /// # Panics + /// + /// Will panic if this OccupiedEntryRef was created through [`EntryRef::insert`]. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::hash_map::{EntryRef, HashMap}; + /// use std::rc::Rc; + /// + /// let mut map: HashMap<Rc<str>, usize> = HashMap::with_capacity(6); + /// let mut keys: Vec<Rc<str>> = Vec::with_capacity(6); + /// + /// for (value, key) in ["a", "b", "c", "d", "e", "f"].into_iter().enumerate() { + /// let rc_key: Rc<str> = Rc::from(key); + /// keys.push(rc_key.clone()); + /// map.insert(rc_key.clone(), value); + /// } + /// + /// assert!(keys.iter().all(|key| Rc::strong_count(key) == 2)); + /// + /// // It doesn't matter that we kind of use a vector with the same keys, + /// // because all keys will be newly created from the references + /// reclaim_memory(&mut map, &keys); + /// + /// assert!(keys.iter().all(|key| Rc::strong_count(key) == 1)); + /// + /// fn reclaim_memory(map: &mut HashMap<Rc<str>, usize>, keys: &[Rc<str>]) { + /// for key in keys { + /// if let EntryRef::Occupied(entry) = map.entry_ref(key.as_ref()) { + /// // Replaces the entry's key with our version of it in `keys`. + /// entry.replace_key(); + /// } + /// } + /// } + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn replace_key(self) -> K + where + K: From<&'b Q>, + { + let entry = unsafe { self.elem.as_mut() }; + mem::replace(&mut entry.0, self.key.unwrap().into_owned()) + } + + /// Provides shared access to the key and owned access to the value of + /// the entry and allows to replace or remove it based on the + /// value of the returned option. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// use hashbrown::hash_map::EntryRef; + /// + /// let mut map: HashMap<String, u32> = HashMap::new(); + /// map.insert("poneyland".to_string(), 42); + /// + /// let entry = match map.entry_ref("poneyland") { + /// EntryRef::Occupied(e) => { + /// e.replace_entry_with(|k, v| { + /// assert_eq!(k, "poneyland"); + /// assert_eq!(v, 42); + /// Some(v + 1) + /// }) + /// } + /// EntryRef::Vacant(_) => panic!(), + /// }; + /// + /// match entry { + /// EntryRef::Occupied(e) => { + /// assert_eq!(e.key(), "poneyland"); + /// assert_eq!(e.get(), &43); + /// } + /// EntryRef::Vacant(_) => panic!(), + /// } + /// + /// assert_eq!(map["poneyland"], 43); + /// + /// let entry = match map.entry_ref("poneyland") { + /// EntryRef::Occupied(e) => e.replace_entry_with(|_k, _v| None), + /// EntryRef::Vacant(_) => panic!(), + /// }; + /// + /// match entry { + /// EntryRef::Vacant(e) => { + /// assert_eq!(e.key(), "poneyland"); + /// } + /// EntryRef::Occupied(_) => panic!(), + /// } + /// + /// assert!(!map.contains_key("poneyland")); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn replace_entry_with<F>(self, f: F) -> EntryRef<'a, 'b, K, Q, V, S, A> + where + F: FnOnce(&K, V) -> Option<V>, + { + unsafe { + let mut spare_key = None; + + self.table + .table + .replace_bucket_with(self.elem.clone(), |(key, value)| { + if let Some(new_value) = f(&key, value) { + Some((key, new_value)) + } else { + spare_key = Some(KeyOrRef::Owned(key)); + None + } + }); + + if let Some(key) = spare_key { + EntryRef::Vacant(VacantEntryRef { + hash: self.hash, + key, + table: self.table, + }) + } else { + EntryRef::Occupied(self) + } + } + } +} + +impl<'a, 'b, K, Q: ?Sized, V, S, A: Allocator> VacantEntryRef<'a, 'b, K, Q, V, S, A> { + /// Gets a reference to the key that would be used when inserting a value + /// through the `VacantEntryRef`. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let mut map: HashMap<String, u32> = HashMap::new(); + /// let key: &str = "poneyland"; + /// assert_eq!(map.entry_ref(key).key(), "poneyland"); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn key(&self) -> &Q + where + K: Borrow<Q>, + { + self.key.as_ref() + } + + /// Take ownership of the key. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::hash_map::{EntryRef, HashMap}; + /// + /// let mut map: HashMap<String, u32> = HashMap::new(); + /// let key: &str = "poneyland"; + /// + /// match map.entry_ref(key) { + /// EntryRef::Occupied(_) => panic!(), + /// EntryRef::Vacant(v) => assert_eq!(v.into_key(), "poneyland".to_owned()), + /// } + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn into_key(self) -> K + where + K: From<&'b Q>, + { + self.key.into_owned() + } + + /// Sets the value of the entry with the VacantEntryRef's key, + /// and returns a mutable reference to it. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// use hashbrown::hash_map::EntryRef; + /// + /// let mut map: HashMap<String, u32> = HashMap::new(); + /// let key: &str = "poneyland"; + /// + /// if let EntryRef::Vacant(o) = map.entry_ref(key) { + /// o.insert(37); + /// } + /// assert_eq!(map["poneyland"], 37); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn insert(self, value: V) -> &'a mut V + where + K: Hash + From<&'b Q>, + S: BuildHasher, + { + let table = &mut self.table.table; + let entry = table.insert_entry( + self.hash, + (self.key.into_owned(), value), + make_hasher::<_, V, S>(&self.table.hash_builder), + ); + &mut entry.1 + } + + #[cfg_attr(feature = "inline-more", inline)] + fn insert_entry(self, value: V) -> OccupiedEntryRef<'a, 'b, K, Q, V, S, A> + where + K: Hash + From<&'b Q>, + S: BuildHasher, + { + let elem = self.table.table.insert( + self.hash, + (self.key.into_owned(), value), + make_hasher::<_, V, S>(&self.table.hash_builder), + ); + OccupiedEntryRef { + hash: self.hash, + key: None, + elem, + table: self.table, + } + } +} + +impl<K, V, S, A> FromIterator<(K, V)> for HashMap<K, V, S, A> +where + K: Eq + Hash, + S: BuildHasher + Default, + A: Default + Allocator, +{ + #[cfg_attr(feature = "inline-more", inline)] + fn from_iter<T: IntoIterator<Item = (K, V)>>(iter: T) -> Self { + let iter = iter.into_iter(); + let mut map = + Self::with_capacity_and_hasher_in(iter.size_hint().0, S::default(), A::default()); + iter.for_each(|(k, v)| { + map.insert(k, v); + }); + map + } +} + +/// Inserts all new key-values from the iterator and replaces values with existing +/// keys with new values returned from the iterator. +impl<K, V, S, A> Extend<(K, V)> for HashMap<K, V, S, A> +where + K: Eq + Hash, + S: BuildHasher, + A: Allocator, +{ + /// Inserts all new key-values from the iterator to existing `HashMap<K, V, S, A>`. + /// Replace values with existing keys with new values returned from the iterator. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::hash_map::HashMap; + /// + /// let mut map = HashMap::new(); + /// map.insert(1, 100); + /// + /// let some_iter = [(1, 1), (2, 2)].into_iter(); + /// map.extend(some_iter); + /// // Replace values with existing keys with new values returned from the iterator. + /// // So that the map.get(&1) doesn't return Some(&100). + /// assert_eq!(map.get(&1), Some(&1)); + /// + /// let some_vec: Vec<_> = vec![(3, 3), (4, 4)]; + /// map.extend(some_vec); + /// + /// let some_arr = [(5, 5), (6, 6)]; + /// map.extend(some_arr); + /// let old_map_len = map.len(); + /// + /// // You can also extend from another HashMap + /// let mut new_map = HashMap::new(); + /// new_map.extend(map); + /// assert_eq!(new_map.len(), old_map_len); + /// + /// let mut vec: Vec<_> = new_map.into_iter().collect(); + /// // The `IntoIter` iterator produces items in arbitrary order, so the + /// // items must be sorted to test them against a sorted array. + /// vec.sort_unstable(); + /// assert_eq!(vec, [(1, 1), (2, 2), (3, 3), (4, 4), (5, 5), (6, 6)]); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + fn extend<T: IntoIterator<Item = (K, V)>>(&mut self, iter: T) { + // Keys may be already present or show multiple times in the iterator. + // Reserve the entire hint lower bound if the map is empty. + // Otherwise reserve half the hint (rounded up), so the map + // will only resize twice in the worst case. + let iter = iter.into_iter(); + let reserve = if self.is_empty() { + iter.size_hint().0 + } else { + (iter.size_hint().0 + 1) / 2 + }; + self.reserve(reserve); + iter.for_each(move |(k, v)| { + self.insert(k, v); + }); + } + + #[inline] + #[cfg(feature = "nightly")] + fn extend_one(&mut self, (k, v): (K, V)) { + self.insert(k, v); + } + + #[inline] + #[cfg(feature = "nightly")] + fn extend_reserve(&mut self, additional: usize) { + // Keys may be already present or show multiple times in the iterator. + // Reserve the entire hint lower bound if the map is empty. + // Otherwise reserve half the hint (rounded up), so the map + // will only resize twice in the worst case. + let reserve = if self.is_empty() { + additional + } else { + (additional + 1) / 2 + }; + self.reserve(reserve); + } +} + +/// Inserts all new key-values from the iterator and replaces values with existing +/// keys with new values returned from the iterator. +impl<'a, K, V, S, A> Extend<(&'a K, &'a V)> for HashMap<K, V, S, A> +where + K: Eq + Hash + Copy, + V: Copy, + S: BuildHasher, + A: Allocator, +{ + /// Inserts all new key-values from the iterator to existing `HashMap<K, V, S, A>`. + /// Replace values with existing keys with new values returned from the iterator. + /// The keys and values must implement [`Copy`] trait. + /// + /// [`Copy`]: https://doc.rust-lang.org/core/marker/trait.Copy.html + /// + /// # Examples + /// + /// ``` + /// use hashbrown::hash_map::HashMap; + /// + /// let mut map = HashMap::new(); + /// map.insert(1, 100); + /// + /// let arr = [(1, 1), (2, 2)]; + /// let some_iter = arr.iter().map(|(k, v)| (k, v)); + /// map.extend(some_iter); + /// // Replace values with existing keys with new values returned from the iterator. + /// // So that the map.get(&1) doesn't return Some(&100). + /// assert_eq!(map.get(&1), Some(&1)); + /// + /// let some_vec: Vec<_> = vec![(3, 3), (4, 4)]; + /// map.extend(some_vec.iter().map(|(k, v)| (k, v))); + /// + /// let some_arr = [(5, 5), (6, 6)]; + /// map.extend(some_arr.iter().map(|(k, v)| (k, v))); + /// + /// // You can also extend from another HashMap + /// let mut new_map = HashMap::new(); + /// new_map.extend(&map); + /// assert_eq!(new_map, map); + /// + /// let mut vec: Vec<_> = new_map.into_iter().collect(); + /// // The `IntoIter` iterator produces items in arbitrary order, so the + /// // items must be sorted to test them against a sorted array. + /// vec.sort_unstable(); + /// assert_eq!(vec, [(1, 1), (2, 2), (3, 3), (4, 4), (5, 5), (6, 6)]); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + fn extend<T: IntoIterator<Item = (&'a K, &'a V)>>(&mut self, iter: T) { + self.extend(iter.into_iter().map(|(&key, &value)| (key, value))); + } + + #[inline] + #[cfg(feature = "nightly")] + fn extend_one(&mut self, (k, v): (&'a K, &'a V)) { + self.insert(*k, *v); + } + + #[inline] + #[cfg(feature = "nightly")] + fn extend_reserve(&mut self, additional: usize) { + Extend::<(K, V)>::extend_reserve(self, additional); + } +} + +/// Inserts all new key-values from the iterator and replaces values with existing +/// keys with new values returned from the iterator. +impl<'a, K, V, S, A> Extend<&'a (K, V)> for HashMap<K, V, S, A> +where + K: Eq + Hash + Copy, + V: Copy, + S: BuildHasher, + A: Allocator, +{ + /// Inserts all new key-values from the iterator to existing `HashMap<K, V, S, A>`. + /// Replace values with existing keys with new values returned from the iterator. + /// The keys and values must implement [`Copy`] trait. + /// + /// [`Copy`]: https://doc.rust-lang.org/core/marker/trait.Copy.html + /// + /// # Examples + /// + /// ``` + /// use hashbrown::hash_map::HashMap; + /// + /// let mut map = HashMap::new(); + /// map.insert(1, 100); + /// + /// let arr = [(1, 1), (2, 2)]; + /// let some_iter = arr.iter(); + /// map.extend(some_iter); + /// // Replace values with existing keys with new values returned from the iterator. + /// // So that the map.get(&1) doesn't return Some(&100). + /// assert_eq!(map.get(&1), Some(&1)); + /// + /// let some_vec: Vec<_> = vec![(3, 3), (4, 4)]; + /// map.extend(&some_vec); + /// + /// let some_arr = [(5, 5), (6, 6)]; + /// map.extend(&some_arr); + /// + /// let mut vec: Vec<_> = map.into_iter().collect(); + /// // The `IntoIter` iterator produces items in arbitrary order, so the + /// // items must be sorted to test them against a sorted array. + /// vec.sort_unstable(); + /// assert_eq!(vec, [(1, 1), (2, 2), (3, 3), (4, 4), (5, 5), (6, 6)]); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + fn extend<T: IntoIterator<Item = &'a (K, V)>>(&mut self, iter: T) { + self.extend(iter.into_iter().map(|&(key, value)| (key, value))); + } + + #[inline] + #[cfg(feature = "nightly")] + fn extend_one(&mut self, &(k, v): &'a (K, V)) { + self.insert(k, v); + } + + #[inline] + #[cfg(feature = "nightly")] + fn extend_reserve(&mut self, additional: usize) { + Extend::<(K, V)>::extend_reserve(self, additional); + } +} + +#[allow(dead_code)] +fn assert_covariance() { + fn map_key<'new>(v: HashMap<&'static str, u8>) -> HashMap<&'new str, u8> { + v + } + fn map_val<'new>(v: HashMap<u8, &'static str>) -> HashMap<u8, &'new str> { + v + } + fn iter_key<'a, 'new>(v: Iter<'a, &'static str, u8>) -> Iter<'a, &'new str, u8> { + v + } + fn iter_val<'a, 'new>(v: Iter<'a, u8, &'static str>) -> Iter<'a, u8, &'new str> { + v + } + fn into_iter_key<'new, A: Allocator>( + v: IntoIter<&'static str, u8, A>, + ) -> IntoIter<&'new str, u8, A> { + v + } + fn into_iter_val<'new, A: Allocator>( + v: IntoIter<u8, &'static str, A>, + ) -> IntoIter<u8, &'new str, A> { + v + } + fn keys_key<'a, 'new>(v: Keys<'a, &'static str, u8>) -> Keys<'a, &'new str, u8> { + v + } + fn keys_val<'a, 'new>(v: Keys<'a, u8, &'static str>) -> Keys<'a, u8, &'new str> { + v + } + fn values_key<'a, 'new>(v: Values<'a, &'static str, u8>) -> Values<'a, &'new str, u8> { + v + } + fn values_val<'a, 'new>(v: Values<'a, u8, &'static str>) -> Values<'a, u8, &'new str> { + v + } + fn drain<'new>( + d: Drain<'static, &'static str, &'static str>, + ) -> Drain<'new, &'new str, &'new str> { + d + } +} + +#[cfg(test)] +mod test_map { + use super::DefaultHashBuilder; + use super::Entry::{Occupied, Vacant}; + use super::EntryRef; + use super::{HashMap, RawEntryMut}; + use alloc::string::{String, ToString}; + use alloc::sync::Arc; + use allocator_api2::alloc::{AllocError, Allocator, Global}; + use core::alloc::Layout; + use core::ptr::NonNull; + use core::sync::atomic::{AtomicI8, Ordering}; + use rand::{rngs::SmallRng, Rng, SeedableRng}; + use std::borrow::ToOwned; + use std::cell::RefCell; + use std::usize; + use std::vec::Vec; + + #[test] + fn test_zero_capacities() { + type HM = HashMap<i32, i32>; + + let m = HM::new(); + assert_eq!(m.capacity(), 0); + + let m = HM::default(); + assert_eq!(m.capacity(), 0); + + let m = HM::with_hasher(DefaultHashBuilder::default()); + assert_eq!(m.capacity(), 0); + + let m = HM::with_capacity(0); + assert_eq!(m.capacity(), 0); + + let m = HM::with_capacity_and_hasher(0, DefaultHashBuilder::default()); + assert_eq!(m.capacity(), 0); + + let mut m = HM::new(); + m.insert(1, 1); + m.insert(2, 2); + m.remove(&1); + m.remove(&2); + m.shrink_to_fit(); + assert_eq!(m.capacity(), 0); + + let mut m = HM::new(); + m.reserve(0); + assert_eq!(m.capacity(), 0); + } + + #[test] + fn test_create_capacity_zero() { + let mut m = HashMap::with_capacity(0); + + assert!(m.insert(1, 1).is_none()); + + assert!(m.contains_key(&1)); + assert!(!m.contains_key(&0)); + } + + #[test] + fn test_insert() { + let mut m = HashMap::new(); + assert_eq!(m.len(), 0); + assert!(m.insert(1, 2).is_none()); + assert_eq!(m.len(), 1); + assert!(m.insert(2, 4).is_none()); + assert_eq!(m.len(), 2); + assert_eq!(*m.get(&1).unwrap(), 2); + assert_eq!(*m.get(&2).unwrap(), 4); + } + + #[test] + fn test_clone() { + let mut m = HashMap::new(); + assert_eq!(m.len(), 0); + assert!(m.insert(1, 2).is_none()); + assert_eq!(m.len(), 1); + assert!(m.insert(2, 4).is_none()); + assert_eq!(m.len(), 2); + #[allow(clippy::redundant_clone)] + let m2 = m.clone(); + assert_eq!(*m2.get(&1).unwrap(), 2); + assert_eq!(*m2.get(&2).unwrap(), 4); + assert_eq!(m2.len(), 2); + } + + #[test] + fn test_clone_from() { + let mut m = HashMap::new(); + let mut m2 = HashMap::new(); + assert_eq!(m.len(), 0); + assert!(m.insert(1, 2).is_none()); + assert_eq!(m.len(), 1); + assert!(m.insert(2, 4).is_none()); + assert_eq!(m.len(), 2); + m2.clone_from(&m); + assert_eq!(*m2.get(&1).unwrap(), 2); + assert_eq!(*m2.get(&2).unwrap(), 4); + assert_eq!(m2.len(), 2); + } + + thread_local! { static DROP_VECTOR: RefCell<Vec<i32>> = RefCell::new(Vec::new()) } + + #[derive(Hash, PartialEq, Eq)] + struct Droppable { + k: usize, + } + + impl Droppable { + fn new(k: usize) -> Droppable { + DROP_VECTOR.with(|slot| { + slot.borrow_mut()[k] += 1; + }); + + Droppable { k } + } + } + + impl Drop for Droppable { + fn drop(&mut self) { + DROP_VECTOR.with(|slot| { + slot.borrow_mut()[self.k] -= 1; + }); + } + } + + impl Clone for Droppable { + fn clone(&self) -> Self { + Droppable::new(self.k) + } + } + + #[test] + fn test_drops() { + DROP_VECTOR.with(|slot| { + *slot.borrow_mut() = vec![0; 200]; + }); + + { + let mut m = HashMap::new(); + + DROP_VECTOR.with(|v| { + for i in 0..200 { + assert_eq!(v.borrow()[i], 0); + } + }); + + for i in 0..100 { + let d1 = Droppable::new(i); + let d2 = Droppable::new(i + 100); + m.insert(d1, d2); + } + + DROP_VECTOR.with(|v| { + for i in 0..200 { + assert_eq!(v.borrow()[i], 1); + } + }); + + for i in 0..50 { + let k = Droppable::new(i); + let v = m.remove(&k); + + assert!(v.is_some()); + + DROP_VECTOR.with(|v| { + assert_eq!(v.borrow()[i], 1); + assert_eq!(v.borrow()[i + 100], 1); + }); + } + + DROP_VECTOR.with(|v| { + for i in 0..50 { + assert_eq!(v.borrow()[i], 0); + assert_eq!(v.borrow()[i + 100], 0); + } + + for i in 50..100 { + assert_eq!(v.borrow()[i], 1); + assert_eq!(v.borrow()[i + 100], 1); + } + }); + } + + DROP_VECTOR.with(|v| { + for i in 0..200 { + assert_eq!(v.borrow()[i], 0); + } + }); + } + + #[test] + fn test_into_iter_drops() { + DROP_VECTOR.with(|v| { + *v.borrow_mut() = vec![0; 200]; + }); + + let hm = { + let mut hm = HashMap::new(); + + DROP_VECTOR.with(|v| { + for i in 0..200 { + assert_eq!(v.borrow()[i], 0); + } + }); + + for i in 0..100 { + let d1 = Droppable::new(i); + let d2 = Droppable::new(i + 100); + hm.insert(d1, d2); + } + + DROP_VECTOR.with(|v| { + for i in 0..200 { + assert_eq!(v.borrow()[i], 1); + } + }); + + hm + }; + + // By the way, ensure that cloning doesn't screw up the dropping. + drop(hm.clone()); + + { + let mut half = hm.into_iter().take(50); + + DROP_VECTOR.with(|v| { + for i in 0..200 { + assert_eq!(v.borrow()[i], 1); + } + }); + + for _ in half.by_ref() {} + + DROP_VECTOR.with(|v| { + let nk = (0..100).filter(|&i| v.borrow()[i] == 1).count(); + + let nv = (0..100).filter(|&i| v.borrow()[i + 100] == 1).count(); + + assert_eq!(nk, 50); + assert_eq!(nv, 50); + }); + }; + + DROP_VECTOR.with(|v| { + for i in 0..200 { + assert_eq!(v.borrow()[i], 0); + } + }); + } + + #[test] + fn test_empty_remove() { + let mut m: HashMap<i32, bool> = HashMap::new(); + assert_eq!(m.remove(&0), None); + } + + #[test] + fn test_empty_entry() { + let mut m: HashMap<i32, bool> = HashMap::new(); + match m.entry(0) { + Occupied(_) => panic!(), + Vacant(_) => {} + } + assert!(*m.entry(0).or_insert(true)); + assert_eq!(m.len(), 1); + } + + #[test] + fn test_empty_entry_ref() { + let mut m: HashMap<std::string::String, bool> = HashMap::new(); + match m.entry_ref("poneyland") { + EntryRef::Occupied(_) => panic!(), + EntryRef::Vacant(_) => {} + } + assert!(*m.entry_ref("poneyland").or_insert(true)); + assert_eq!(m.len(), 1); + } + + #[test] + fn test_empty_iter() { + let mut m: HashMap<i32, bool> = HashMap::new(); + assert_eq!(m.drain().next(), None); + assert_eq!(m.keys().next(), None); + assert_eq!(m.values().next(), None); + assert_eq!(m.values_mut().next(), None); + assert_eq!(m.iter().next(), None); + assert_eq!(m.iter_mut().next(), None); + assert_eq!(m.len(), 0); + assert!(m.is_empty()); + assert_eq!(m.into_iter().next(), None); + } + + #[test] + #[cfg_attr(miri, ignore)] // FIXME: takes too long + fn test_lots_of_insertions() { + let mut m = HashMap::new(); + + // Try this a few times to make sure we never screw up the hashmap's + // internal state. + for _ in 0..10 { + assert!(m.is_empty()); + + for i in 1..1001 { + assert!(m.insert(i, i).is_none()); + + for j in 1..=i { + let r = m.get(&j); + assert_eq!(r, Some(&j)); + } + + for j in i + 1..1001 { + let r = m.get(&j); + assert_eq!(r, None); + } + } + + for i in 1001..2001 { + assert!(!m.contains_key(&i)); + } + + // remove forwards + for i in 1..1001 { + assert!(m.remove(&i).is_some()); + + for j in 1..=i { + assert!(!m.contains_key(&j)); + } + + for j in i + 1..1001 { + assert!(m.contains_key(&j)); + } + } + + for i in 1..1001 { + assert!(!m.contains_key(&i)); + } + + for i in 1..1001 { + assert!(m.insert(i, i).is_none()); + } + + // remove backwards + for i in (1..1001).rev() { + assert!(m.remove(&i).is_some()); + + for j in i..1001 { + assert!(!m.contains_key(&j)); + } + + for j in 1..i { + assert!(m.contains_key(&j)); + } + } + } + } + + #[test] + fn test_find_mut() { + let mut m = HashMap::new(); + assert!(m.insert(1, 12).is_none()); + assert!(m.insert(2, 8).is_none()); + assert!(m.insert(5, 14).is_none()); + let new = 100; + match m.get_mut(&5) { + None => panic!(), + Some(x) => *x = new, + } + assert_eq!(m.get(&5), Some(&new)); + } + + #[test] + fn test_insert_overwrite() { + let mut m = HashMap::new(); + assert!(m.insert(1, 2).is_none()); + assert_eq!(*m.get(&1).unwrap(), 2); + assert!(m.insert(1, 3).is_some()); + assert_eq!(*m.get(&1).unwrap(), 3); + } + + #[test] + fn test_insert_conflicts() { + let mut m = HashMap::with_capacity(4); + assert!(m.insert(1, 2).is_none()); + assert!(m.insert(5, 3).is_none()); + assert!(m.insert(9, 4).is_none()); + assert_eq!(*m.get(&9).unwrap(), 4); + assert_eq!(*m.get(&5).unwrap(), 3); + assert_eq!(*m.get(&1).unwrap(), 2); + } + + #[test] + fn test_conflict_remove() { + let mut m = HashMap::with_capacity(4); + assert!(m.insert(1, 2).is_none()); + assert_eq!(*m.get(&1).unwrap(), 2); + assert!(m.insert(5, 3).is_none()); + assert_eq!(*m.get(&1).unwrap(), 2); + assert_eq!(*m.get(&5).unwrap(), 3); + assert!(m.insert(9, 4).is_none()); + assert_eq!(*m.get(&1).unwrap(), 2); + assert_eq!(*m.get(&5).unwrap(), 3); + assert_eq!(*m.get(&9).unwrap(), 4); + assert!(m.remove(&1).is_some()); + assert_eq!(*m.get(&9).unwrap(), 4); + assert_eq!(*m.get(&5).unwrap(), 3); + } + + #[test] + fn test_insert_unique_unchecked() { + let mut map = HashMap::new(); + let (k1, v1) = map.insert_unique_unchecked(10, 11); + assert_eq!((&10, &mut 11), (k1, v1)); + let (k2, v2) = map.insert_unique_unchecked(20, 21); + assert_eq!((&20, &mut 21), (k2, v2)); + assert_eq!(Some(&11), map.get(&10)); + assert_eq!(Some(&21), map.get(&20)); + assert_eq!(None, map.get(&30)); + } + + #[test] + fn test_is_empty() { + let mut m = HashMap::with_capacity(4); + assert!(m.insert(1, 2).is_none()); + assert!(!m.is_empty()); + assert!(m.remove(&1).is_some()); + assert!(m.is_empty()); + } + + #[test] + fn test_remove() { + let mut m = HashMap::new(); + m.insert(1, 2); + assert_eq!(m.remove(&1), Some(2)); + assert_eq!(m.remove(&1), None); + } + + #[test] + fn test_remove_entry() { + let mut m = HashMap::new(); + m.insert(1, 2); + assert_eq!(m.remove_entry(&1), Some((1, 2))); + assert_eq!(m.remove(&1), None); + } + + #[test] + fn test_iterate() { + let mut m = HashMap::with_capacity(4); + for i in 0..32 { + assert!(m.insert(i, i * 2).is_none()); + } + assert_eq!(m.len(), 32); + + let mut observed: u32 = 0; + + for (k, v) in &m { + assert_eq!(*v, *k * 2); + observed |= 1 << *k; + } + assert_eq!(observed, 0xFFFF_FFFF); + } + + #[test] + fn test_keys() { + let vec = vec![(1, 'a'), (2, 'b'), (3, 'c')]; + let map: HashMap<_, _> = vec.into_iter().collect(); + let keys: Vec<_> = map.keys().copied().collect(); + assert_eq!(keys.len(), 3); + assert!(keys.contains(&1)); + assert!(keys.contains(&2)); + assert!(keys.contains(&3)); + } + + #[test] + fn test_values() { + let vec = vec![(1, 'a'), (2, 'b'), (3, 'c')]; + let map: HashMap<_, _> = vec.into_iter().collect(); + let values: Vec<_> = map.values().copied().collect(); + assert_eq!(values.len(), 3); + assert!(values.contains(&'a')); + assert!(values.contains(&'b')); + assert!(values.contains(&'c')); + } + + #[test] + fn test_values_mut() { + let vec = vec![(1, 1), (2, 2), (3, 3)]; + let mut map: HashMap<_, _> = vec.into_iter().collect(); + for value in map.values_mut() { + *value *= 2; + } + let values: Vec<_> = map.values().copied().collect(); + assert_eq!(values.len(), 3); + assert!(values.contains(&2)); + assert!(values.contains(&4)); + assert!(values.contains(&6)); + } + + #[test] + fn test_into_keys() { + let vec = vec![(1, 'a'), (2, 'b'), (3, 'c')]; + let map: HashMap<_, _> = vec.into_iter().collect(); + let keys: Vec<_> = map.into_keys().collect(); + + assert_eq!(keys.len(), 3); + assert!(keys.contains(&1)); + assert!(keys.contains(&2)); + assert!(keys.contains(&3)); + } + + #[test] + fn test_into_values() { + let vec = vec![(1, 'a'), (2, 'b'), (3, 'c')]; + let map: HashMap<_, _> = vec.into_iter().collect(); + let values: Vec<_> = map.into_values().collect(); + + assert_eq!(values.len(), 3); + assert!(values.contains(&'a')); + assert!(values.contains(&'b')); + assert!(values.contains(&'c')); + } + + #[test] + fn test_find() { + let mut m = HashMap::new(); + assert!(m.get(&1).is_none()); + m.insert(1, 2); + match m.get(&1) { + None => panic!(), + Some(v) => assert_eq!(*v, 2), + } + } + + #[test] + fn test_eq() { + let mut m1 = HashMap::new(); + m1.insert(1, 2); + m1.insert(2, 3); + m1.insert(3, 4); + + let mut m2 = HashMap::new(); + m2.insert(1, 2); + m2.insert(2, 3); + + assert!(m1 != m2); + + m2.insert(3, 4); + + assert_eq!(m1, m2); + } + + #[test] + fn test_show() { + let mut map = HashMap::new(); + let empty: HashMap<i32, i32> = HashMap::new(); + + map.insert(1, 2); + map.insert(3, 4); + + let map_str = format!("{map:?}"); + + assert!(map_str == "{1: 2, 3: 4}" || map_str == "{3: 4, 1: 2}"); + assert_eq!(format!("{empty:?}"), "{}"); + } + + #[test] + fn test_expand() { + let mut m = HashMap::new(); + + assert_eq!(m.len(), 0); + assert!(m.is_empty()); + + let mut i = 0; + let old_raw_cap = m.raw_capacity(); + while old_raw_cap == m.raw_capacity() { + m.insert(i, i); + i += 1; + } + + assert_eq!(m.len(), i); + assert!(!m.is_empty()); + } + + #[test] + fn test_behavior_resize_policy() { + let mut m = HashMap::new(); + + assert_eq!(m.len(), 0); + assert_eq!(m.raw_capacity(), 1); + assert!(m.is_empty()); + + m.insert(0, 0); + m.remove(&0); + assert!(m.is_empty()); + let initial_raw_cap = m.raw_capacity(); + m.reserve(initial_raw_cap); + let raw_cap = m.raw_capacity(); + + assert_eq!(raw_cap, initial_raw_cap * 2); + + let mut i = 0; + for _ in 0..raw_cap * 3 / 4 { + m.insert(i, i); + i += 1; + } + // three quarters full + + assert_eq!(m.len(), i); + assert_eq!(m.raw_capacity(), raw_cap); + + for _ in 0..raw_cap / 4 { + m.insert(i, i); + i += 1; + } + // half full + + let new_raw_cap = m.raw_capacity(); + assert_eq!(new_raw_cap, raw_cap * 2); + + for _ in 0..raw_cap / 2 - 1 { + i -= 1; + m.remove(&i); + assert_eq!(m.raw_capacity(), new_raw_cap); + } + // A little more than one quarter full. + m.shrink_to_fit(); + assert_eq!(m.raw_capacity(), raw_cap); + // again, a little more than half full + for _ in 0..raw_cap / 2 { + i -= 1; + m.remove(&i); + } + m.shrink_to_fit(); + + assert_eq!(m.len(), i); + assert!(!m.is_empty()); + assert_eq!(m.raw_capacity(), initial_raw_cap); + } + + #[test] + fn test_reserve_shrink_to_fit() { + let mut m = HashMap::new(); + m.insert(0, 0); + m.remove(&0); + assert!(m.capacity() >= m.len()); + for i in 0..128 { + m.insert(i, i); + } + m.reserve(256); + + let usable_cap = m.capacity(); + for i in 128..(128 + 256) { + m.insert(i, i); + assert_eq!(m.capacity(), usable_cap); + } + + for i in 100..(128 + 256) { + assert_eq!(m.remove(&i), Some(i)); + } + m.shrink_to_fit(); + + assert_eq!(m.len(), 100); + assert!(!m.is_empty()); + assert!(m.capacity() >= m.len()); + + for i in 0..100 { + assert_eq!(m.remove(&i), Some(i)); + } + m.shrink_to_fit(); + m.insert(0, 0); + + assert_eq!(m.len(), 1); + assert!(m.capacity() >= m.len()); + assert_eq!(m.remove(&0), Some(0)); + } + + #[test] + fn test_from_iter() { + let xs = [(1, 1), (2, 2), (2, 2), (3, 3), (4, 4), (5, 5), (6, 6)]; + + let map: HashMap<_, _> = xs.iter().copied().collect(); + + for &(k, v) in &xs { + assert_eq!(map.get(&k), Some(&v)); + } + + assert_eq!(map.iter().len(), xs.len() - 1); + } + + #[test] + fn test_size_hint() { + let xs = [(1, 1), (2, 2), (3, 3), (4, 4), (5, 5), (6, 6)]; + + let map: HashMap<_, _> = xs.iter().copied().collect(); + + let mut iter = map.iter(); + + for _ in iter.by_ref().take(3) {} + + assert_eq!(iter.size_hint(), (3, Some(3))); + } + + #[test] + fn test_iter_len() { + let xs = [(1, 1), (2, 2), (3, 3), (4, 4), (5, 5), (6, 6)]; + + let map: HashMap<_, _> = xs.iter().copied().collect(); + + let mut iter = map.iter(); + + for _ in iter.by_ref().take(3) {} + + assert_eq!(iter.len(), 3); + } + + #[test] + fn test_mut_size_hint() { + let xs = [(1, 1), (2, 2), (3, 3), (4, 4), (5, 5), (6, 6)]; + + let mut map: HashMap<_, _> = xs.iter().copied().collect(); + + let mut iter = map.iter_mut(); + + for _ in iter.by_ref().take(3) {} + + assert_eq!(iter.size_hint(), (3, Some(3))); + } + + #[test] + fn test_iter_mut_len() { + let xs = [(1, 1), (2, 2), (3, 3), (4, 4), (5, 5), (6, 6)]; + + let mut map: HashMap<_, _> = xs.iter().copied().collect(); + + let mut iter = map.iter_mut(); + + for _ in iter.by_ref().take(3) {} + + assert_eq!(iter.len(), 3); + } + + #[test] + fn test_index() { + let mut map = HashMap::new(); + + map.insert(1, 2); + map.insert(2, 1); + map.insert(3, 4); + + assert_eq!(map[&2], 1); + } + + #[test] + #[should_panic] + fn test_index_nonexistent() { + let mut map = HashMap::new(); + + map.insert(1, 2); + map.insert(2, 1); + map.insert(3, 4); + + #[allow(clippy::no_effect)] // false positive lint + map[&4]; + } + + #[test] + fn test_entry() { + let xs = [(1, 10), (2, 20), (3, 30), (4, 40), (5, 50), (6, 60)]; + + let mut map: HashMap<_, _> = xs.iter().copied().collect(); + + // Existing key (insert) + match map.entry(1) { + Vacant(_) => unreachable!(), + Occupied(mut view) => { + assert_eq!(view.get(), &10); + assert_eq!(view.insert(100), 10); + } + } + assert_eq!(map.get(&1).unwrap(), &100); + assert_eq!(map.len(), 6); + + // Existing key (update) + match map.entry(2) { + Vacant(_) => unreachable!(), + Occupied(mut view) => { + let v = view.get_mut(); + let new_v = (*v) * 10; + *v = new_v; + } + } + assert_eq!(map.get(&2).unwrap(), &200); + assert_eq!(map.len(), 6); + + // Existing key (take) + match map.entry(3) { + Vacant(_) => unreachable!(), + Occupied(view) => { + assert_eq!(view.remove(), 30); + } + } + assert_eq!(map.get(&3), None); + assert_eq!(map.len(), 5); + + // Inexistent key (insert) + match map.entry(10) { + Occupied(_) => unreachable!(), + Vacant(view) => { + assert_eq!(*view.insert(1000), 1000); + } + } + assert_eq!(map.get(&10).unwrap(), &1000); + assert_eq!(map.len(), 6); + } + + #[test] + fn test_entry_ref() { + let xs = [ + ("One".to_owned(), 10), + ("Two".to_owned(), 20), + ("Three".to_owned(), 30), + ("Four".to_owned(), 40), + ("Five".to_owned(), 50), + ("Six".to_owned(), 60), + ]; + + let mut map: HashMap<_, _> = xs.iter().cloned().collect(); + + // Existing key (insert) + match map.entry_ref("One") { + EntryRef::Vacant(_) => unreachable!(), + EntryRef::Occupied(mut view) => { + assert_eq!(view.get(), &10); + assert_eq!(view.insert(100), 10); + } + } + assert_eq!(map.get("One").unwrap(), &100); + assert_eq!(map.len(), 6); + + // Existing key (update) + match map.entry_ref("Two") { + EntryRef::Vacant(_) => unreachable!(), + EntryRef::Occupied(mut view) => { + let v = view.get_mut(); + let new_v = (*v) * 10; + *v = new_v; + } + } + assert_eq!(map.get("Two").unwrap(), &200); + assert_eq!(map.len(), 6); + + // Existing key (take) + match map.entry_ref("Three") { + EntryRef::Vacant(_) => unreachable!(), + EntryRef::Occupied(view) => { + assert_eq!(view.remove(), 30); + } + } + assert_eq!(map.get("Three"), None); + assert_eq!(map.len(), 5); + + // Inexistent key (insert) + match map.entry_ref("Ten") { + EntryRef::Occupied(_) => unreachable!(), + EntryRef::Vacant(view) => { + assert_eq!(*view.insert(1000), 1000); + } + } + assert_eq!(map.get("Ten").unwrap(), &1000); + assert_eq!(map.len(), 6); + } + + #[test] + fn test_entry_take_doesnt_corrupt() { + #![allow(deprecated)] //rand + // Test for #19292 + fn check(m: &HashMap<i32, ()>) { + for k in m.keys() { + assert!(m.contains_key(k), "{k} is in keys() but not in the map?"); + } + } + + let mut m = HashMap::new(); + + let mut rng = { + let seed = u64::from_le_bytes(*b"testseed"); + SmallRng::seed_from_u64(seed) + }; + + // Populate the map with some items. + for _ in 0..50 { + let x = rng.gen_range(-10..10); + m.insert(x, ()); + } + + for _ in 0..1000 { + let x = rng.gen_range(-10..10); + match m.entry(x) { + Vacant(_) => {} + Occupied(e) => { + e.remove(); + } + } + + check(&m); + } + } + + #[test] + fn test_entry_ref_take_doesnt_corrupt() { + #![allow(deprecated)] //rand + // Test for #19292 + fn check(m: &HashMap<std::string::String, ()>) { + for k in m.keys() { + assert!(m.contains_key(k), "{k} is in keys() but not in the map?"); + } + } + + let mut m = HashMap::new(); + + let mut rng = { + let seed = u64::from_le_bytes(*b"testseed"); + SmallRng::seed_from_u64(seed) + }; + + // Populate the map with some items. + for _ in 0..50 { + let mut x = std::string::String::with_capacity(1); + x.push(rng.gen_range('a'..='z')); + m.insert(x, ()); + } + + for _ in 0..1000 { + let mut x = std::string::String::with_capacity(1); + x.push(rng.gen_range('a'..='z')); + match m.entry_ref(x.as_str()) { + EntryRef::Vacant(_) => {} + EntryRef::Occupied(e) => { + e.remove(); + } + } + + check(&m); + } + } + + #[test] + fn test_extend_ref_k_ref_v() { + let mut a = HashMap::new(); + a.insert(1, "one"); + let mut b = HashMap::new(); + b.insert(2, "two"); + b.insert(3, "three"); + + a.extend(&b); + + assert_eq!(a.len(), 3); + assert_eq!(a[&1], "one"); + assert_eq!(a[&2], "two"); + assert_eq!(a[&3], "three"); + } + + #[test] + #[allow(clippy::needless_borrow)] + fn test_extend_ref_kv_tuple() { + use std::ops::AddAssign; + let mut a = HashMap::new(); + a.insert(0, 0); + + fn create_arr<T: AddAssign<T> + Copy, const N: usize>(start: T, step: T) -> [(T, T); N] { + let mut outs: [(T, T); N] = [(start, start); N]; + let mut element = step; + outs.iter_mut().skip(1).for_each(|(k, v)| { + *k += element; + *v += element; + element += step; + }); + outs + } + + let for_iter: Vec<_> = (0..100).map(|i| (i, i)).collect(); + let iter = for_iter.iter(); + let vec: Vec<_> = (100..200).map(|i| (i, i)).collect(); + a.extend(iter); + a.extend(&vec); + a.extend(create_arr::<i32, 100>(200, 1)); + + assert_eq!(a.len(), 300); + + for item in 0..300 { + assert_eq!(a[&item], item); + } + } + + #[test] + fn test_capacity_not_less_than_len() { + let mut a = HashMap::new(); + let mut item = 0; + + for _ in 0..116 { + a.insert(item, 0); + item += 1; + } + + assert!(a.capacity() > a.len()); + + let free = a.capacity() - a.len(); + for _ in 0..free { + a.insert(item, 0); + item += 1; + } + + assert_eq!(a.len(), a.capacity()); + + // Insert at capacity should cause allocation. + a.insert(item, 0); + assert!(a.capacity() > a.len()); + } + + #[test] + fn test_occupied_entry_key() { + let mut a = HashMap::new(); + let key = "hello there"; + let value = "value goes here"; + assert!(a.is_empty()); + a.insert(key, value); + assert_eq!(a.len(), 1); + assert_eq!(a[key], value); + + match a.entry(key) { + Vacant(_) => panic!(), + Occupied(e) => assert_eq!(key, *e.key()), + } + assert_eq!(a.len(), 1); + assert_eq!(a[key], value); + } + + #[test] + fn test_occupied_entry_ref_key() { + let mut a = HashMap::new(); + let key = "hello there"; + let value = "value goes here"; + assert!(a.is_empty()); + a.insert(key.to_owned(), value); + assert_eq!(a.len(), 1); + assert_eq!(a[key], value); + + match a.entry_ref(key) { + EntryRef::Vacant(_) => panic!(), + EntryRef::Occupied(e) => assert_eq!(key, e.key()), + } + assert_eq!(a.len(), 1); + assert_eq!(a[key], value); + } + + #[test] + fn test_vacant_entry_key() { + let mut a = HashMap::new(); + let key = "hello there"; + let value = "value goes here"; + + assert!(a.is_empty()); + match a.entry(key) { + Occupied(_) => panic!(), + Vacant(e) => { + assert_eq!(key, *e.key()); + e.insert(value); + } + } + assert_eq!(a.len(), 1); + assert_eq!(a[key], value); + } + + #[test] + fn test_vacant_entry_ref_key() { + let mut a: HashMap<std::string::String, &str> = HashMap::new(); + let key = "hello there"; + let value = "value goes here"; + + assert!(a.is_empty()); + match a.entry_ref(key) { + EntryRef::Occupied(_) => panic!(), + EntryRef::Vacant(e) => { + assert_eq!(key, e.key()); + e.insert(value); + } + } + assert_eq!(a.len(), 1); + assert_eq!(a[key], value); + } + + #[test] + fn test_occupied_entry_replace_entry_with() { + let mut a = HashMap::new(); + + let key = "a key"; + let value = "an initial value"; + let new_value = "a new value"; + + let entry = a.entry(key).insert(value).replace_entry_with(|k, v| { + assert_eq!(k, &key); + assert_eq!(v, value); + Some(new_value) + }); + + match entry { + Occupied(e) => { + assert_eq!(e.key(), &key); + assert_eq!(e.get(), &new_value); + } + Vacant(_) => panic!(), + } + + assert_eq!(a[key], new_value); + assert_eq!(a.len(), 1); + + let entry = match a.entry(key) { + Occupied(e) => e.replace_entry_with(|k, v| { + assert_eq!(k, &key); + assert_eq!(v, new_value); + None + }), + Vacant(_) => panic!(), + }; + + match entry { + Vacant(e) => assert_eq!(e.key(), &key), + Occupied(_) => panic!(), + } + + assert!(!a.contains_key(key)); + assert_eq!(a.len(), 0); + } + + #[test] + fn test_occupied_entry_ref_replace_entry_with() { + let mut a: HashMap<std::string::String, &str> = HashMap::new(); + + let key = "a key"; + let value = "an initial value"; + let new_value = "a new value"; + + let entry = a.entry_ref(key).insert(value).replace_entry_with(|k, v| { + assert_eq!(k, key); + assert_eq!(v, value); + Some(new_value) + }); + + match entry { + EntryRef::Occupied(e) => { + assert_eq!(e.key(), key); + assert_eq!(e.get(), &new_value); + } + EntryRef::Vacant(_) => panic!(), + } + + assert_eq!(a[key], new_value); + assert_eq!(a.len(), 1); + + let entry = match a.entry_ref(key) { + EntryRef::Occupied(e) => e.replace_entry_with(|k, v| { + assert_eq!(k, key); + assert_eq!(v, new_value); + None + }), + EntryRef::Vacant(_) => panic!(), + }; + + match entry { + EntryRef::Vacant(e) => assert_eq!(e.key(), key), + EntryRef::Occupied(_) => panic!(), + } + + assert!(!a.contains_key(key)); + assert_eq!(a.len(), 0); + } + + #[test] + fn test_entry_and_replace_entry_with() { + let mut a = HashMap::new(); + + let key = "a key"; + let value = "an initial value"; + let new_value = "a new value"; + + let entry = a.entry(key).and_replace_entry_with(|_, _| panic!()); + + match entry { + Vacant(e) => assert_eq!(e.key(), &key), + Occupied(_) => panic!(), + } + + a.insert(key, value); + + let entry = a.entry(key).and_replace_entry_with(|k, v| { + assert_eq!(k, &key); + assert_eq!(v, value); + Some(new_value) + }); + + match entry { + Occupied(e) => { + assert_eq!(e.key(), &key); + assert_eq!(e.get(), &new_value); + } + Vacant(_) => panic!(), + } + + assert_eq!(a[key], new_value); + assert_eq!(a.len(), 1); + + let entry = a.entry(key).and_replace_entry_with(|k, v| { + assert_eq!(k, &key); + assert_eq!(v, new_value); + None + }); + + match entry { + Vacant(e) => assert_eq!(e.key(), &key), + Occupied(_) => panic!(), + } + + assert!(!a.contains_key(key)); + assert_eq!(a.len(), 0); + } + + #[test] + fn test_entry_ref_and_replace_entry_with() { + let mut a = HashMap::new(); + + let key = "a key"; + let value = "an initial value"; + let new_value = "a new value"; + + let entry = a.entry_ref(key).and_replace_entry_with(|_, _| panic!()); + + match entry { + EntryRef::Vacant(e) => assert_eq!(e.key(), key), + EntryRef::Occupied(_) => panic!(), + } + + a.insert(key.to_owned(), value); + + let entry = a.entry_ref(key).and_replace_entry_with(|k, v| { + assert_eq!(k, key); + assert_eq!(v, value); + Some(new_value) + }); + + match entry { + EntryRef::Occupied(e) => { + assert_eq!(e.key(), key); + assert_eq!(e.get(), &new_value); + } + EntryRef::Vacant(_) => panic!(), + } + + assert_eq!(a[key], new_value); + assert_eq!(a.len(), 1); + + let entry = a.entry_ref(key).and_replace_entry_with(|k, v| { + assert_eq!(k, key); + assert_eq!(v, new_value); + None + }); + + match entry { + EntryRef::Vacant(e) => assert_eq!(e.key(), key), + EntryRef::Occupied(_) => panic!(), + } + + assert!(!a.contains_key(key)); + assert_eq!(a.len(), 0); + } + + #[test] + fn test_raw_occupied_entry_replace_entry_with() { + let mut a = HashMap::new(); + + let key = "a key"; + let value = "an initial value"; + let new_value = "a new value"; + + let entry = a + .raw_entry_mut() + .from_key(&key) + .insert(key, value) + .replace_entry_with(|k, v| { + assert_eq!(k, &key); + assert_eq!(v, value); + Some(new_value) + }); + + match entry { + RawEntryMut::Occupied(e) => { + assert_eq!(e.key(), &key); + assert_eq!(e.get(), &new_value); + } + RawEntryMut::Vacant(_) => panic!(), + } + + assert_eq!(a[key], new_value); + assert_eq!(a.len(), 1); + + let entry = match a.raw_entry_mut().from_key(&key) { + RawEntryMut::Occupied(e) => e.replace_entry_with(|k, v| { + assert_eq!(k, &key); + assert_eq!(v, new_value); + None + }), + RawEntryMut::Vacant(_) => panic!(), + }; + + match entry { + RawEntryMut::Vacant(_) => {} + RawEntryMut::Occupied(_) => panic!(), + } + + assert!(!a.contains_key(key)); + assert_eq!(a.len(), 0); + } + + #[test] + fn test_raw_entry_and_replace_entry_with() { + let mut a = HashMap::new(); + + let key = "a key"; + let value = "an initial value"; + let new_value = "a new value"; + + let entry = a + .raw_entry_mut() + .from_key(&key) + .and_replace_entry_with(|_, _| panic!()); + + match entry { + RawEntryMut::Vacant(_) => {} + RawEntryMut::Occupied(_) => panic!(), + } + + a.insert(key, value); + + let entry = a + .raw_entry_mut() + .from_key(&key) + .and_replace_entry_with(|k, v| { + assert_eq!(k, &key); + assert_eq!(v, value); + Some(new_value) + }); + + match entry { + RawEntryMut::Occupied(e) => { + assert_eq!(e.key(), &key); + assert_eq!(e.get(), &new_value); + } + RawEntryMut::Vacant(_) => panic!(), + } + + assert_eq!(a[key], new_value); + assert_eq!(a.len(), 1); + + let entry = a + .raw_entry_mut() + .from_key(&key) + .and_replace_entry_with(|k, v| { + assert_eq!(k, &key); + assert_eq!(v, new_value); + None + }); + + match entry { + RawEntryMut::Vacant(_) => {} + RawEntryMut::Occupied(_) => panic!(), + } + + assert!(!a.contains_key(key)); + assert_eq!(a.len(), 0); + } + + #[test] + fn test_replace_entry_with_doesnt_corrupt() { + #![allow(deprecated)] //rand + // Test for #19292 + fn check(m: &HashMap<i32, ()>) { + for k in m.keys() { + assert!(m.contains_key(k), "{k} is in keys() but not in the map?"); + } + } + + let mut m = HashMap::new(); + + let mut rng = { + let seed = u64::from_le_bytes(*b"testseed"); + SmallRng::seed_from_u64(seed) + }; + + // Populate the map with some items. + for _ in 0..50 { + let x = rng.gen_range(-10..10); + m.insert(x, ()); + } + + for _ in 0..1000 { + let x = rng.gen_range(-10..10); + m.entry(x).and_replace_entry_with(|_, _| None); + check(&m); + } + } + + #[test] + fn test_replace_entry_ref_with_doesnt_corrupt() { + #![allow(deprecated)] //rand + // Test for #19292 + fn check(m: &HashMap<std::string::String, ()>) { + for k in m.keys() { + assert!(m.contains_key(k), "{k} is in keys() but not in the map?"); + } + } + + let mut m = HashMap::new(); + + let mut rng = { + let seed = u64::from_le_bytes(*b"testseed"); + SmallRng::seed_from_u64(seed) + }; + + // Populate the map with some items. + for _ in 0..50 { + let mut x = std::string::String::with_capacity(1); + x.push(rng.gen_range('a'..='z')); + m.insert(x, ()); + } + + for _ in 0..1000 { + let mut x = std::string::String::with_capacity(1); + x.push(rng.gen_range('a'..='z')); + m.entry_ref(x.as_str()).and_replace_entry_with(|_, _| None); + check(&m); + } + } + + #[test] + fn test_retain() { + let mut map: HashMap<i32, i32> = (0..100).map(|x| (x, x * 10)).collect(); + + map.retain(|&k, _| k % 2 == 0); + assert_eq!(map.len(), 50); + assert_eq!(map[&2], 20); + assert_eq!(map[&4], 40); + assert_eq!(map[&6], 60); + } + + #[test] + fn test_extract_if() { + { + let mut map: HashMap<i32, i32> = (0..8).map(|x| (x, x * 10)).collect(); + let drained = map.extract_if(|&k, _| k % 2 == 0); + let mut out = drained.collect::<Vec<_>>(); + out.sort_unstable(); + assert_eq!(vec![(0, 0), (2, 20), (4, 40), (6, 60)], out); + assert_eq!(map.len(), 4); + } + { + let mut map: HashMap<i32, i32> = (0..8).map(|x| (x, x * 10)).collect(); + map.extract_if(|&k, _| k % 2 == 0).for_each(drop); + assert_eq!(map.len(), 4); + } + } + + #[test] + #[cfg_attr(miri, ignore)] // FIXME: no OOM signalling (https://github.com/rust-lang/miri/issues/613) + fn test_try_reserve() { + use crate::TryReserveError::{AllocError, CapacityOverflow}; + + const MAX_ISIZE: usize = isize::MAX as usize; + + let mut empty_bytes: HashMap<u8, u8> = HashMap::new(); + + if let Err(CapacityOverflow) = empty_bytes.try_reserve(usize::MAX) { + } else { + panic!("usize::MAX should trigger an overflow!"); + } + + if let Err(CapacityOverflow) = empty_bytes.try_reserve(MAX_ISIZE) { + } else { + panic!("isize::MAX should trigger an overflow!"); + } + + if let Err(AllocError { .. }) = empty_bytes.try_reserve(MAX_ISIZE / 5) { + } else { + // This may succeed if there is enough free memory. Attempt to + // allocate a few more hashmaps to ensure the allocation will fail. + let mut empty_bytes2: HashMap<u8, u8> = HashMap::new(); + let _ = empty_bytes2.try_reserve(MAX_ISIZE / 5); + let mut empty_bytes3: HashMap<u8, u8> = HashMap::new(); + let _ = empty_bytes3.try_reserve(MAX_ISIZE / 5); + let mut empty_bytes4: HashMap<u8, u8> = HashMap::new(); + if let Err(AllocError { .. }) = empty_bytes4.try_reserve(MAX_ISIZE / 5) { + } else { + panic!("isize::MAX / 5 should trigger an OOM!"); + } + } + } + + #[test] + fn test_raw_entry() { + use super::RawEntryMut::{Occupied, Vacant}; + + let xs = [(1_i32, 10_i32), (2, 20), (3, 30), (4, 40), (5, 50), (6, 60)]; + + let mut map: HashMap<_, _> = xs.iter().copied().collect(); + + let compute_hash = |map: &HashMap<i32, i32>, k: i32| -> u64 { + super::make_hash::<i32, _>(map.hasher(), &k) + }; + + // Existing key (insert) + match map.raw_entry_mut().from_key(&1) { + Vacant(_) => unreachable!(), + Occupied(mut view) => { + assert_eq!(view.get(), &10); + assert_eq!(view.insert(100), 10); + } + } + let hash1 = compute_hash(&map, 1); + assert_eq!(map.raw_entry().from_key(&1).unwrap(), (&1, &100)); + assert_eq!( + map.raw_entry().from_hash(hash1, |k| *k == 1).unwrap(), + (&1, &100) + ); + assert_eq!( + map.raw_entry().from_key_hashed_nocheck(hash1, &1).unwrap(), + (&1, &100) + ); + assert_eq!(map.len(), 6); + + // Existing key (update) + match map.raw_entry_mut().from_key(&2) { + Vacant(_) => unreachable!(), + Occupied(mut view) => { + let v = view.get_mut(); + let new_v = (*v) * 10; + *v = new_v; + } + } + let hash2 = compute_hash(&map, 2); + assert_eq!(map.raw_entry().from_key(&2).unwrap(), (&2, &200)); + assert_eq!( + map.raw_entry().from_hash(hash2, |k| *k == 2).unwrap(), + (&2, &200) + ); + assert_eq!( + map.raw_entry().from_key_hashed_nocheck(hash2, &2).unwrap(), + (&2, &200) + ); + assert_eq!(map.len(), 6); + + // Existing key (take) + let hash3 = compute_hash(&map, 3); + match map.raw_entry_mut().from_key_hashed_nocheck(hash3, &3) { + Vacant(_) => unreachable!(), + Occupied(view) => { + assert_eq!(view.remove_entry(), (3, 30)); + } + } + assert_eq!(map.raw_entry().from_key(&3), None); + assert_eq!(map.raw_entry().from_hash(hash3, |k| *k == 3), None); + assert_eq!(map.raw_entry().from_key_hashed_nocheck(hash3, &3), None); + assert_eq!(map.len(), 5); + + // Nonexistent key (insert) + match map.raw_entry_mut().from_key(&10) { + Occupied(_) => unreachable!(), + Vacant(view) => { + assert_eq!(view.insert(10, 1000), (&mut 10, &mut 1000)); + } + } + assert_eq!(map.raw_entry().from_key(&10).unwrap(), (&10, &1000)); + assert_eq!(map.len(), 6); + + // Ensure all lookup methods produce equivalent results. + for k in 0..12 { + let hash = compute_hash(&map, k); + let v = map.get(&k).copied(); + let kv = v.as_ref().map(|v| (&k, v)); + + assert_eq!(map.raw_entry().from_key(&k), kv); + assert_eq!(map.raw_entry().from_hash(hash, |q| *q == k), kv); + assert_eq!(map.raw_entry().from_key_hashed_nocheck(hash, &k), kv); + + match map.raw_entry_mut().from_key(&k) { + Occupied(o) => assert_eq!(Some(o.get_key_value()), kv), + Vacant(_) => assert_eq!(v, None), + } + match map.raw_entry_mut().from_key_hashed_nocheck(hash, &k) { + Occupied(o) => assert_eq!(Some(o.get_key_value()), kv), + Vacant(_) => assert_eq!(v, None), + } + match map.raw_entry_mut().from_hash(hash, |q| *q == k) { + Occupied(o) => assert_eq!(Some(o.get_key_value()), kv), + Vacant(_) => assert_eq!(v, None), + } + } + } + + #[test] + fn test_key_without_hash_impl() { + #[derive(Debug)] + struct IntWrapper(u64); + + let mut m: HashMap<IntWrapper, (), ()> = HashMap::default(); + { + assert!(m.raw_entry().from_hash(0, |k| k.0 == 0).is_none()); + } + { + let vacant_entry = match m.raw_entry_mut().from_hash(0, |k| k.0 == 0) { + RawEntryMut::Occupied(..) => panic!("Found entry for key 0"), + RawEntryMut::Vacant(e) => e, + }; + vacant_entry.insert_with_hasher(0, IntWrapper(0), (), |k| k.0); + } + { + assert!(m.raw_entry().from_hash(0, |k| k.0 == 0).is_some()); + assert!(m.raw_entry().from_hash(1, |k| k.0 == 1).is_none()); + assert!(m.raw_entry().from_hash(2, |k| k.0 == 2).is_none()); + } + { + let vacant_entry = match m.raw_entry_mut().from_hash(1, |k| k.0 == 1) { + RawEntryMut::Occupied(..) => panic!("Found entry for key 1"), + RawEntryMut::Vacant(e) => e, + }; + vacant_entry.insert_with_hasher(1, IntWrapper(1), (), |k| k.0); + } + { + assert!(m.raw_entry().from_hash(0, |k| k.0 == 0).is_some()); + assert!(m.raw_entry().from_hash(1, |k| k.0 == 1).is_some()); + assert!(m.raw_entry().from_hash(2, |k| k.0 == 2).is_none()); + } + { + let occupied_entry = match m.raw_entry_mut().from_hash(0, |k| k.0 == 0) { + RawEntryMut::Occupied(e) => e, + RawEntryMut::Vacant(..) => panic!("Couldn't find entry for key 0"), + }; + occupied_entry.remove(); + } + assert!(m.raw_entry().from_hash(0, |k| k.0 == 0).is_none()); + assert!(m.raw_entry().from_hash(1, |k| k.0 == 1).is_some()); + assert!(m.raw_entry().from_hash(2, |k| k.0 == 2).is_none()); + } + + #[test] + #[cfg(feature = "raw")] + fn test_into_iter_refresh() { + #[cfg(miri)] + const N: usize = 32; + #[cfg(not(miri))] + const N: usize = 128; + + let mut rng = rand::thread_rng(); + for n in 0..N { + let mut map = HashMap::new(); + for i in 0..n { + assert!(map.insert(i, 2 * i).is_none()); + } + let hash_builder = map.hasher().clone(); + + let mut it = unsafe { map.table.iter() }; + assert_eq!(it.len(), n); + + let mut i = 0; + let mut left = n; + let mut removed = Vec::new(); + loop { + // occasionally remove some elements + if i < n && rng.gen_bool(0.1) { + let hash_value = super::make_hash(&hash_builder, &i); + + unsafe { + let e = map.table.find(hash_value, |q| q.0.eq(&i)); + if let Some(e) = e { + it.reflect_remove(&e); + let t = map.table.remove(e).0; + removed.push(t); + left -= 1; + } else { + assert!(removed.contains(&(i, 2 * i)), "{i} not in {removed:?}"); + let e = map.table.insert( + hash_value, + (i, 2 * i), + super::make_hasher::<_, usize, _>(&hash_builder), + ); + it.reflect_insert(&e); + if let Some(p) = removed.iter().position(|e| e == &(i, 2 * i)) { + removed.swap_remove(p); + } + left += 1; + } + } + } + + let e = it.next(); + if e.is_none() { + break; + } + assert!(i < n); + let t = unsafe { e.unwrap().as_ref() }; + assert!(!removed.contains(t)); + let (key, value) = t; + assert_eq!(*value, 2 * key); + i += 1; + } + assert!(i <= n); + + // just for safety: + assert_eq!(map.table.len(), left); + } + } + + #[test] + fn test_const_with_hasher() { + use core::hash::BuildHasher; + use std::collections::hash_map::DefaultHasher; + + #[derive(Clone)] + struct MyHasher; + impl BuildHasher for MyHasher { + type Hasher = DefaultHasher; + + fn build_hasher(&self) -> DefaultHasher { + DefaultHasher::new() + } + } + + const EMPTY_MAP: HashMap<u32, std::string::String, MyHasher> = + HashMap::with_hasher(MyHasher); + + let mut map = EMPTY_MAP; + map.insert(17, "seventeen".to_owned()); + assert_eq!("seventeen", map[&17]); + } + + #[test] + fn test_get_each_mut() { + let mut map = HashMap::new(); + map.insert("foo".to_owned(), 0); + map.insert("bar".to_owned(), 10); + map.insert("baz".to_owned(), 20); + map.insert("qux".to_owned(), 30); + + let xs = map.get_many_mut(["foo", "qux"]); + assert_eq!(xs, Some([&mut 0, &mut 30])); + + let xs = map.get_many_mut(["foo", "dud"]); + assert_eq!(xs, None); + + let xs = map.get_many_mut(["foo", "foo"]); + assert_eq!(xs, None); + + let ys = map.get_many_key_value_mut(["bar", "baz"]); + assert_eq!( + ys, + Some([(&"bar".to_owned(), &mut 10), (&"baz".to_owned(), &mut 20),]), + ); + + let ys = map.get_many_key_value_mut(["bar", "dip"]); + assert_eq!(ys, None); + + let ys = map.get_many_key_value_mut(["baz", "baz"]); + assert_eq!(ys, None); + } + + #[test] + #[should_panic = "panic in drop"] + fn test_clone_from_double_drop() { + #[derive(Clone)] + struct CheckedDrop { + panic_in_drop: bool, + dropped: bool, + } + impl Drop for CheckedDrop { + fn drop(&mut self) { + if self.panic_in_drop { + self.dropped = true; + panic!("panic in drop"); + } + if self.dropped { + panic!("double drop"); + } + self.dropped = true; + } + } + const DISARMED: CheckedDrop = CheckedDrop { + panic_in_drop: false, + dropped: false, + }; + const ARMED: CheckedDrop = CheckedDrop { + panic_in_drop: true, + dropped: false, + }; + + let mut map1 = HashMap::new(); + map1.insert(1, DISARMED); + map1.insert(2, DISARMED); + map1.insert(3, DISARMED); + map1.insert(4, DISARMED); + + let mut map2 = HashMap::new(); + map2.insert(1, DISARMED); + map2.insert(2, ARMED); + map2.insert(3, DISARMED); + map2.insert(4, DISARMED); + + map2.clone_from(&map1); + } + + #[test] + #[should_panic = "panic in clone"] + fn test_clone_from_memory_leaks() { + use ::alloc::vec::Vec; + + struct CheckedClone { + panic_in_clone: bool, + need_drop: Vec<i32>, + } + impl Clone for CheckedClone { + fn clone(&self) -> Self { + if self.panic_in_clone { + panic!("panic in clone") + } + Self { + panic_in_clone: self.panic_in_clone, + need_drop: self.need_drop.clone(), + } + } + } + let mut map1 = HashMap::new(); + map1.insert( + 1, + CheckedClone { + panic_in_clone: false, + need_drop: vec![0, 1, 2], + }, + ); + map1.insert( + 2, + CheckedClone { + panic_in_clone: false, + need_drop: vec![3, 4, 5], + }, + ); + map1.insert( + 3, + CheckedClone { + panic_in_clone: true, + need_drop: vec![6, 7, 8], + }, + ); + let _map2 = map1.clone(); + } + + struct MyAllocInner { + drop_count: Arc<AtomicI8>, + } + + #[derive(Clone)] + struct MyAlloc { + _inner: Arc<MyAllocInner>, + } + + impl MyAlloc { + fn new(drop_count: Arc<AtomicI8>) -> Self { + MyAlloc { + _inner: Arc::new(MyAllocInner { drop_count }), + } + } + } + + impl Drop for MyAllocInner { + fn drop(&mut self) { + println!("MyAlloc freed."); + self.drop_count.fetch_sub(1, Ordering::SeqCst); + } + } + + unsafe impl Allocator for MyAlloc { + fn allocate(&self, layout: Layout) -> std::result::Result<NonNull<[u8]>, AllocError> { + let g = Global; + g.allocate(layout) + } + + unsafe fn deallocate(&self, ptr: NonNull<u8>, layout: Layout) { + let g = Global; + g.deallocate(ptr, layout) + } + } + + #[test] + fn test_hashmap_into_iter_bug() { + let dropped: Arc<AtomicI8> = Arc::new(AtomicI8::new(1)); + + { + let mut map = HashMap::with_capacity_in(10, MyAlloc::new(dropped.clone())); + for i in 0..10 { + map.entry(i).or_insert_with(|| "i".to_string()); + } + + for (k, v) in map { + println!("{}, {}", k, v); + } + } + + // All allocator clones should already be dropped. + assert_eq!(dropped.load(Ordering::SeqCst), 0); + } + + #[derive(Debug)] + struct CheckedCloneDrop<T> { + panic_in_clone: bool, + panic_in_drop: bool, + dropped: bool, + data: T, + } + + impl<T> CheckedCloneDrop<T> { + fn new(panic_in_clone: bool, panic_in_drop: bool, data: T) -> Self { + CheckedCloneDrop { + panic_in_clone, + panic_in_drop, + dropped: false, + data, + } + } + } + + impl<T: Clone> Clone for CheckedCloneDrop<T> { + fn clone(&self) -> Self { + if self.panic_in_clone { + panic!("panic in clone") + } + Self { + panic_in_clone: self.panic_in_clone, + panic_in_drop: self.panic_in_drop, + dropped: self.dropped, + data: self.data.clone(), + } + } + } + + impl<T> Drop for CheckedCloneDrop<T> { + fn drop(&mut self) { + if self.panic_in_drop { + self.dropped = true; + panic!("panic in drop"); + } + if self.dropped { + panic!("double drop"); + } + self.dropped = true; + } + } + + /// Return hashmap with predefined distribution of elements. + /// All elements will be located in the same order as elements + /// returned by iterator. + /// + /// This function does not panic, but returns an error as a `String` + /// to distinguish between a test panic and an error in the input data. + fn get_test_map<I, T, A>( + iter: I, + mut fun: impl FnMut(u64) -> T, + alloc: A, + ) -> Result<HashMap<u64, CheckedCloneDrop<T>, DefaultHashBuilder, A>, String> + where + I: Iterator<Item = (bool, bool)> + Clone + ExactSizeIterator, + A: Allocator, + T: PartialEq + core::fmt::Debug, + { + use crate::scopeguard::guard; + + let mut map: HashMap<u64, CheckedCloneDrop<T>, _, A> = + HashMap::with_capacity_in(iter.size_hint().0, alloc); + { + let mut guard = guard(&mut map, |map| { + for (_, value) in map.iter_mut() { + value.panic_in_drop = false + } + }); + + let mut count = 0; + // Hash and Key must be equal to each other for controlling the elements placement. + for (panic_in_clone, panic_in_drop) in iter.clone() { + if core::mem::needs_drop::<T>() && panic_in_drop { + return Err(String::from( + "panic_in_drop can be set with a type that doesn't need to be dropped", + )); + } + guard.table.insert( + count, + ( + count, + CheckedCloneDrop::new(panic_in_clone, panic_in_drop, fun(count)), + ), + |(k, _)| *k, + ); + count += 1; + } + + // Let's check that all elements are located as we wanted + let mut check_count = 0; + for ((key, value), (panic_in_clone, panic_in_drop)) in guard.iter().zip(iter) { + if *key != check_count { + return Err(format!( + "key != check_count,\nkey: `{}`,\ncheck_count: `{}`", + key, check_count + )); + } + if value.dropped + || value.panic_in_clone != panic_in_clone + || value.panic_in_drop != panic_in_drop + || value.data != fun(check_count) + { + return Err(format!( + "Value is not equal to expected,\nvalue: `{:?}`,\nexpected: \ + `CheckedCloneDrop {{ panic_in_clone: {}, panic_in_drop: {}, dropped: {}, data: {:?} }}`", + value, panic_in_clone, panic_in_drop, false, fun(check_count) + )); + } + check_count += 1; + } + + if guard.len() != check_count as usize { + return Err(format!( + "map.len() != check_count,\nmap.len(): `{}`,\ncheck_count: `{}`", + guard.len(), + check_count + )); + } + + if count != check_count { + return Err(format!( + "count != check_count,\ncount: `{}`,\ncheck_count: `{}`", + count, check_count + )); + } + core::mem::forget(guard); + } + Ok(map) + } + + const DISARMED: bool = false; + const ARMED: bool = true; + + const ARMED_FLAGS: [bool; 8] = [ + DISARMED, DISARMED, DISARMED, ARMED, DISARMED, DISARMED, DISARMED, DISARMED, + ]; + + const DISARMED_FLAGS: [bool; 8] = [ + DISARMED, DISARMED, DISARMED, DISARMED, DISARMED, DISARMED, DISARMED, DISARMED, + ]; + + #[test] + #[should_panic = "panic in clone"] + fn test_clone_memory_leaks_and_double_drop_one() { + let dropped: Arc<AtomicI8> = Arc::new(AtomicI8::new(2)); + + { + assert_eq!(ARMED_FLAGS.len(), DISARMED_FLAGS.len()); + + let map: HashMap<u64, CheckedCloneDrop<Vec<u64>>, DefaultHashBuilder, MyAlloc> = + match get_test_map( + ARMED_FLAGS.into_iter().zip(DISARMED_FLAGS), + |n| vec![n], + MyAlloc::new(dropped.clone()), + ) { + Ok(map) => map, + Err(msg) => panic!("{msg}"), + }; + + // Clone should normally clone a few elements, and then (when the + // clone function panics), deallocate both its own memory, memory + // of `dropped: Arc<AtomicI8>` and the memory of already cloned + // elements (Vec<i32> memory inside CheckedCloneDrop). + let _map2 = map.clone(); + } + } + + #[test] + #[should_panic = "panic in drop"] + fn test_clone_memory_leaks_and_double_drop_two() { + let dropped: Arc<AtomicI8> = Arc::new(AtomicI8::new(2)); + + { + assert_eq!(ARMED_FLAGS.len(), DISARMED_FLAGS.len()); + + let map: HashMap<u64, CheckedCloneDrop<u64>, DefaultHashBuilder, _> = match get_test_map( + DISARMED_FLAGS.into_iter().zip(DISARMED_FLAGS), + |n| n, + MyAlloc::new(dropped.clone()), + ) { + Ok(map) => map, + Err(msg) => panic!("{msg}"), + }; + + let mut map2 = match get_test_map( + DISARMED_FLAGS.into_iter().zip(ARMED_FLAGS), + |n| n, + MyAlloc::new(dropped.clone()), + ) { + Ok(map) => map, + Err(msg) => panic!("{msg}"), + }; + + // The `clone_from` should try to drop the elements of `map2` without + // double drop and leaking the allocator. Elements that have not been + // dropped leak their memory. + map2.clone_from(&map); + } + } + + /// We check that we have a working table if the clone operation from another + /// thread ended in a panic (when buckets of maps are equal to each other). + #[test] + fn test_catch_panic_clone_from_when_len_is_equal() { + use std::thread; + + let dropped: Arc<AtomicI8> = Arc::new(AtomicI8::new(2)); + + { + assert_eq!(ARMED_FLAGS.len(), DISARMED_FLAGS.len()); + + let mut map = match get_test_map( + DISARMED_FLAGS.into_iter().zip(DISARMED_FLAGS), + |n| vec![n], + MyAlloc::new(dropped.clone()), + ) { + Ok(map) => map, + Err(msg) => panic!("{msg}"), + }; + + thread::scope(|s| { + let result: thread::ScopedJoinHandle<'_, String> = s.spawn(|| { + let scope_map = + match get_test_map(ARMED_FLAGS.into_iter().zip(DISARMED_FLAGS), |n| vec![n * 2], MyAlloc::new(dropped.clone())) { + Ok(map) => map, + Err(msg) => return msg, + }; + if map.table.buckets() != scope_map.table.buckets() { + return format!( + "map.table.buckets() != scope_map.table.buckets(),\nleft: `{}`,\nright: `{}`", + map.table.buckets(), scope_map.table.buckets() + ); + } + map.clone_from(&scope_map); + "We must fail the cloning!!!".to_owned() + }); + if let Ok(msg) = result.join() { + panic!("{msg}") + } + }); + + // Let's check that all iterators work fine and do not return elements + // (especially `RawIterRange`, which does not depend on the number of + // elements in the table, but looks directly at the control bytes) + // + // SAFETY: We know for sure that `RawTable` will outlive + // the returned `RawIter / RawIterRange` iterator. + assert_eq!(map.len(), 0); + assert_eq!(map.iter().count(), 0); + assert_eq!(unsafe { map.table.iter().count() }, 0); + assert_eq!(unsafe { map.table.iter().iter.count() }, 0); + + for idx in 0..map.table.buckets() { + let idx = idx as u64; + assert!( + map.table.find(idx, |(k, _)| *k == idx).is_none(), + "Index: {idx}" + ); + } + } + + // All allocator clones should already be dropped. + assert_eq!(dropped.load(Ordering::SeqCst), 0); + } + + /// We check that we have a working table if the clone operation from another + /// thread ended in a panic (when buckets of maps are not equal to each other). + #[test] + fn test_catch_panic_clone_from_when_len_is_not_equal() { + use std::thread; + + let dropped: Arc<AtomicI8> = Arc::new(AtomicI8::new(2)); + + { + assert_eq!(ARMED_FLAGS.len(), DISARMED_FLAGS.len()); + + let mut map = match get_test_map( + [DISARMED].into_iter().zip([DISARMED]), + |n| vec![n], + MyAlloc::new(dropped.clone()), + ) { + Ok(map) => map, + Err(msg) => panic!("{msg}"), + }; + + thread::scope(|s| { + let result: thread::ScopedJoinHandle<'_, String> = s.spawn(|| { + let scope_map = match get_test_map( + ARMED_FLAGS.into_iter().zip(DISARMED_FLAGS), + |n| vec![n * 2], + MyAlloc::new(dropped.clone()), + ) { + Ok(map) => map, + Err(msg) => return msg, + }; + if map.table.buckets() == scope_map.table.buckets() { + return format!( + "map.table.buckets() == scope_map.table.buckets(): `{}`", + map.table.buckets() + ); + } + map.clone_from(&scope_map); + "We must fail the cloning!!!".to_owned() + }); + if let Ok(msg) = result.join() { + panic!("{msg}") + } + }); + + // Let's check that all iterators work fine and do not return elements + // (especially `RawIterRange`, which does not depend on the number of + // elements in the table, but looks directly at the control bytes) + // + // SAFETY: We know for sure that `RawTable` will outlive + // the returned `RawIter / RawIterRange` iterator. + assert_eq!(map.len(), 0); + assert_eq!(map.iter().count(), 0); + assert_eq!(unsafe { map.table.iter().count() }, 0); + assert_eq!(unsafe { map.table.iter().iter.count() }, 0); + + for idx in 0..map.table.buckets() { + let idx = idx as u64; + assert!( + map.table.find(idx, |(k, _)| *k == idx).is_none(), + "Index: {idx}" + ); + } + } + + // All allocator clones should already be dropped. + assert_eq!(dropped.load(Ordering::SeqCst), 0); + } +} diff --git a/vendor/hashbrown/src/raw/alloc.rs b/vendor/hashbrown/src/raw/alloc.rs new file mode 100644 index 0000000..15299e7 --- /dev/null +++ b/vendor/hashbrown/src/raw/alloc.rs @@ -0,0 +1,86 @@ +pub(crate) use self::inner::{do_alloc, Allocator, Global}; + +// Nightly-case. +// Use unstable `allocator_api` feature. +// This is compatible with `allocator-api2` which can be enabled or not. +// This is used when building for `std`. +#[cfg(feature = "nightly")] +mod inner { + use crate::alloc::alloc::Layout; + pub use crate::alloc::alloc::{Allocator, Global}; + use core::ptr::NonNull; + + #[allow(clippy::map_err_ignore)] + pub(crate) fn do_alloc<A: Allocator>(alloc: &A, layout: Layout) -> Result<NonNull<u8>, ()> { + match alloc.allocate(layout) { + Ok(ptr) => Ok(ptr.as_non_null_ptr()), + Err(_) => Err(()), + } + } +} + +// Basic non-nightly case. +// This uses `allocator-api2` enabled by default. +// If any crate enables "nightly" in `allocator-api2`, +// this will be equivalent to the nightly case, +// since `allocator_api2::alloc::Allocator` would be re-export of +// `core::alloc::Allocator`. +#[cfg(all(not(feature = "nightly"), feature = "allocator-api2"))] +mod inner { + use crate::alloc::alloc::Layout; + pub use allocator_api2::alloc::{Allocator, Global}; + use core::ptr::NonNull; + + #[allow(clippy::map_err_ignore)] + pub(crate) fn do_alloc<A: Allocator>(alloc: &A, layout: Layout) -> Result<NonNull<u8>, ()> { + match alloc.allocate(layout) { + Ok(ptr) => Ok(ptr.cast()), + Err(_) => Err(()), + } + } +} + +// No-defaults case. +// When building with default-features turned off and +// neither `nightly` nor `allocator-api2` is enabled, +// this will be used. +// Making it impossible to use any custom allocator with collections defined +// in this crate. +// Any crate in build-tree can enable `allocator-api2`, +// or `nightly` without disturbing users that don't want to use it. +#[cfg(not(any(feature = "nightly", feature = "allocator-api2")))] +mod inner { + use crate::alloc::alloc::{alloc, dealloc, Layout}; + use core::ptr::NonNull; + + #[allow(clippy::missing_safety_doc)] // not exposed outside of this crate + pub unsafe trait Allocator { + fn allocate(&self, layout: Layout) -> Result<NonNull<u8>, ()>; + unsafe fn deallocate(&self, ptr: NonNull<u8>, layout: Layout); + } + + #[derive(Copy, Clone)] + pub struct Global; + + unsafe impl Allocator for Global { + #[inline] + fn allocate(&self, layout: Layout) -> Result<NonNull<u8>, ()> { + unsafe { NonNull::new(alloc(layout)).ok_or(()) } + } + #[inline] + unsafe fn deallocate(&self, ptr: NonNull<u8>, layout: Layout) { + dealloc(ptr.as_ptr(), layout); + } + } + + impl Default for Global { + #[inline] + fn default() -> Self { + Global + } + } + + pub(crate) fn do_alloc<A: Allocator>(alloc: &A, layout: Layout) -> Result<NonNull<u8>, ()> { + alloc.allocate(layout) + } +} diff --git a/vendor/hashbrown/src/raw/bitmask.rs b/vendor/hashbrown/src/raw/bitmask.rs new file mode 100644 index 0000000..6576b3c --- /dev/null +++ b/vendor/hashbrown/src/raw/bitmask.rs @@ -0,0 +1,133 @@ +use super::imp::{ + BitMaskWord, NonZeroBitMaskWord, BITMASK_ITER_MASK, BITMASK_MASK, BITMASK_STRIDE, +}; + +/// A bit mask which contains the result of a `Match` operation on a `Group` and +/// allows iterating through them. +/// +/// The bit mask is arranged so that low-order bits represent lower memory +/// addresses for group match results. +/// +/// For implementation reasons, the bits in the set may be sparsely packed with +/// groups of 8 bits representing one element. If any of these bits are non-zero +/// then this element is considered to true in the mask. If this is the +/// case, `BITMASK_STRIDE` will be 8 to indicate a divide-by-8 should be +/// performed on counts/indices to normalize this difference. `BITMASK_MASK` is +/// similarly a mask of all the actually-used bits. +/// +/// To iterate over a bit mask, it must be converted to a form where only 1 bit +/// is set per element. This is done by applying `BITMASK_ITER_MASK` on the +/// mask bits. +#[derive(Copy, Clone)] +pub(crate) struct BitMask(pub(crate) BitMaskWord); + +#[allow(clippy::use_self)] +impl BitMask { + /// Returns a new `BitMask` with all bits inverted. + #[inline] + #[must_use] + #[allow(dead_code)] + pub(crate) fn invert(self) -> Self { + BitMask(self.0 ^ BITMASK_MASK) + } + + /// Returns a new `BitMask` with the lowest bit removed. + #[inline] + #[must_use] + fn remove_lowest_bit(self) -> Self { + BitMask(self.0 & (self.0 - 1)) + } + + /// Returns whether the `BitMask` has at least one set bit. + #[inline] + pub(crate) fn any_bit_set(self) -> bool { + self.0 != 0 + } + + /// Returns the first set bit in the `BitMask`, if there is one. + #[inline] + pub(crate) fn lowest_set_bit(self) -> Option<usize> { + if let Some(nonzero) = NonZeroBitMaskWord::new(self.0) { + Some(Self::nonzero_trailing_zeros(nonzero)) + } else { + None + } + } + + /// Returns the number of trailing zeroes in the `BitMask`. + #[inline] + pub(crate) fn trailing_zeros(self) -> usize { + // ARM doesn't have a trailing_zeroes instruction, and instead uses + // reverse_bits (RBIT) + leading_zeroes (CLZ). However older ARM + // versions (pre-ARMv7) don't have RBIT and need to emulate it + // instead. Since we only have 1 bit set in each byte on ARM, we can + // use swap_bytes (REV) + leading_zeroes instead. + if cfg!(target_arch = "arm") && BITMASK_STRIDE % 8 == 0 { + self.0.swap_bytes().leading_zeros() as usize / BITMASK_STRIDE + } else { + self.0.trailing_zeros() as usize / BITMASK_STRIDE + } + } + + /// Same as above but takes a `NonZeroBitMaskWord`. + #[inline] + fn nonzero_trailing_zeros(nonzero: NonZeroBitMaskWord) -> usize { + if cfg!(target_arch = "arm") && BITMASK_STRIDE % 8 == 0 { + // SAFETY: A byte-swapped non-zero value is still non-zero. + let swapped = unsafe { NonZeroBitMaskWord::new_unchecked(nonzero.get().swap_bytes()) }; + swapped.leading_zeros() as usize / BITMASK_STRIDE + } else { + nonzero.trailing_zeros() as usize / BITMASK_STRIDE + } + } + + /// Returns the number of leading zeroes in the `BitMask`. + #[inline] + pub(crate) fn leading_zeros(self) -> usize { + self.0.leading_zeros() as usize / BITMASK_STRIDE + } +} + +impl IntoIterator for BitMask { + type Item = usize; + type IntoIter = BitMaskIter; + + #[inline] + fn into_iter(self) -> BitMaskIter { + // A BitMask only requires each element (group of bits) to be non-zero. + // However for iteration we need each element to only contain 1 bit. + BitMaskIter(BitMask(self.0 & BITMASK_ITER_MASK)) + } +} + +/// Iterator over the contents of a `BitMask`, returning the indices of set +/// bits. +#[derive(Copy, Clone)] +pub(crate) struct BitMaskIter(pub(crate) BitMask); + +impl BitMaskIter { + /// Flip the bit in the mask for the entry at the given index. + /// + /// Returns the bit's previous state. + #[inline] + #[allow(clippy::cast_ptr_alignment)] + #[cfg(feature = "raw")] + pub(crate) unsafe fn flip(&mut self, index: usize) -> bool { + // NOTE: The + BITMASK_STRIDE - 1 is to set the high bit. + let mask = 1 << (index * BITMASK_STRIDE + BITMASK_STRIDE - 1); + self.0 .0 ^= mask; + // The bit was set if the bit is now 0. + self.0 .0 & mask == 0 + } +} + +impl Iterator for BitMaskIter { + type Item = usize; + + #[inline] + fn next(&mut self) -> Option<usize> { + let bit = self.0.lowest_set_bit()?; + self.0 = self.0.remove_lowest_bit(); + Some(bit) + } +} diff --git a/vendor/hashbrown/src/raw/generic.rs b/vendor/hashbrown/src/raw/generic.rs new file mode 100644 index 0000000..c668b06 --- /dev/null +++ b/vendor/hashbrown/src/raw/generic.rs @@ -0,0 +1,157 @@ +use super::bitmask::BitMask; +use super::EMPTY; +use core::{mem, ptr}; + +// Use the native word size as the group size. Using a 64-bit group size on +// a 32-bit architecture will just end up being more expensive because +// shifts and multiplies will need to be emulated. + +cfg_if! { + if #[cfg(any( + target_pointer_width = "64", + target_arch = "aarch64", + target_arch = "x86_64", + target_arch = "wasm32", + ))] { + type GroupWord = u64; + type NonZeroGroupWord = core::num::NonZeroU64; + } else { + type GroupWord = u32; + type NonZeroGroupWord = core::num::NonZeroU32; + } +} + +pub(crate) type BitMaskWord = GroupWord; +pub(crate) type NonZeroBitMaskWord = NonZeroGroupWord; +pub(crate) const BITMASK_STRIDE: usize = 8; +// We only care about the highest bit of each byte for the mask. +#[allow(clippy::cast_possible_truncation, clippy::unnecessary_cast)] +pub(crate) const BITMASK_MASK: BitMaskWord = 0x8080_8080_8080_8080_u64 as GroupWord; +pub(crate) const BITMASK_ITER_MASK: BitMaskWord = !0; + +/// Helper function to replicate a byte across a `GroupWord`. +#[inline] +fn repeat(byte: u8) -> GroupWord { + GroupWord::from_ne_bytes([byte; Group::WIDTH]) +} + +/// Abstraction over a group of control bytes which can be scanned in +/// parallel. +/// +/// This implementation uses a word-sized integer. +#[derive(Copy, Clone)] +pub(crate) struct Group(GroupWord); + +// We perform all operations in the native endianness, and convert to +// little-endian just before creating a BitMask. The can potentially +// enable the compiler to eliminate unnecessary byte swaps if we are +// only checking whether a BitMask is empty. +#[allow(clippy::use_self)] +impl Group { + /// Number of bytes in the group. + pub(crate) const WIDTH: usize = mem::size_of::<Self>(); + + /// Returns a full group of empty bytes, suitable for use as the initial + /// value for an empty hash table. + /// + /// This is guaranteed to be aligned to the group size. + #[inline] + pub(crate) const fn static_empty() -> &'static [u8; Group::WIDTH] { + #[repr(C)] + struct AlignedBytes { + _align: [Group; 0], + bytes: [u8; Group::WIDTH], + } + const ALIGNED_BYTES: AlignedBytes = AlignedBytes { + _align: [], + bytes: [EMPTY; Group::WIDTH], + }; + &ALIGNED_BYTES.bytes + } + + /// Loads a group of bytes starting at the given address. + #[inline] + #[allow(clippy::cast_ptr_alignment)] // unaligned load + pub(crate) unsafe fn load(ptr: *const u8) -> Self { + Group(ptr::read_unaligned(ptr.cast())) + } + + /// Loads a group of bytes starting at the given address, which must be + /// aligned to `mem::align_of::<Group>()`. + #[inline] + #[allow(clippy::cast_ptr_alignment)] + pub(crate) unsafe fn load_aligned(ptr: *const u8) -> Self { + // FIXME: use align_offset once it stabilizes + debug_assert_eq!(ptr as usize & (mem::align_of::<Self>() - 1), 0); + Group(ptr::read(ptr.cast())) + } + + /// Stores the group of bytes to the given address, which must be + /// aligned to `mem::align_of::<Group>()`. + #[inline] + #[allow(clippy::cast_ptr_alignment)] + pub(crate) unsafe fn store_aligned(self, ptr: *mut u8) { + // FIXME: use align_offset once it stabilizes + debug_assert_eq!(ptr as usize & (mem::align_of::<Self>() - 1), 0); + ptr::write(ptr.cast(), self.0); + } + + /// Returns a `BitMask` indicating all bytes in the group which *may* + /// have the given value. + /// + /// This function may return a false positive in certain cases where + /// the byte in the group differs from the searched value only in its + /// lowest bit. This is fine because: + /// - This never happens for `EMPTY` and `DELETED`, only full entries. + /// - The check for key equality will catch these. + /// - This only happens if there is at least 1 true match. + /// - The chance of this happening is very low (< 1% chance per byte). + #[inline] + pub(crate) fn match_byte(self, byte: u8) -> BitMask { + // This algorithm is derived from + // https://graphics.stanford.edu/~seander/bithacks.html##ValueInWord + let cmp = self.0 ^ repeat(byte); + BitMask((cmp.wrapping_sub(repeat(0x01)) & !cmp & repeat(0x80)).to_le()) + } + + /// Returns a `BitMask` indicating all bytes in the group which are + /// `EMPTY`. + #[inline] + pub(crate) fn match_empty(self) -> BitMask { + // If the high bit is set, then the byte must be either: + // 1111_1111 (EMPTY) or 1000_0000 (DELETED). + // So we can just check if the top two bits are 1 by ANDing them. + BitMask((self.0 & (self.0 << 1) & repeat(0x80)).to_le()) + } + + /// Returns a `BitMask` indicating all bytes in the group which are + /// `EMPTY` or `DELETED`. + #[inline] + pub(crate) fn match_empty_or_deleted(self) -> BitMask { + // A byte is EMPTY or DELETED iff the high bit is set + BitMask((self.0 & repeat(0x80)).to_le()) + } + + /// Returns a `BitMask` indicating all bytes in the group which are full. + #[inline] + pub(crate) fn match_full(self) -> BitMask { + self.match_empty_or_deleted().invert() + } + + /// Performs the following transformation on all bytes in the group: + /// - `EMPTY => EMPTY` + /// - `DELETED => EMPTY` + /// - `FULL => DELETED` + #[inline] + pub(crate) fn convert_special_to_empty_and_full_to_deleted(self) -> Self { + // Map high_bit = 1 (EMPTY or DELETED) to 1111_1111 + // and high_bit = 0 (FULL) to 1000_0000 + // + // Here's this logic expanded to concrete values: + // let full = 1000_0000 (true) or 0000_0000 (false) + // !1000_0000 + 1 = 0111_1111 + 1 = 1000_0000 (no carry) + // !0000_0000 + 0 = 1111_1111 + 0 = 1111_1111 (no carry) + let full = !self.0 & repeat(0x80); + Group(!full + (full >> 7)) + } +} diff --git a/vendor/hashbrown/src/raw/mod.rs b/vendor/hashbrown/src/raw/mod.rs new file mode 100644 index 0000000..789c3a5 --- /dev/null +++ b/vendor/hashbrown/src/raw/mod.rs @@ -0,0 +1,4478 @@ +use crate::alloc::alloc::{handle_alloc_error, Layout}; +use crate::scopeguard::{guard, ScopeGuard}; +use crate::TryReserveError; +use core::iter::FusedIterator; +use core::marker::PhantomData; +use core::mem; +use core::mem::MaybeUninit; +use core::ptr::NonNull; +use core::{hint, ptr}; + +cfg_if! { + // Use the SSE2 implementation if possible: it allows us to scan 16 buckets + // at once instead of 8. We don't bother with AVX since it would require + // runtime dispatch and wouldn't gain us much anyways: the probability of + // finding a match drops off drastically after the first few buckets. + // + // I attempted an implementation on ARM using NEON instructions, but it + // turns out that most NEON instructions have multi-cycle latency, which in + // the end outweighs any gains over the generic implementation. + if #[cfg(all( + target_feature = "sse2", + any(target_arch = "x86", target_arch = "x86_64"), + not(miri) + ))] { + mod sse2; + use sse2 as imp; + } else if #[cfg(all(target_arch = "aarch64", target_feature = "neon"))] { + mod neon; + use neon as imp; + } else { + mod generic; + use generic as imp; + } +} + +mod alloc; +pub(crate) use self::alloc::{do_alloc, Allocator, Global}; + +mod bitmask; + +use self::bitmask::BitMaskIter; +use self::imp::Group; + +// Branch prediction hint. This is currently only available on nightly but it +// consistently improves performance by 10-15%. +#[cfg(not(feature = "nightly"))] +use core::convert::identity as likely; +#[cfg(not(feature = "nightly"))] +use core::convert::identity as unlikely; +#[cfg(feature = "nightly")] +use core::intrinsics::{likely, unlikely}; + +// Use strict provenance functions if available. +#[cfg(feature = "nightly")] +use core::ptr::invalid_mut; +// Implement it with a cast otherwise. +#[cfg(not(feature = "nightly"))] +#[inline(always)] +fn invalid_mut<T>(addr: usize) -> *mut T { + addr as *mut T +} + +#[inline] +unsafe fn offset_from<T>(to: *const T, from: *const T) -> usize { + to.offset_from(from) as usize +} + +/// Whether memory allocation errors should return an error or abort. +#[derive(Copy, Clone)] +enum Fallibility { + Fallible, + Infallible, +} + +impl Fallibility { + /// Error to return on capacity overflow. + #[cfg_attr(feature = "inline-more", inline)] + fn capacity_overflow(self) -> TryReserveError { + match self { + Fallibility::Fallible => TryReserveError::CapacityOverflow, + Fallibility::Infallible => panic!("Hash table capacity overflow"), + } + } + + /// Error to return on allocation error. + #[cfg_attr(feature = "inline-more", inline)] + fn alloc_err(self, layout: Layout) -> TryReserveError { + match self { + Fallibility::Fallible => TryReserveError::AllocError { layout }, + Fallibility::Infallible => handle_alloc_error(layout), + } + } +} + +trait SizedTypeProperties: Sized { + const IS_ZERO_SIZED: bool = mem::size_of::<Self>() == 0; + const NEEDS_DROP: bool = mem::needs_drop::<Self>(); +} + +impl<T> SizedTypeProperties for T {} + +/// Control byte value for an empty bucket. +const EMPTY: u8 = 0b1111_1111; + +/// Control byte value for a deleted bucket. +const DELETED: u8 = 0b1000_0000; + +/// Checks whether a control byte represents a full bucket (top bit is clear). +#[inline] +fn is_full(ctrl: u8) -> bool { + ctrl & 0x80 == 0 +} + +/// Checks whether a control byte represents a special value (top bit is set). +#[inline] +fn is_special(ctrl: u8) -> bool { + ctrl & 0x80 != 0 +} + +/// Checks whether a special control value is EMPTY (just check 1 bit). +#[inline] +fn special_is_empty(ctrl: u8) -> bool { + debug_assert!(is_special(ctrl)); + ctrl & 0x01 != 0 +} + +/// Primary hash function, used to select the initial bucket to probe from. +#[inline] +#[allow(clippy::cast_possible_truncation)] +fn h1(hash: u64) -> usize { + // On 32-bit platforms we simply ignore the higher hash bits. + hash as usize +} + +// Constant for h2 function that grabing the top 7 bits of the hash. +const MIN_HASH_LEN: usize = if mem::size_of::<usize>() < mem::size_of::<u64>() { + mem::size_of::<usize>() +} else { + mem::size_of::<u64>() +}; + +/// Secondary hash function, saved in the low 7 bits of the control byte. +#[inline] +#[allow(clippy::cast_possible_truncation)] +fn h2(hash: u64) -> u8 { + // Grab the top 7 bits of the hash. While the hash is normally a full 64-bit + // value, some hash functions (such as FxHash) produce a usize result + // instead, which means that the top 32 bits are 0 on 32-bit platforms. + // So we use MIN_HASH_LEN constant to handle this. + let top7 = hash >> (MIN_HASH_LEN * 8 - 7); + (top7 & 0x7f) as u8 // truncation +} + +/// Probe sequence based on triangular numbers, which is guaranteed (since our +/// table size is a power of two) to visit every group of elements exactly once. +/// +/// A triangular probe has us jump by 1 more group every time. So first we +/// jump by 1 group (meaning we just continue our linear scan), then 2 groups +/// (skipping over 1 group), then 3 groups (skipping over 2 groups), and so on. +/// +/// Proof that the probe will visit every group in the table: +/// <https://fgiesen.wordpress.com/2015/02/22/triangular-numbers-mod-2n/> +struct ProbeSeq { + pos: usize, + stride: usize, +} + +impl ProbeSeq { + #[inline] + fn move_next(&mut self, bucket_mask: usize) { + // We should have found an empty bucket by now and ended the probe. + debug_assert!( + self.stride <= bucket_mask, + "Went past end of probe sequence" + ); + + self.stride += Group::WIDTH; + self.pos += self.stride; + self.pos &= bucket_mask; + } +} + +/// Returns the number of buckets needed to hold the given number of items, +/// taking the maximum load factor into account. +/// +/// Returns `None` if an overflow occurs. +// Workaround for emscripten bug emscripten-core/emscripten-fastcomp#258 +#[cfg_attr(target_os = "emscripten", inline(never))] +#[cfg_attr(not(target_os = "emscripten"), inline)] +fn capacity_to_buckets(cap: usize) -> Option<usize> { + debug_assert_ne!(cap, 0); + + // For small tables we require at least 1 empty bucket so that lookups are + // guaranteed to terminate if an element doesn't exist in the table. + if cap < 8 { + // We don't bother with a table size of 2 buckets since that can only + // hold a single element. Instead we skip directly to a 4 bucket table + // which can hold 3 elements. + return Some(if cap < 4 { 4 } else { 8 }); + } + + // Otherwise require 1/8 buckets to be empty (87.5% load) + // + // Be careful when modifying this, calculate_layout relies on the + // overflow check here. + let adjusted_cap = cap.checked_mul(8)? / 7; + + // Any overflows will have been caught by the checked_mul. Also, any + // rounding errors from the division above will be cleaned up by + // next_power_of_two (which can't overflow because of the previous division). + Some(adjusted_cap.next_power_of_two()) +} + +/// Returns the maximum effective capacity for the given bucket mask, taking +/// the maximum load factor into account. +#[inline] +fn bucket_mask_to_capacity(bucket_mask: usize) -> usize { + if bucket_mask < 8 { + // For tables with 1/2/4/8 buckets, we always reserve one empty slot. + // Keep in mind that the bucket mask is one less than the bucket count. + bucket_mask + } else { + // For larger tables we reserve 12.5% of the slots as empty. + ((bucket_mask + 1) / 8) * 7 + } +} + +/// Helper which allows the max calculation for ctrl_align to be statically computed for each T +/// while keeping the rest of `calculate_layout_for` independent of `T` +#[derive(Copy, Clone)] +struct TableLayout { + size: usize, + ctrl_align: usize, +} + +impl TableLayout { + #[inline] + const fn new<T>() -> Self { + let layout = Layout::new::<T>(); + Self { + size: layout.size(), + ctrl_align: if layout.align() > Group::WIDTH { + layout.align() + } else { + Group::WIDTH + }, + } + } + + #[inline] + fn calculate_layout_for(self, buckets: usize) -> Option<(Layout, usize)> { + debug_assert!(buckets.is_power_of_two()); + + let TableLayout { size, ctrl_align } = self; + // Manual layout calculation since Layout methods are not yet stable. + let ctrl_offset = + size.checked_mul(buckets)?.checked_add(ctrl_align - 1)? & !(ctrl_align - 1); + let len = ctrl_offset.checked_add(buckets + Group::WIDTH)?; + + // We need an additional check to ensure that the allocation doesn't + // exceed `isize::MAX` (https://github.com/rust-lang/rust/pull/95295). + if len > isize::MAX as usize - (ctrl_align - 1) { + return None; + } + + Some(( + unsafe { Layout::from_size_align_unchecked(len, ctrl_align) }, + ctrl_offset, + )) + } +} + +/// A reference to an empty bucket into which an can be inserted. +pub struct InsertSlot { + index: usize, +} + +/// A reference to a hash table bucket containing a `T`. +/// +/// This is usually just a pointer to the element itself. However if the element +/// is a ZST, then we instead track the index of the element in the table so +/// that `erase` works properly. +pub struct Bucket<T> { + // Actually it is pointer to next element than element itself + // this is needed to maintain pointer arithmetic invariants + // keeping direct pointer to element introduces difficulty. + // Using `NonNull` for variance and niche layout + ptr: NonNull<T>, +} + +// This Send impl is needed for rayon support. This is safe since Bucket is +// never exposed in a public API. +unsafe impl<T> Send for Bucket<T> {} + +impl<T> Clone for Bucket<T> { + #[inline] + fn clone(&self) -> Self { + Self { ptr: self.ptr } + } +} + +impl<T> Bucket<T> { + /// Creates a [`Bucket`] that contain pointer to the data. + /// The pointer calculation is performed by calculating the + /// offset from given `base` pointer (convenience for + /// `base.as_ptr().sub(index)`). + /// + /// `index` is in units of `T`; e.g., an `index` of 3 represents a pointer + /// offset of `3 * size_of::<T>()` bytes. + /// + /// If the `T` is a ZST, then we instead track the index of the element + /// in the table so that `erase` works properly (return + /// `NonNull::new_unchecked((index + 1) as *mut T)`) + /// + /// # Safety + /// + /// If `mem::size_of::<T>() != 0`, then the safety rules are directly derived + /// from the safety rules for [`<*mut T>::sub`] method of `*mut T` and the safety + /// rules of [`NonNull::new_unchecked`] function. + /// + /// Thus, in order to uphold the safety contracts for the [`<*mut T>::sub`] method + /// and [`NonNull::new_unchecked`] function, as well as for the correct + /// logic of the work of this crate, the following rules are necessary and + /// sufficient: + /// + /// * the `base` pointer must not be `dangling` and must points to the + /// end of the first `value element` from the `data part` of the table, i.e. + /// must be the pointer that returned by [`RawTable::data_end`] or by + /// [`RawTableInner::data_end<T>`]; + /// + /// * `index` must not be greater than `RawTableInner.bucket_mask`, i.e. + /// `index <= RawTableInner.bucket_mask` or, in other words, `(index + 1)` + /// must be no greater than the number returned by the function + /// [`RawTable::buckets`] or [`RawTableInner::buckets`]. + /// + /// If `mem::size_of::<T>() == 0`, then the only requirement is that the + /// `index` must not be greater than `RawTableInner.bucket_mask`, i.e. + /// `index <= RawTableInner.bucket_mask` or, in other words, `(index + 1)` + /// must be no greater than the number returned by the function + /// [`RawTable::buckets`] or [`RawTableInner::buckets`]. + /// + /// [`Bucket`]: crate::raw::Bucket + /// [`<*mut T>::sub`]: https://doc.rust-lang.org/core/primitive.pointer.html#method.sub-1 + /// [`NonNull::new_unchecked`]: https://doc.rust-lang.org/stable/std/ptr/struct.NonNull.html#method.new_unchecked + /// [`RawTable::data_end`]: crate::raw::RawTable::data_end + /// [`RawTableInner::data_end<T>`]: RawTableInner::data_end<T> + /// [`RawTable::buckets`]: crate::raw::RawTable::buckets + /// [`RawTableInner::buckets`]: RawTableInner::buckets + #[inline] + unsafe fn from_base_index(base: NonNull<T>, index: usize) -> Self { + // If mem::size_of::<T>() != 0 then return a pointer to an `element` in + // the data part of the table (we start counting from "0", so that + // in the expression T[last], the "last" index actually one less than the + // "buckets" number in the table, i.e. "last = RawTableInner.bucket_mask"): + // + // `from_base_index(base, 1).as_ptr()` returns a pointer that + // points here in the data part of the table + // (to the start of T1) + // | + // | `base: NonNull<T>` must point here + // | (to the end of T0 or to the start of C0) + // v v + // [Padding], Tlast, ..., |T1|, T0, |C0, C1, ..., Clast + // ^ + // `from_base_index(base, 1)` returns a pointer + // that points here in the data part of the table + // (to the end of T1) + // + // where: T0...Tlast - our stored data; C0...Clast - control bytes + // or metadata for data. + let ptr = if T::IS_ZERO_SIZED { + // won't overflow because index must be less than length (bucket_mask) + // and bucket_mask is guaranteed to be less than `isize::MAX` + // (see TableLayout::calculate_layout_for method) + invalid_mut(index + 1) + } else { + base.as_ptr().sub(index) + }; + Self { + ptr: NonNull::new_unchecked(ptr), + } + } + + /// Calculates the index of a [`Bucket`] as distance between two pointers + /// (convenience for `base.as_ptr().offset_from(self.ptr.as_ptr()) as usize`). + /// The returned value is in units of T: the distance in bytes divided by + /// [`core::mem::size_of::<T>()`]. + /// + /// If the `T` is a ZST, then we return the index of the element in + /// the table so that `erase` works properly (return `self.ptr.as_ptr() as usize - 1`). + /// + /// This function is the inverse of [`from_base_index`]. + /// + /// # Safety + /// + /// If `mem::size_of::<T>() != 0`, then the safety rules are directly derived + /// from the safety rules for [`<*const T>::offset_from`] method of `*const T`. + /// + /// Thus, in order to uphold the safety contracts for [`<*const T>::offset_from`] + /// method, as well as for the correct logic of the work of this crate, the + /// following rules are necessary and sufficient: + /// + /// * `base` contained pointer must not be `dangling` and must point to the + /// end of the first `element` from the `data part` of the table, i.e. + /// must be a pointer that returns by [`RawTable::data_end`] or by + /// [`RawTableInner::data_end<T>`]; + /// + /// * `self` also must not contain dangling pointer; + /// + /// * both `self` and `base` must be created from the same [`RawTable`] + /// (or [`RawTableInner`]). + /// + /// If `mem::size_of::<T>() == 0`, this function is always safe. + /// + /// [`Bucket`]: crate::raw::Bucket + /// [`from_base_index`]: crate::raw::Bucket::from_base_index + /// [`RawTable::data_end`]: crate::raw::RawTable::data_end + /// [`RawTableInner::data_end<T>`]: RawTableInner::data_end<T> + /// [`RawTable`]: crate::raw::RawTable + /// [`RawTableInner`]: RawTableInner + /// [`<*const T>::offset_from`]: https://doc.rust-lang.org/nightly/core/primitive.pointer.html#method.offset_from + #[inline] + unsafe fn to_base_index(&self, base: NonNull<T>) -> usize { + // If mem::size_of::<T>() != 0 then return an index under which we used to store the + // `element` in the data part of the table (we start counting from "0", so + // that in the expression T[last], the "last" index actually is one less than the + // "buckets" number in the table, i.e. "last = RawTableInner.bucket_mask"). + // For example for 5th element in table calculation is performed like this: + // + // mem::size_of::<T>() + // | + // | `self = from_base_index(base, 5)` that returns pointer + // | that points here in tha data part of the table + // | (to the end of T5) + // | | `base: NonNull<T>` must point here + // v | (to the end of T0 or to the start of C0) + // /???\ v v + // [Padding], Tlast, ..., |T10|, ..., T5|, T4, T3, T2, T1, T0, |C0, C1, C2, C3, C4, C5, ..., C10, ..., Clast + // \__________ __________/ + // \/ + // `bucket.to_base_index(base)` = 5 + // (base.as_ptr() as usize - self.ptr.as_ptr() as usize) / mem::size_of::<T>() + // + // where: T0...Tlast - our stored data; C0...Clast - control bytes or metadata for data. + if T::IS_ZERO_SIZED { + // this can not be UB + self.ptr.as_ptr() as usize - 1 + } else { + offset_from(base.as_ptr(), self.ptr.as_ptr()) + } + } + + /// Acquires the underlying raw pointer `*mut T` to `data`. + /// + /// # Note + /// + /// If `T` is not [`Copy`], do not use `*mut T` methods that can cause calling the + /// destructor of `T` (for example the [`<*mut T>::drop_in_place`] method), because + /// for properly dropping the data we also need to clear `data` control bytes. If we + /// drop data, but do not clear `data control byte` it leads to double drop when + /// [`RawTable`] goes out of scope. + /// + /// If you modify an already initialized `value`, so [`Hash`] and [`Eq`] on the new + /// `T` value and its borrowed form *must* match those for the old `T` value, as the map + /// will not re-evaluate where the new value should go, meaning the value may become + /// "lost" if their location does not reflect their state. + /// + /// [`RawTable`]: crate::raw::RawTable + /// [`<*mut T>::drop_in_place`]: https://doc.rust-lang.org/core/primitive.pointer.html#method.drop_in_place + /// [`Hash`]: https://doc.rust-lang.org/core/hash/trait.Hash.html + /// [`Eq`]: https://doc.rust-lang.org/core/cmp/trait.Eq.html + /// + /// # Examples + /// + /// ``` + /// # #[cfg(feature = "raw")] + /// # fn test() { + /// use core::hash::{BuildHasher, Hash}; + /// use hashbrown::raw::{Bucket, RawTable}; + /// + /// type NewHashBuilder = core::hash::BuildHasherDefault<ahash::AHasher>; + /// + /// fn make_hash<K: Hash + ?Sized, S: BuildHasher>(hash_builder: &S, key: &K) -> u64 { + /// use core::hash::Hasher; + /// let mut state = hash_builder.build_hasher(); + /// key.hash(&mut state); + /// state.finish() + /// } + /// + /// let hash_builder = NewHashBuilder::default(); + /// let mut table = RawTable::new(); + /// + /// let value = ("a", 100); + /// let hash = make_hash(&hash_builder, &value.0); + /// + /// table.insert(hash, value.clone(), |val| make_hash(&hash_builder, &val.0)); + /// + /// let bucket: Bucket<(&str, i32)> = table.find(hash, |(k1, _)| k1 == &value.0).unwrap(); + /// + /// assert_eq!(unsafe { &*bucket.as_ptr() }, &("a", 100)); + /// # } + /// # fn main() { + /// # #[cfg(feature = "raw")] + /// # test() + /// # } + /// ``` + #[inline] + pub fn as_ptr(&self) -> *mut T { + if T::IS_ZERO_SIZED { + // Just return an arbitrary ZST pointer which is properly aligned + // invalid pointer is good enough for ZST + invalid_mut(mem::align_of::<T>()) + } else { + unsafe { self.ptr.as_ptr().sub(1) } + } + } + + /// Create a new [`Bucket`] that is offset from the `self` by the given + /// `offset`. The pointer calculation is performed by calculating the + /// offset from `self` pointer (convenience for `self.ptr.as_ptr().sub(offset)`). + /// This function is used for iterators. + /// + /// `offset` is in units of `T`; e.g., a `offset` of 3 represents a pointer + /// offset of `3 * size_of::<T>()` bytes. + /// + /// # Safety + /// + /// If `mem::size_of::<T>() != 0`, then the safety rules are directly derived + /// from the safety rules for [`<*mut T>::sub`] method of `*mut T` and safety + /// rules of [`NonNull::new_unchecked`] function. + /// + /// Thus, in order to uphold the safety contracts for [`<*mut T>::sub`] method + /// and [`NonNull::new_unchecked`] function, as well as for the correct + /// logic of the work of this crate, the following rules are necessary and + /// sufficient: + /// + /// * `self` contained pointer must not be `dangling`; + /// + /// * `self.to_base_index() + ofset` must not be greater than `RawTableInner.bucket_mask`, + /// i.e. `(self.to_base_index() + ofset) <= RawTableInner.bucket_mask` or, in other + /// words, `self.to_base_index() + ofset + 1` must be no greater than the number returned + /// by the function [`RawTable::buckets`] or [`RawTableInner::buckets`]. + /// + /// If `mem::size_of::<T>() == 0`, then the only requirement is that the + /// `self.to_base_index() + ofset` must not be greater than `RawTableInner.bucket_mask`, + /// i.e. `(self.to_base_index() + ofset) <= RawTableInner.bucket_mask` or, in other words, + /// `self.to_base_index() + ofset + 1` must be no greater than the number returned by the + /// function [`RawTable::buckets`] or [`RawTableInner::buckets`]. + /// + /// [`Bucket`]: crate::raw::Bucket + /// [`<*mut T>::sub`]: https://doc.rust-lang.org/core/primitive.pointer.html#method.sub-1 + /// [`NonNull::new_unchecked`]: https://doc.rust-lang.org/stable/std/ptr/struct.NonNull.html#method.new_unchecked + /// [`RawTable::buckets`]: crate::raw::RawTable::buckets + /// [`RawTableInner::buckets`]: RawTableInner::buckets + #[inline] + unsafe fn next_n(&self, offset: usize) -> Self { + let ptr = if T::IS_ZERO_SIZED { + // invalid pointer is good enough for ZST + invalid_mut(self.ptr.as_ptr() as usize + offset) + } else { + self.ptr.as_ptr().sub(offset) + }; + Self { + ptr: NonNull::new_unchecked(ptr), + } + } + + /// Executes the destructor (if any) of the pointed-to `data`. + /// + /// # Safety + /// + /// See [`ptr::drop_in_place`] for safety concerns. + /// + /// You should use [`RawTable::erase`] instead of this function, + /// or be careful with calling this function directly, because for + /// properly dropping the data we need also clear `data` control bytes. + /// If we drop data, but do not erase `data control byte` it leads to + /// double drop when [`RawTable`] goes out of scope. + /// + /// [`ptr::drop_in_place`]: https://doc.rust-lang.org/core/ptr/fn.drop_in_place.html + /// [`RawTable`]: crate::raw::RawTable + /// [`RawTable::erase`]: crate::raw::RawTable::erase + #[cfg_attr(feature = "inline-more", inline)] + pub(crate) unsafe fn drop(&self) { + self.as_ptr().drop_in_place(); + } + + /// Reads the `value` from `self` without moving it. This leaves the + /// memory in `self` unchanged. + /// + /// # Safety + /// + /// See [`ptr::read`] for safety concerns. + /// + /// You should use [`RawTable::remove`] instead of this function, + /// or be careful with calling this function directly, because compiler + /// calls its destructor when readed `value` goes out of scope. It + /// can cause double dropping when [`RawTable`] goes out of scope, + /// because of not erased `data control byte`. + /// + /// [`ptr::read`]: https://doc.rust-lang.org/core/ptr/fn.read.html + /// [`RawTable`]: crate::raw::RawTable + /// [`RawTable::remove`]: crate::raw::RawTable::remove + #[inline] + pub(crate) unsafe fn read(&self) -> T { + self.as_ptr().read() + } + + /// Overwrites a memory location with the given `value` without reading + /// or dropping the old value (like [`ptr::write`] function). + /// + /// # Safety + /// + /// See [`ptr::write`] for safety concerns. + /// + /// # Note + /// + /// [`Hash`] and [`Eq`] on the new `T` value and its borrowed form *must* match + /// those for the old `T` value, as the map will not re-evaluate where the new + /// value should go, meaning the value may become "lost" if their location + /// does not reflect their state. + /// + /// [`ptr::write`]: https://doc.rust-lang.org/core/ptr/fn.write.html + /// [`Hash`]: https://doc.rust-lang.org/core/hash/trait.Hash.html + /// [`Eq`]: https://doc.rust-lang.org/core/cmp/trait.Eq.html + #[inline] + pub(crate) unsafe fn write(&self, val: T) { + self.as_ptr().write(val); + } + + /// Returns a shared immutable reference to the `value`. + /// + /// # Safety + /// + /// See [`NonNull::as_ref`] for safety concerns. + /// + /// [`NonNull::as_ref`]: https://doc.rust-lang.org/core/ptr/struct.NonNull.html#method.as_ref + /// + /// # Examples + /// + /// ``` + /// # #[cfg(feature = "raw")] + /// # fn test() { + /// use core::hash::{BuildHasher, Hash}; + /// use hashbrown::raw::{Bucket, RawTable}; + /// + /// type NewHashBuilder = core::hash::BuildHasherDefault<ahash::AHasher>; + /// + /// fn make_hash<K: Hash + ?Sized, S: BuildHasher>(hash_builder: &S, key: &K) -> u64 { + /// use core::hash::Hasher; + /// let mut state = hash_builder.build_hasher(); + /// key.hash(&mut state); + /// state.finish() + /// } + /// + /// let hash_builder = NewHashBuilder::default(); + /// let mut table = RawTable::new(); + /// + /// let value: (&str, String) = ("A pony", "is a small horse".to_owned()); + /// let hash = make_hash(&hash_builder, &value.0); + /// + /// table.insert(hash, value.clone(), |val| make_hash(&hash_builder, &val.0)); + /// + /// let bucket: Bucket<(&str, String)> = table.find(hash, |(k, _)| k == &value.0).unwrap(); + /// + /// assert_eq!( + /// unsafe { bucket.as_ref() }, + /// &("A pony", "is a small horse".to_owned()) + /// ); + /// # } + /// # fn main() { + /// # #[cfg(feature = "raw")] + /// # test() + /// # } + /// ``` + #[inline] + pub unsafe fn as_ref<'a>(&self) -> &'a T { + &*self.as_ptr() + } + + /// Returns a unique mutable reference to the `value`. + /// + /// # Safety + /// + /// See [`NonNull::as_mut`] for safety concerns. + /// + /// # Note + /// + /// [`Hash`] and [`Eq`] on the new `T` value and its borrowed form *must* match + /// those for the old `T` value, as the map will not re-evaluate where the new + /// value should go, meaning the value may become "lost" if their location + /// does not reflect their state. + /// + /// [`NonNull::as_mut`]: https://doc.rust-lang.org/core/ptr/struct.NonNull.html#method.as_mut + /// [`Hash`]: https://doc.rust-lang.org/core/hash/trait.Hash.html + /// [`Eq`]: https://doc.rust-lang.org/core/cmp/trait.Eq.html + /// + /// # Examples + /// + /// ``` + /// # #[cfg(feature = "raw")] + /// # fn test() { + /// use core::hash::{BuildHasher, Hash}; + /// use hashbrown::raw::{Bucket, RawTable}; + /// + /// type NewHashBuilder = core::hash::BuildHasherDefault<ahash::AHasher>; + /// + /// fn make_hash<K: Hash + ?Sized, S: BuildHasher>(hash_builder: &S, key: &K) -> u64 { + /// use core::hash::Hasher; + /// let mut state = hash_builder.build_hasher(); + /// key.hash(&mut state); + /// state.finish() + /// } + /// + /// let hash_builder = NewHashBuilder::default(); + /// let mut table = RawTable::new(); + /// + /// let value: (&str, String) = ("A pony", "is a small horse".to_owned()); + /// let hash = make_hash(&hash_builder, &value.0); + /// + /// table.insert(hash, value.clone(), |val| make_hash(&hash_builder, &val.0)); + /// + /// let bucket: Bucket<(&str, String)> = table.find(hash, |(k, _)| k == &value.0).unwrap(); + /// + /// unsafe { + /// bucket + /// .as_mut() + /// .1 + /// .push_str(" less than 147 cm at the withers") + /// }; + /// assert_eq!( + /// unsafe { bucket.as_ref() }, + /// &( + /// "A pony", + /// "is a small horse less than 147 cm at the withers".to_owned() + /// ) + /// ); + /// # } + /// # fn main() { + /// # #[cfg(feature = "raw")] + /// # test() + /// # } + /// ``` + #[inline] + pub unsafe fn as_mut<'a>(&self) -> &'a mut T { + &mut *self.as_ptr() + } + + /// Copies `size_of<T>` bytes from `other` to `self`. The source + /// and destination may *not* overlap. + /// + /// # Safety + /// + /// See [`ptr::copy_nonoverlapping`] for safety concerns. + /// + /// Like [`read`], `copy_nonoverlapping` creates a bitwise copy of `T`, regardless of + /// whether `T` is [`Copy`]. If `T` is not [`Copy`], using *both* the values + /// in the region beginning at `*self` and the region beginning at `*other` can + /// [violate memory safety]. + /// + /// # Note + /// + /// [`Hash`] and [`Eq`] on the new `T` value and its borrowed form *must* match + /// those for the old `T` value, as the map will not re-evaluate where the new + /// value should go, meaning the value may become "lost" if their location + /// does not reflect their state. + /// + /// [`ptr::copy_nonoverlapping`]: https://doc.rust-lang.org/core/ptr/fn.copy_nonoverlapping.html + /// [`read`]: https://doc.rust-lang.org/core/ptr/fn.read.html + /// [violate memory safety]: https://doc.rust-lang.org/std/ptr/fn.read.html#ownership-of-the-returned-value + /// [`Hash`]: https://doc.rust-lang.org/core/hash/trait.Hash.html + /// [`Eq`]: https://doc.rust-lang.org/core/cmp/trait.Eq.html + #[cfg(feature = "raw")] + #[inline] + pub unsafe fn copy_from_nonoverlapping(&self, other: &Self) { + self.as_ptr().copy_from_nonoverlapping(other.as_ptr(), 1); + } +} + +/// A raw hash table with an unsafe API. +pub struct RawTable<T, A: Allocator = Global> { + table: RawTableInner, + alloc: A, + // Tell dropck that we own instances of T. + marker: PhantomData<T>, +} + +/// Non-generic part of `RawTable` which allows functions to be instantiated only once regardless +/// of how many different key-value types are used. +struct RawTableInner { + // Mask to get an index from a hash value. The value is one less than the + // number of buckets in the table. + bucket_mask: usize, + + // [Padding], T1, T2, ..., Tlast, C1, C2, ... + // ^ points here + ctrl: NonNull<u8>, + + // Number of elements that can be inserted before we need to grow the table + growth_left: usize, + + // Number of elements in the table, only really used by len() + items: usize, +} + +impl<T> RawTable<T, Global> { + /// Creates a new empty hash table without allocating any memory. + /// + /// In effect this returns a table with exactly 1 bucket. However we can + /// leave the data pointer dangling since that bucket is never written to + /// due to our load factor forcing us to always have at least 1 free bucket. + #[inline] + pub const fn new() -> Self { + Self { + table: RawTableInner::NEW, + alloc: Global, + marker: PhantomData, + } + } + + /// Attempts to allocate a new hash table with at least enough capacity + /// for inserting the given number of elements without reallocating. + #[cfg(feature = "raw")] + pub fn try_with_capacity(capacity: usize) -> Result<Self, TryReserveError> { + Self::try_with_capacity_in(capacity, Global) + } + + /// Allocates a new hash table with at least enough capacity for inserting + /// the given number of elements without reallocating. + pub fn with_capacity(capacity: usize) -> Self { + Self::with_capacity_in(capacity, Global) + } +} + +impl<T, A: Allocator> RawTable<T, A> { + const TABLE_LAYOUT: TableLayout = TableLayout::new::<T>(); + + /// Creates a new empty hash table without allocating any memory, using the + /// given allocator. + /// + /// In effect this returns a table with exactly 1 bucket. However we can + /// leave the data pointer dangling since that bucket is never written to + /// due to our load factor forcing us to always have at least 1 free bucket. + #[inline] + pub const fn new_in(alloc: A) -> Self { + Self { + table: RawTableInner::NEW, + alloc, + marker: PhantomData, + } + } + + /// Allocates a new hash table with the given number of buckets. + /// + /// The control bytes are left uninitialized. + #[cfg_attr(feature = "inline-more", inline)] + unsafe fn new_uninitialized( + alloc: A, + buckets: usize, + fallibility: Fallibility, + ) -> Result<Self, TryReserveError> { + debug_assert!(buckets.is_power_of_two()); + + Ok(Self { + table: RawTableInner::new_uninitialized( + &alloc, + Self::TABLE_LAYOUT, + buckets, + fallibility, + )?, + alloc, + marker: PhantomData, + }) + } + + /// Attempts to allocate a new hash table using the given allocator, with at least enough + /// capacity for inserting the given number of elements without reallocating. + #[cfg(feature = "raw")] + pub fn try_with_capacity_in(capacity: usize, alloc: A) -> Result<Self, TryReserveError> { + Ok(Self { + table: RawTableInner::fallible_with_capacity( + &alloc, + Self::TABLE_LAYOUT, + capacity, + Fallibility::Fallible, + )?, + alloc, + marker: PhantomData, + }) + } + + /// Allocates a new hash table using the given allocator, with at least enough capacity for + /// inserting the given number of elements without reallocating. + pub fn with_capacity_in(capacity: usize, alloc: A) -> Self { + Self { + table: RawTableInner::with_capacity(&alloc, Self::TABLE_LAYOUT, capacity), + alloc, + marker: PhantomData, + } + } + + /// Returns a reference to the underlying allocator. + #[inline] + pub fn allocator(&self) -> &A { + &self.alloc + } + + /// Returns pointer to one past last element of data table. + #[inline] + pub unsafe fn data_end(&self) -> NonNull<T> { + NonNull::new_unchecked(self.table.ctrl.as_ptr().cast()) + } + + /// Returns pointer to start of data table. + #[inline] + #[cfg(any(feature = "raw", feature = "nightly"))] + pub unsafe fn data_start(&self) -> NonNull<T> { + NonNull::new_unchecked(self.data_end().as_ptr().wrapping_sub(self.buckets())) + } + + /// Return the information about memory allocated by the table. + /// + /// `RawTable` allocates single memory block to store both data and metadata. + /// This function returns allocation size and alignment and the beginning of the area. + /// These are the arguments which will be passed to `dealloc` when the table is dropped. + /// + /// This function might be useful for memory profiling. + #[inline] + #[cfg(feature = "raw")] + pub fn allocation_info(&self) -> (NonNull<u8>, Layout) { + // SAFETY: We use the same `table_layout` that was used to allocate + // this table. + unsafe { self.table.allocation_info_or_zero(Self::TABLE_LAYOUT) } + } + + /// Returns the index of a bucket from a `Bucket`. + #[inline] + pub unsafe fn bucket_index(&self, bucket: &Bucket<T>) -> usize { + bucket.to_base_index(self.data_end()) + } + + /// Returns a pointer to an element in the table. + #[inline] + pub unsafe fn bucket(&self, index: usize) -> Bucket<T> { + debug_assert_ne!(self.table.bucket_mask, 0); + debug_assert!(index < self.buckets()); + Bucket::from_base_index(self.data_end(), index) + } + + /// Erases an element from the table without dropping it. + #[cfg_attr(feature = "inline-more", inline)] + unsafe fn erase_no_drop(&mut self, item: &Bucket<T>) { + let index = self.bucket_index(item); + self.table.erase(index); + } + + /// Erases an element from the table, dropping it in place. + #[cfg_attr(feature = "inline-more", inline)] + #[allow(clippy::needless_pass_by_value)] + pub unsafe fn erase(&mut self, item: Bucket<T>) { + // Erase the element from the table first since drop might panic. + self.erase_no_drop(&item); + item.drop(); + } + + /// Finds and erases an element from the table, dropping it in place. + /// Returns true if an element was found. + #[cfg(feature = "raw")] + #[cfg_attr(feature = "inline-more", inline)] + pub fn erase_entry(&mut self, hash: u64, eq: impl FnMut(&T) -> bool) -> bool { + // Avoid `Option::map` because it bloats LLVM IR. + if let Some(bucket) = self.find(hash, eq) { + unsafe { + self.erase(bucket); + } + true + } else { + false + } + } + + /// Removes an element from the table, returning it. + /// + /// This also returns an `InsertSlot` pointing to the newly free bucket. + #[cfg_attr(feature = "inline-more", inline)] + #[allow(clippy::needless_pass_by_value)] + pub unsafe fn remove(&mut self, item: Bucket<T>) -> (T, InsertSlot) { + self.erase_no_drop(&item); + ( + item.read(), + InsertSlot { + index: self.bucket_index(&item), + }, + ) + } + + /// Finds and removes an element from the table, returning it. + #[cfg_attr(feature = "inline-more", inline)] + pub fn remove_entry(&mut self, hash: u64, eq: impl FnMut(&T) -> bool) -> Option<T> { + // Avoid `Option::map` because it bloats LLVM IR. + match self.find(hash, eq) { + Some(bucket) => Some(unsafe { self.remove(bucket).0 }), + None => None, + } + } + + /// Marks all table buckets as empty without dropping their contents. + #[cfg_attr(feature = "inline-more", inline)] + pub fn clear_no_drop(&mut self) { + self.table.clear_no_drop(); + } + + /// Removes all elements from the table without freeing the backing memory. + #[cfg_attr(feature = "inline-more", inline)] + pub fn clear(&mut self) { + if self.is_empty() { + // Special case empty table to avoid surprising O(capacity) time. + return; + } + // Ensure that the table is reset even if one of the drops panic + let mut self_ = guard(self, |self_| self_.clear_no_drop()); + unsafe { + // SAFETY: ScopeGuard sets to zero the `items` field of the table + // even in case of panic during the dropping of the elements so + // that there will be no double drop of the elements. + self_.table.drop_elements::<T>(); + } + } + + /// Shrinks the table to fit `max(self.len(), min_size)` elements. + #[cfg_attr(feature = "inline-more", inline)] + pub fn shrink_to(&mut self, min_size: usize, hasher: impl Fn(&T) -> u64) { + // Calculate the minimal number of elements that we need to reserve + // space for. + let min_size = usize::max(self.table.items, min_size); + if min_size == 0 { + let mut old_inner = mem::replace(&mut self.table, RawTableInner::NEW); + unsafe { + // SAFETY: + // 1. We call the function only once; + // 2. We know for sure that `alloc` and `table_layout` matches the [`Allocator`] + // and [`TableLayout`] that were used to allocate this table. + // 3. If any elements' drop function panics, then there will only be a memory leak, + // because we have replaced the inner table with a new one. + old_inner.drop_inner_table::<T, _>(&self.alloc, Self::TABLE_LAYOUT); + } + return; + } + + // Calculate the number of buckets that we need for this number of + // elements. If the calculation overflows then the requested bucket + // count must be larger than what we have right and nothing needs to be + // done. + let min_buckets = match capacity_to_buckets(min_size) { + Some(buckets) => buckets, + None => return, + }; + + // If we have more buckets than we need, shrink the table. + if min_buckets < self.buckets() { + // Fast path if the table is empty + if self.table.items == 0 { + let new_inner = + RawTableInner::with_capacity(&self.alloc, Self::TABLE_LAYOUT, min_size); + let mut old_inner = mem::replace(&mut self.table, new_inner); + unsafe { + // SAFETY: + // 1. We call the function only once; + // 2. We know for sure that `alloc` and `table_layout` matches the [`Allocator`] + // and [`TableLayout`] that were used to allocate this table. + // 3. If any elements' drop function panics, then there will only be a memory leak, + // because we have replaced the inner table with a new one. + old_inner.drop_inner_table::<T, _>(&self.alloc, Self::TABLE_LAYOUT); + } + } else { + // Avoid `Result::unwrap_or_else` because it bloats LLVM IR. + unsafe { + // SAFETY: + // 1. We know for sure that `min_size >= self.table.items`. + // 2. The [`RawTableInner`] must already have properly initialized control bytes since + // we will never expose RawTable::new_uninitialized in a public API. + if self + .resize(min_size, hasher, Fallibility::Infallible) + .is_err() + { + // SAFETY: The result of calling the `resize` function cannot be an error + // because `fallibility == Fallibility::Infallible. + hint::unreachable_unchecked() + } + } + } + } + } + + /// Ensures that at least `additional` items can be inserted into the table + /// without reallocation. + #[cfg_attr(feature = "inline-more", inline)] + pub fn reserve(&mut self, additional: usize, hasher: impl Fn(&T) -> u64) { + if unlikely(additional > self.table.growth_left) { + // Avoid `Result::unwrap_or_else` because it bloats LLVM IR. + unsafe { + // SAFETY: The [`RawTableInner`] must already have properly initialized control + // bytes since we will never expose RawTable::new_uninitialized in a public API. + if self + .reserve_rehash(additional, hasher, Fallibility::Infallible) + .is_err() + { + // SAFETY: All allocation errors will be caught inside `RawTableInner::reserve_rehash`. + hint::unreachable_unchecked() + } + } + } + } + + /// Tries to ensure that at least `additional` items can be inserted into + /// the table without reallocation. + #[cfg_attr(feature = "inline-more", inline)] + pub fn try_reserve( + &mut self, + additional: usize, + hasher: impl Fn(&T) -> u64, + ) -> Result<(), TryReserveError> { + if additional > self.table.growth_left { + // SAFETY: The [`RawTableInner`] must already have properly initialized control + // bytes since we will never expose RawTable::new_uninitialized in a public API. + unsafe { self.reserve_rehash(additional, hasher, Fallibility::Fallible) } + } else { + Ok(()) + } + } + + /// Out-of-line slow path for `reserve` and `try_reserve`. + /// + /// # Safety + /// + /// The [`RawTableInner`] must have properly initialized control bytes, + /// otherwise calling this function results in [`undefined behavior`] + /// + /// [`undefined behavior`]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html + #[cold] + #[inline(never)] + unsafe fn reserve_rehash( + &mut self, + additional: usize, + hasher: impl Fn(&T) -> u64, + fallibility: Fallibility, + ) -> Result<(), TryReserveError> { + unsafe { + // SAFETY: + // 1. We know for sure that `alloc` and `layout` matches the [`Allocator`] and + // [`TableLayout`] that were used to allocate this table. + // 2. The `drop` function is the actual drop function of the elements stored in + // the table. + // 3. The caller ensures that the control bytes of the `RawTableInner` + // are already initialized. + self.table.reserve_rehash_inner( + &self.alloc, + additional, + &|table, index| hasher(table.bucket::<T>(index).as_ref()), + fallibility, + Self::TABLE_LAYOUT, + if T::NEEDS_DROP { + Some(mem::transmute(ptr::drop_in_place::<T> as unsafe fn(*mut T))) + } else { + None + }, + ) + } + } + + /// Allocates a new table of a different size and moves the contents of the + /// current table into it. + /// + /// # Safety + /// + /// The [`RawTableInner`] must have properly initialized control bytes, + /// otherwise calling this function results in [`undefined behavior`] + /// + /// The caller of this function must ensure that `capacity >= self.table.items` + /// otherwise: + /// + /// * If `self.table.items != 0`, calling of this function with `capacity` + /// equal to 0 (`capacity == 0`) results in [`undefined behavior`]. + /// + /// * If `capacity_to_buckets(capacity) < Group::WIDTH` and + /// `self.table.items > capacity_to_buckets(capacity)` + /// calling this function results in [`undefined behavior`]. + /// + /// * If `capacity_to_buckets(capacity) >= Group::WIDTH` and + /// `self.table.items > capacity_to_buckets(capacity)` + /// calling this function are never return (will go into an + /// infinite loop). + /// + /// See [`RawTableInner::find_insert_slot`] for more information. + /// + /// [`RawTableInner::find_insert_slot`]: RawTableInner::find_insert_slot + /// [`undefined behavior`]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html + unsafe fn resize( + &mut self, + capacity: usize, + hasher: impl Fn(&T) -> u64, + fallibility: Fallibility, + ) -> Result<(), TryReserveError> { + // SAFETY: + // 1. The caller of this function guarantees that `capacity >= self.table.items`. + // 2. We know for sure that `alloc` and `layout` matches the [`Allocator`] and + // [`TableLayout`] that were used to allocate this table. + // 3. The caller ensures that the control bytes of the `RawTableInner` + // are already initialized. + self.table.resize_inner( + &self.alloc, + capacity, + &|table, index| hasher(table.bucket::<T>(index).as_ref()), + fallibility, + Self::TABLE_LAYOUT, + ) + } + + /// Inserts a new element into the table, and returns its raw bucket. + /// + /// This does not check if the given element already exists in the table. + #[cfg_attr(feature = "inline-more", inline)] + pub fn insert(&mut self, hash: u64, value: T, hasher: impl Fn(&T) -> u64) -> Bucket<T> { + unsafe { + // SAFETY: + // 1. The [`RawTableInner`] must already have properly initialized control bytes since + // we will never expose `RawTable::new_uninitialized` in a public API. + // + // 2. We reserve additional space (if necessary) right after calling this function. + let mut slot = self.table.find_insert_slot(hash); + + // We can avoid growing the table once we have reached our load factor if we are replacing + // a tombstone. This works since the number of EMPTY slots does not change in this case. + // + // SAFETY: The function is guaranteed to return [`InsertSlot`] that contains an index + // in the range `0..=self.buckets()`. + let old_ctrl = *self.table.ctrl(slot.index); + if unlikely(self.table.growth_left == 0 && special_is_empty(old_ctrl)) { + self.reserve(1, hasher); + // SAFETY: We know for sure that `RawTableInner` has control bytes + // initialized and that there is extra space in the table. + slot = self.table.find_insert_slot(hash); + } + + self.insert_in_slot(hash, slot, value) + } + } + + /// Attempts to insert a new element without growing the table and return its raw bucket. + /// + /// Returns an `Err` containing the given element if inserting it would require growing the + /// table. + /// + /// This does not check if the given element already exists in the table. + #[cfg(feature = "raw")] + #[cfg_attr(feature = "inline-more", inline)] + pub fn try_insert_no_grow(&mut self, hash: u64, value: T) -> Result<Bucket<T>, T> { + unsafe { + match self.table.prepare_insert_no_grow(hash) { + Ok(index) => { + let bucket = self.bucket(index); + bucket.write(value); + Ok(bucket) + } + Err(()) => Err(value), + } + } + } + + /// Inserts a new element into the table, and returns a mutable reference to it. + /// + /// This does not check if the given element already exists in the table. + #[cfg_attr(feature = "inline-more", inline)] + pub fn insert_entry(&mut self, hash: u64, value: T, hasher: impl Fn(&T) -> u64) -> &mut T { + unsafe { self.insert(hash, value, hasher).as_mut() } + } + + /// Inserts a new element into the table, without growing the table. + /// + /// There must be enough space in the table to insert the new element. + /// + /// This does not check if the given element already exists in the table. + #[cfg_attr(feature = "inline-more", inline)] + #[cfg(any(feature = "raw", feature = "rustc-internal-api"))] + pub unsafe fn insert_no_grow(&mut self, hash: u64, value: T) -> Bucket<T> { + let (index, old_ctrl) = self.table.prepare_insert_slot(hash); + let bucket = self.table.bucket(index); + + // If we are replacing a DELETED entry then we don't need to update + // the load counter. + self.table.growth_left -= special_is_empty(old_ctrl) as usize; + + bucket.write(value); + self.table.items += 1; + bucket + } + + /// Temporary removes a bucket, applying the given function to the removed + /// element and optionally put back the returned value in the same bucket. + /// + /// Returns `true` if the bucket still contains an element + /// + /// This does not check if the given bucket is actually occupied. + #[cfg_attr(feature = "inline-more", inline)] + pub unsafe fn replace_bucket_with<F>(&mut self, bucket: Bucket<T>, f: F) -> bool + where + F: FnOnce(T) -> Option<T>, + { + let index = self.bucket_index(&bucket); + let old_ctrl = *self.table.ctrl(index); + debug_assert!(self.is_bucket_full(index)); + let old_growth_left = self.table.growth_left; + let item = self.remove(bucket).0; + if let Some(new_item) = f(item) { + self.table.growth_left = old_growth_left; + self.table.set_ctrl(index, old_ctrl); + self.table.items += 1; + self.bucket(index).write(new_item); + true + } else { + false + } + } + + /// Searches for an element in the table. If the element is not found, + /// returns `Err` with the position of a slot where an element with the + /// same hash could be inserted. + /// + /// This function may resize the table if additional space is required for + /// inserting an element. + #[inline] + pub fn find_or_find_insert_slot( + &mut self, + hash: u64, + mut eq: impl FnMut(&T) -> bool, + hasher: impl Fn(&T) -> u64, + ) -> Result<Bucket<T>, InsertSlot> { + self.reserve(1, hasher); + + unsafe { + // SAFETY: + // 1. We know for sure that there is at least one empty `bucket` in the table. + // 2. The [`RawTableInner`] must already have properly initialized control bytes since we will + // never expose `RawTable::new_uninitialized` in a public API. + // 3. The `find_or_find_insert_slot_inner` function returns the `index` of only the full bucket, + // which is in the range `0..self.buckets()` (since there is at least one empty `bucket` in + // the table), so calling `self.bucket(index)` and `Bucket::as_ref` is safe. + match self + .table + .find_or_find_insert_slot_inner(hash, &mut |index| eq(self.bucket(index).as_ref())) + { + // SAFETY: See explanation above. + Ok(index) => Ok(self.bucket(index)), + Err(slot) => Err(slot), + } + } + } + + /// Inserts a new element into the table in the given slot, and returns its + /// raw bucket. + /// + /// # Safety + /// + /// `slot` must point to a slot previously returned by + /// `find_or_find_insert_slot`, and no mutation of the table must have + /// occurred since that call. + #[inline] + pub unsafe fn insert_in_slot(&mut self, hash: u64, slot: InsertSlot, value: T) -> Bucket<T> { + let old_ctrl = *self.table.ctrl(slot.index); + self.table.record_item_insert_at(slot.index, old_ctrl, hash); + + let bucket = self.bucket(slot.index); + bucket.write(value); + bucket + } + + /// Searches for an element in the table. + #[inline] + pub fn find(&self, hash: u64, mut eq: impl FnMut(&T) -> bool) -> Option<Bucket<T>> { + unsafe { + // SAFETY: + // 1. The [`RawTableInner`] must already have properly initialized control bytes since we + // will never expose `RawTable::new_uninitialized` in a public API. + // 1. The `find_inner` function returns the `index` of only the full bucket, which is in + // the range `0..self.buckets()`, so calling `self.bucket(index)` and `Bucket::as_ref` + // is safe. + let result = self + .table + .find_inner(hash, &mut |index| eq(self.bucket(index).as_ref())); + + // Avoid `Option::map` because it bloats LLVM IR. + match result { + // SAFETY: See explanation above. + Some(index) => Some(self.bucket(index)), + None => None, + } + } + } + + /// Gets a reference to an element in the table. + #[inline] + pub fn get(&self, hash: u64, eq: impl FnMut(&T) -> bool) -> Option<&T> { + // Avoid `Option::map` because it bloats LLVM IR. + match self.find(hash, eq) { + Some(bucket) => Some(unsafe { bucket.as_ref() }), + None => None, + } + } + + /// Gets a mutable reference to an element in the table. + #[inline] + pub fn get_mut(&mut self, hash: u64, eq: impl FnMut(&T) -> bool) -> Option<&mut T> { + // Avoid `Option::map` because it bloats LLVM IR. + match self.find(hash, eq) { + Some(bucket) => Some(unsafe { bucket.as_mut() }), + None => None, + } + } + + /// Attempts to get mutable references to `N` entries in the table at once. + /// + /// Returns an array of length `N` with the results of each query. + /// + /// At most one mutable reference will be returned to any entry. `None` will be returned if any + /// of the hashes are duplicates. `None` will be returned if the hash is not found. + /// + /// The `eq` argument should be a closure such that `eq(i, k)` returns true if `k` is equal to + /// the `i`th key to be looked up. + pub fn get_many_mut<const N: usize>( + &mut self, + hashes: [u64; N], + eq: impl FnMut(usize, &T) -> bool, + ) -> Option<[&'_ mut T; N]> { + unsafe { + let ptrs = self.get_many_mut_pointers(hashes, eq)?; + + for (i, &cur) in ptrs.iter().enumerate() { + if ptrs[..i].iter().any(|&prev| ptr::eq::<T>(prev, cur)) { + return None; + } + } + // All bucket are distinct from all previous buckets so we're clear to return the result + // of the lookup. + + // TODO use `MaybeUninit::array_assume_init` here instead once that's stable. + Some(mem::transmute_copy(&ptrs)) + } + } + + pub unsafe fn get_many_unchecked_mut<const N: usize>( + &mut self, + hashes: [u64; N], + eq: impl FnMut(usize, &T) -> bool, + ) -> Option<[&'_ mut T; N]> { + let ptrs = self.get_many_mut_pointers(hashes, eq)?; + Some(mem::transmute_copy(&ptrs)) + } + + unsafe fn get_many_mut_pointers<const N: usize>( + &mut self, + hashes: [u64; N], + mut eq: impl FnMut(usize, &T) -> bool, + ) -> Option<[*mut T; N]> { + // TODO use `MaybeUninit::uninit_array` here instead once that's stable. + let mut outs: MaybeUninit<[*mut T; N]> = MaybeUninit::uninit(); + let outs_ptr = outs.as_mut_ptr(); + + for (i, &hash) in hashes.iter().enumerate() { + let cur = self.find(hash, |k| eq(i, k))?; + *(*outs_ptr).get_unchecked_mut(i) = cur.as_mut(); + } + + // TODO use `MaybeUninit::array_assume_init` here instead once that's stable. + Some(outs.assume_init()) + } + + /// Returns the number of elements the map can hold without reallocating. + /// + /// This number is a lower bound; the table might be able to hold + /// more, but is guaranteed to be able to hold at least this many. + #[inline] + pub fn capacity(&self) -> usize { + self.table.items + self.table.growth_left + } + + /// Returns the number of elements in the table. + #[inline] + pub fn len(&self) -> usize { + self.table.items + } + + /// Returns `true` if the table contains no elements. + #[inline] + pub fn is_empty(&self) -> bool { + self.len() == 0 + } + + /// Returns the number of buckets in the table. + #[inline] + pub fn buckets(&self) -> usize { + self.table.bucket_mask + 1 + } + + /// Checks whether the bucket at `index` is full. + /// + /// # Safety + /// + /// The caller must ensure `index` is less than the number of buckets. + #[inline] + pub unsafe fn is_bucket_full(&self, index: usize) -> bool { + self.table.is_bucket_full(index) + } + + /// Returns an iterator over every element in the table. It is up to + /// the caller to ensure that the `RawTable` outlives the `RawIter`. + /// Because we cannot make the `next` method unsafe on the `RawIter` + /// struct, we have to make the `iter` method unsafe. + #[inline] + pub unsafe fn iter(&self) -> RawIter<T> { + // SAFETY: + // 1. The caller must uphold the safety contract for `iter` method. + // 2. The [`RawTableInner`] must already have properly initialized control bytes since + // we will never expose RawTable::new_uninitialized in a public API. + self.table.iter() + } + + /// Returns an iterator over occupied buckets that could match a given hash. + /// + /// `RawTable` only stores 7 bits of the hash value, so this iterator may + /// return items that have a hash value different than the one provided. You + /// should always validate the returned values before using them. + /// + /// It is up to the caller to ensure that the `RawTable` outlives the + /// `RawIterHash`. Because we cannot make the `next` method unsafe on the + /// `RawIterHash` struct, we have to make the `iter_hash` method unsafe. + #[cfg_attr(feature = "inline-more", inline)] + #[cfg(feature = "raw")] + pub unsafe fn iter_hash(&self, hash: u64) -> RawIterHash<T> { + RawIterHash::new(self, hash) + } + + /// Returns an iterator which removes all elements from the table without + /// freeing the memory. + #[cfg_attr(feature = "inline-more", inline)] + pub fn drain(&mut self) -> RawDrain<'_, T, A> { + unsafe { + let iter = self.iter(); + self.drain_iter_from(iter) + } + } + + /// Returns an iterator which removes all elements from the table without + /// freeing the memory. + /// + /// Iteration starts at the provided iterator's current location. + /// + /// It is up to the caller to ensure that the iterator is valid for this + /// `RawTable` and covers all items that remain in the table. + #[cfg_attr(feature = "inline-more", inline)] + pub unsafe fn drain_iter_from(&mut self, iter: RawIter<T>) -> RawDrain<'_, T, A> { + debug_assert_eq!(iter.len(), self.len()); + RawDrain { + iter, + table: mem::replace(&mut self.table, RawTableInner::NEW), + orig_table: NonNull::from(&mut self.table), + marker: PhantomData, + } + } + + /// Returns an iterator which consumes all elements from the table. + /// + /// Iteration starts at the provided iterator's current location. + /// + /// It is up to the caller to ensure that the iterator is valid for this + /// `RawTable` and covers all items that remain in the table. + pub unsafe fn into_iter_from(self, iter: RawIter<T>) -> RawIntoIter<T, A> { + debug_assert_eq!(iter.len(), self.len()); + + let allocation = self.into_allocation(); + RawIntoIter { + iter, + allocation, + marker: PhantomData, + } + } + + /// Converts the table into a raw allocation. The contents of the table + /// should be dropped using a `RawIter` before freeing the allocation. + #[cfg_attr(feature = "inline-more", inline)] + pub(crate) fn into_allocation(self) -> Option<(NonNull<u8>, Layout, A)> { + let alloc = if self.table.is_empty_singleton() { + None + } else { + // Avoid `Option::unwrap_or_else` because it bloats LLVM IR. + let (layout, ctrl_offset) = + match Self::TABLE_LAYOUT.calculate_layout_for(self.table.buckets()) { + Some(lco) => lco, + None => unsafe { hint::unreachable_unchecked() }, + }; + Some(( + unsafe { NonNull::new_unchecked(self.table.ctrl.as_ptr().sub(ctrl_offset)) }, + layout, + unsafe { ptr::read(&self.alloc) }, + )) + }; + mem::forget(self); + alloc + } +} + +unsafe impl<T, A: Allocator> Send for RawTable<T, A> +where + T: Send, + A: Send, +{ +} +unsafe impl<T, A: Allocator> Sync for RawTable<T, A> +where + T: Sync, + A: Sync, +{ +} + +impl RawTableInner { + const NEW: Self = RawTableInner::new(); + + /// Creates a new empty hash table without allocating any memory. + /// + /// In effect this returns a table with exactly 1 bucket. However we can + /// leave the data pointer dangling since that bucket is never accessed + /// due to our load factor forcing us to always have at least 1 free bucket. + #[inline] + const fn new() -> Self { + Self { + // Be careful to cast the entire slice to a raw pointer. + ctrl: unsafe { NonNull::new_unchecked(Group::static_empty() as *const _ as *mut u8) }, + bucket_mask: 0, + items: 0, + growth_left: 0, + } + } +} + +impl RawTableInner { + /// Allocates a new [`RawTableInner`] with the given number of buckets. + /// The control bytes and buckets are left uninitialized. + /// + /// # Safety + /// + /// The caller of this function must ensure that the `buckets` is power of two + /// and also initialize all control bytes of the length `self.bucket_mask + 1 + + /// Group::WIDTH` with the [`EMPTY`] bytes. + /// + /// See also [`Allocator`] API for other safety concerns. + /// + /// [`Allocator`]: https://doc.rust-lang.org/alloc/alloc/trait.Allocator.html + #[cfg_attr(feature = "inline-more", inline)] + unsafe fn new_uninitialized<A>( + alloc: &A, + table_layout: TableLayout, + buckets: usize, + fallibility: Fallibility, + ) -> Result<Self, TryReserveError> + where + A: Allocator, + { + debug_assert!(buckets.is_power_of_two()); + + // Avoid `Option::ok_or_else` because it bloats LLVM IR. + let (layout, ctrl_offset) = match table_layout.calculate_layout_for(buckets) { + Some(lco) => lco, + None => return Err(fallibility.capacity_overflow()), + }; + + let ptr: NonNull<u8> = match do_alloc(alloc, layout) { + Ok(block) => block.cast(), + Err(_) => return Err(fallibility.alloc_err(layout)), + }; + + // SAFETY: null pointer will be caught in above check + let ctrl = NonNull::new_unchecked(ptr.as_ptr().add(ctrl_offset)); + Ok(Self { + ctrl, + bucket_mask: buckets - 1, + items: 0, + growth_left: bucket_mask_to_capacity(buckets - 1), + }) + } + + /// Attempts to allocate a new [`RawTableInner`] with at least enough + /// capacity for inserting the given number of elements without reallocating. + /// + /// All the control bytes are initialized with the [`EMPTY`] bytes. + #[inline] + fn fallible_with_capacity<A>( + alloc: &A, + table_layout: TableLayout, + capacity: usize, + fallibility: Fallibility, + ) -> Result<Self, TryReserveError> + where + A: Allocator, + { + if capacity == 0 { + Ok(Self::NEW) + } else { + // SAFETY: We checked that we could successfully allocate the new table, and then + // initialized all control bytes with the constant `EMPTY` byte. + unsafe { + let buckets = + capacity_to_buckets(capacity).ok_or_else(|| fallibility.capacity_overflow())?; + + let result = Self::new_uninitialized(alloc, table_layout, buckets, fallibility)?; + // SAFETY: We checked that the table is allocated and therefore the table already has + // `self.bucket_mask + 1 + Group::WIDTH` number of control bytes (see TableLayout::calculate_layout_for) + // so writing `self.num_ctrl_bytes() == bucket_mask + 1 + Group::WIDTH` bytes is safe. + result.ctrl(0).write_bytes(EMPTY, result.num_ctrl_bytes()); + + Ok(result) + } + } + } + + /// Allocates a new [`RawTableInner`] with at least enough capacity for inserting + /// the given number of elements without reallocating. + /// + /// Panics if the new capacity exceeds [`isize::MAX`] bytes and [`abort`] the program + /// in case of allocation error. Use [`fallible_with_capacity`] instead if you want to + /// handle memory allocation failure. + /// + /// All the control bytes are initialized with the [`EMPTY`] bytes. + /// + /// [`fallible_with_capacity`]: RawTableInner::fallible_with_capacity + /// [`abort`]: https://doc.rust-lang.org/alloc/alloc/fn.handle_alloc_error.html + fn with_capacity<A>(alloc: &A, table_layout: TableLayout, capacity: usize) -> Self + where + A: Allocator, + { + // Avoid `Result::unwrap_or_else` because it bloats LLVM IR. + match Self::fallible_with_capacity(alloc, table_layout, capacity, Fallibility::Infallible) { + Ok(table_inner) => table_inner, + // SAFETY: All allocation errors will be caught inside `RawTableInner::new_uninitialized`. + Err(_) => unsafe { hint::unreachable_unchecked() }, + } + } + + /// Fixes up an insertion slot returned by the [`RawTableInner::find_insert_slot_in_group`] method. + /// + /// In tables smaller than the group width (`self.buckets() < Group::WIDTH`), trailing control + /// bytes outside the range of the table are filled with [`EMPTY`] entries. These will unfortunately + /// trigger a match of [`RawTableInner::find_insert_slot_in_group`] function. This is because + /// the `Some(bit)` returned by `group.match_empty_or_deleted().lowest_set_bit()` after masking + /// (`(probe_seq.pos + bit) & self.bucket_mask`) may point to a full bucket that is already occupied. + /// We detect this situation here and perform a second scan starting at the beginning of the table. + /// This second scan is guaranteed to find an empty slot (due to the load factor) before hitting the + /// trailing control bytes (containing [`EMPTY`] bytes). + /// + /// If this function is called correctly, it is guaranteed to return [`InsertSlot`] with an + /// index of an empty or deleted bucket in the range `0..self.buckets()` (see `Warning` and + /// `Safety`). + /// + /// # Warning + /// + /// The table must have at least 1 empty or deleted `bucket`, otherwise if the table is less than + /// the group width (`self.buckets() < Group::WIDTH`) this function returns an index outside of the + /// table indices range `0..self.buckets()` (`0..=self.bucket_mask`). Attempt to write data at that + /// index will cause immediate [`undefined behavior`]. + /// + /// # Safety + /// + /// The safety rules are directly derived from the safety rules for [`RawTableInner::ctrl`] method. + /// Thus, in order to uphold those safety contracts, as well as for the correct logic of the work + /// of this crate, the following rules are necessary and sufficient: + /// + /// * The [`RawTableInner`] must have properly initialized control bytes otherwise calling this + /// function results in [`undefined behavior`]. + /// + /// * This function must only be used on insertion slots found by [`RawTableInner::find_insert_slot_in_group`] + /// (after the `find_insert_slot_in_group` function, but before insertion into the table). + /// + /// * The `index` must not be greater than the `self.bucket_mask`, i.e. `(index + 1) <= self.buckets()` + /// (this one is provided by the [`RawTableInner::find_insert_slot_in_group`] function). + /// + /// Calling this function with an index not provided by [`RawTableInner::find_insert_slot_in_group`] + /// may result in [`undefined behavior`] even if the index satisfies the safety rules of the + /// [`RawTableInner::ctrl`] function (`index < self.bucket_mask + 1 + Group::WIDTH`). + /// + /// [`RawTableInner::ctrl`]: RawTableInner::ctrl + /// [`RawTableInner::find_insert_slot_in_group`]: RawTableInner::find_insert_slot_in_group + /// [`undefined behavior`]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html + #[inline] + unsafe fn fix_insert_slot(&self, mut index: usize) -> InsertSlot { + // SAFETY: The caller of this function ensures that `index` is in the range `0..=self.bucket_mask`. + if unlikely(self.is_bucket_full(index)) { + debug_assert!(self.bucket_mask < Group::WIDTH); + // SAFETY: + // + // * Since the caller of this function ensures that the control bytes are properly + // initialized and `ptr = self.ctrl(0)` points to the start of the array of control + // bytes, therefore: `ctrl` is valid for reads, properly aligned to `Group::WIDTH` + // and points to the properly initialized control bytes (see also + // `TableLayout::calculate_layout_for` and `ptr::read`); + // + // * Because the caller of this function ensures that the index was provided by the + // `self.find_insert_slot_in_group()` function, so for for tables larger than the + // group width (self.buckets() >= Group::WIDTH), we will never end up in the given + // branch, since `(probe_seq.pos + bit) & self.bucket_mask` in `find_insert_slot_in_group` + // cannot return a full bucket index. For tables smaller than the group width, calling + // the `unwrap_unchecked` function is also safe, as the trailing control bytes outside + // the range of the table are filled with EMPTY bytes (and we know for sure that there + // is at least one FULL bucket), so this second scan either finds an empty slot (due to + // the load factor) or hits the trailing control bytes (containing EMPTY). + index = Group::load_aligned(self.ctrl(0)) + .match_empty_or_deleted() + .lowest_set_bit() + .unwrap_unchecked(); + } + InsertSlot { index } + } + + /// Finds the position to insert something in a group. + /// + /// **This may have false positives and must be fixed up with `fix_insert_slot` + /// before it's used.** + /// + /// The function is guaranteed to return the index of an empty or deleted [`Bucket`] + /// in the range `0..self.buckets()` (`0..=self.bucket_mask`). + #[inline] + fn find_insert_slot_in_group(&self, group: &Group, probe_seq: &ProbeSeq) -> Option<usize> { + let bit = group.match_empty_or_deleted().lowest_set_bit(); + + if likely(bit.is_some()) { + // This is the same as `(probe_seq.pos + bit) % self.buckets()` because the number + // of buckets is a power of two, and `self.bucket_mask = self.buckets() - 1`. + Some((probe_seq.pos + bit.unwrap()) & self.bucket_mask) + } else { + None + } + } + + /// Searches for an element in the table, or a potential slot where that element could + /// be inserted (an empty or deleted [`Bucket`] index). + /// + /// This uses dynamic dispatch to reduce the amount of code generated, but that is + /// eliminated by LLVM optimizations. + /// + /// This function does not make any changes to the `data` part of the table, or any + /// changes to the `items` or `growth_left` field of the table. + /// + /// The table must have at least 1 empty or deleted `bucket`, otherwise, if the + /// `eq: &mut dyn FnMut(usize) -> bool` function does not return `true`, this function + /// will never return (will go into an infinite loop) for tables larger than the group + /// width, or return an index outside of the table indices range if the table is less + /// than the group width. + /// + /// This function is guaranteed to provide the `eq: &mut dyn FnMut(usize) -> bool` + /// function with only `FULL` buckets' indices and return the `index` of the found + /// element (as `Ok(index)`). If the element is not found and there is at least 1 + /// empty or deleted [`Bucket`] in the table, the function is guaranteed to return + /// [InsertSlot] with an index in the range `0..self.buckets()`, but in any case, + /// if this function returns [`InsertSlot`], it will contain an index in the range + /// `0..=self.buckets()`. + /// + /// # Safety + /// + /// The [`RawTableInner`] must have properly initialized control bytes otherwise calling + /// this function results in [`undefined behavior`]. + /// + /// Attempt to write data at the [`InsertSlot`] returned by this function when the table is + /// less than the group width and if there was not at least one empty or deleted bucket in + /// the table will cause immediate [`undefined behavior`]. This is because in this case the + /// function will return `self.bucket_mask + 1` as an index due to the trailing [`EMPTY] + /// control bytes outside the table range. + /// + /// [`undefined behavior`]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html + #[inline] + unsafe fn find_or_find_insert_slot_inner( + &self, + hash: u64, + eq: &mut dyn FnMut(usize) -> bool, + ) -> Result<usize, InsertSlot> { + let mut insert_slot = None; + + let h2_hash = h2(hash); + let mut probe_seq = self.probe_seq(hash); + + loop { + // SAFETY: + // * Caller of this function ensures that the control bytes are properly initialized. + // + // * `ProbeSeq.pos` cannot be greater than `self.bucket_mask = self.buckets() - 1` + // of the table due to masking with `self.bucket_mask` and also because mumber of + // buckets is a power of two (see `self.probe_seq` function). + // + // * Even if `ProbeSeq.pos` returns `position == self.bucket_mask`, it is safe to + // call `Group::load` due to the extended control bytes range, which is + // `self.bucket_mask + 1 + Group::WIDTH` (in fact, this means that the last control + // byte will never be read for the allocated table); + // + // * Also, even if `RawTableInner` is not already allocated, `ProbeSeq.pos` will + // always return "0" (zero), so Group::load will read unaligned `Group::static_empty()` + // bytes, which is safe (see RawTableInner::new). + let group = unsafe { Group::load(self.ctrl(probe_seq.pos)) }; + + for bit in group.match_byte(h2_hash) { + let index = (probe_seq.pos + bit) & self.bucket_mask; + + if likely(eq(index)) { + return Ok(index); + } + } + + // We didn't find the element we were looking for in the group, try to get an + // insertion slot from the group if we don't have one yet. + if likely(insert_slot.is_none()) { + insert_slot = self.find_insert_slot_in_group(&group, &probe_seq); + } + + // Only stop the search if the group contains at least one empty element. + // Otherwise, the element that we are looking for might be in a following group. + if likely(group.match_empty().any_bit_set()) { + // We must have found a insert slot by now, since the current group contains at + // least one. For tables smaller than the group width, there will still be an + // empty element in the current (and only) group due to the load factor. + unsafe { + // SAFETY: + // * Caller of this function ensures that the control bytes are properly initialized. + // + // * We use this function with the slot / index found by `self.find_insert_slot_in_group` + return Err(self.fix_insert_slot(insert_slot.unwrap_unchecked())); + } + } + + probe_seq.move_next(self.bucket_mask); + } + } + + /// Searches for an empty or deleted bucket which is suitable for inserting a new + /// element and sets the hash for that slot. Returns an index of that slot and the + /// old control byte stored in the found index. + /// + /// This function does not check if the given element exists in the table. Also, + /// this function does not check if there is enough space in the table to insert + /// a new element. Caller of the funtion must make ensure that the table has at + /// least 1 empty or deleted `bucket`, otherwise this function will never return + /// (will go into an infinite loop) for tables larger than the group width, or + /// return an index outside of the table indices range if the table is less than + /// the group width. + /// + /// If there is at least 1 empty or deleted `bucket` in the table, the function is + /// guaranteed to return an `index` in the range `0..self.buckets()`, but in any case, + /// if this function returns an `index` it will be in the range `0..=self.buckets()`. + /// + /// This function does not make any changes to the `data` parts of the table, + /// or any changes to the the `items` or `growth_left` field of the table. + /// + /// # Safety + /// + /// The safety rules are directly derived from the safety rules for the + /// [`RawTableInner::set_ctrl_h2`] and [`RawTableInner::find_insert_slot`] methods. + /// Thus, in order to uphold the safety contracts for that methods, as well as for + /// the correct logic of the work of this crate, you must observe the following rules + /// when calling this function: + /// + /// * The [`RawTableInner`] has already been allocated and has properly initialized + /// control bytes otherwise calling this function results in [`undefined behavior`]. + /// + /// * The caller of this function must ensure that the "data" parts of the table + /// will have an entry in the returned index (matching the given hash) right + /// after calling this function. + /// + /// Attempt to write data at the `index` returned by this function when the table is + /// less than the group width and if there was not at least one empty or deleted bucket in + /// the table will cause immediate [`undefined behavior`]. This is because in this case the + /// function will return `self.bucket_mask + 1` as an index due to the trailing [`EMPTY] + /// control bytes outside the table range. + /// + /// The caller must independently increase the `items` field of the table, and also, + /// if the old control byte was [`EMPTY`], then decrease the table's `growth_left` + /// field, and do not change it if the old control byte was [`DELETED`]. + /// + /// See also [`Bucket::as_ptr`] method, for more information about of properly removing + /// or saving `element` from / into the [`RawTable`] / [`RawTableInner`]. + /// + /// [`Bucket::as_ptr`]: Bucket::as_ptr + /// [`undefined behavior`]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html + /// [`RawTableInner::ctrl`]: RawTableInner::ctrl + /// [`RawTableInner::set_ctrl_h2`]: RawTableInner::set_ctrl_h2 + /// [`RawTableInner::find_insert_slot`]: RawTableInner::find_insert_slot + #[inline] + unsafe fn prepare_insert_slot(&mut self, hash: u64) -> (usize, u8) { + // SAFETY: Caller of this function ensures that the control bytes are properly initialized. + let index: usize = self.find_insert_slot(hash).index; + // SAFETY: + // 1. The `find_insert_slot` function either returns an `index` less than or + // equal to `self.buckets() = self.bucket_mask + 1` of the table, or never + // returns if it cannot find an empty or deleted slot. + // 2. The caller of this function guarantees that the table has already been + // allocated + let old_ctrl = *self.ctrl(index); + self.set_ctrl_h2(index, hash); + (index, old_ctrl) + } + + /// Searches for an empty or deleted bucket which is suitable for inserting + /// a new element, returning the `index` for the new [`Bucket`]. + /// + /// This function does not make any changes to the `data` part of the table, or any + /// changes to the `items` or `growth_left` field of the table. + /// + /// The table must have at least 1 empty or deleted `bucket`, otherwise this function + /// will never return (will go into an infinite loop) for tables larger than the group + /// width, or return an index outside of the table indices range if the table is less + /// than the group width. + /// + /// If there is at least 1 empty or deleted `bucket` in the table, the function is + /// guaranteed to return [`InsertSlot`] with an index in the range `0..self.buckets()`, + /// but in any case, if this function returns [`InsertSlot`], it will contain an index + /// in the range `0..=self.buckets()`. + /// + /// # Safety + /// + /// The [`RawTableInner`] must have properly initialized control bytes otherwise calling + /// this function results in [`undefined behavior`]. + /// + /// Attempt to write data at the [`InsertSlot`] returned by this function when the table is + /// less than the group width and if there was not at least one empty or deleted bucket in + /// the table will cause immediate [`undefined behavior`]. This is because in this case the + /// function will return `self.bucket_mask + 1` as an index due to the trailing [`EMPTY] + /// control bytes outside the table range. + /// + /// [`undefined behavior`]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html + #[inline] + unsafe fn find_insert_slot(&self, hash: u64) -> InsertSlot { + let mut probe_seq = self.probe_seq(hash); + loop { + // SAFETY: + // * Caller of this function ensures that the control bytes are properly initialized. + // + // * `ProbeSeq.pos` cannot be greater than `self.bucket_mask = self.buckets() - 1` + // of the table due to masking with `self.bucket_mask` and also because mumber of + // buckets is a power of two (see `self.probe_seq` function). + // + // * Even if `ProbeSeq.pos` returns `position == self.bucket_mask`, it is safe to + // call `Group::load` due to the extended control bytes range, which is + // `self.bucket_mask + 1 + Group::WIDTH` (in fact, this means that the last control + // byte will never be read for the allocated table); + // + // * Also, even if `RawTableInner` is not already allocated, `ProbeSeq.pos` will + // always return "0" (zero), so Group::load will read unaligned `Group::static_empty()` + // bytes, which is safe (see RawTableInner::new). + let group = unsafe { Group::load(self.ctrl(probe_seq.pos)) }; + + let index = self.find_insert_slot_in_group(&group, &probe_seq); + if likely(index.is_some()) { + // SAFETY: + // * Caller of this function ensures that the control bytes are properly initialized. + // + // * We use this function with the slot / index found by `self.find_insert_slot_in_group` + unsafe { + return self.fix_insert_slot(index.unwrap_unchecked()); + } + } + probe_seq.move_next(self.bucket_mask); + } + } + + /// Searches for an element in a table, returning the `index` of the found element. + /// This uses dynamic dispatch to reduce the amount of code generated, but it is + /// eliminated by LLVM optimizations. + /// + /// This function does not make any changes to the `data` part of the table, or any + /// changes to the `items` or `growth_left` field of the table. + /// + /// The table must have at least 1 empty `bucket`, otherwise, if the + /// `eq: &mut dyn FnMut(usize) -> bool` function does not return `true`, + /// this function will also never return (will go into an infinite loop). + /// + /// This function is guaranteed to provide the `eq: &mut dyn FnMut(usize) -> bool` + /// function with only `FULL` buckets' indices and return the `index` of the found + /// element as `Some(index)`, so the index will always be in the range + /// `0..self.buckets()`. + /// + /// # Safety + /// + /// The [`RawTableInner`] must have properly initialized control bytes otherwise calling + /// this function results in [`undefined behavior`]. + /// + /// [`undefined behavior`]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html + #[inline(always)] + unsafe fn find_inner(&self, hash: u64, eq: &mut dyn FnMut(usize) -> bool) -> Option<usize> { + let h2_hash = h2(hash); + let mut probe_seq = self.probe_seq(hash); + + loop { + // SAFETY: + // * Caller of this function ensures that the control bytes are properly initialized. + // + // * `ProbeSeq.pos` cannot be greater than `self.bucket_mask = self.buckets() - 1` + // of the table due to masking with `self.bucket_mask`. + // + // * Even if `ProbeSeq.pos` returns `position == self.bucket_mask`, it is safe to + // call `Group::load` due to the extended control bytes range, which is + // `self.bucket_mask + 1 + Group::WIDTH` (in fact, this means that the last control + // byte will never be read for the allocated table); + // + // * Also, even if `RawTableInner` is not already allocated, `ProbeSeq.pos` will + // always return "0" (zero), so Group::load will read unaligned `Group::static_empty()` + // bytes, which is safe (see RawTableInner::new_in). + let group = unsafe { Group::load(self.ctrl(probe_seq.pos)) }; + + for bit in group.match_byte(h2_hash) { + // This is the same as `(probe_seq.pos + bit) % self.buckets()` because the number + // of buckets is a power of two, and `self.bucket_mask = self.buckets() - 1`. + let index = (probe_seq.pos + bit) & self.bucket_mask; + + if likely(eq(index)) { + return Some(index); + } + } + + if likely(group.match_empty().any_bit_set()) { + return None; + } + + probe_seq.move_next(self.bucket_mask); + } + } + + /// Prepares for rehashing data in place (that is, without allocating new memory). + /// Converts all full index `control bytes` to `DELETED` and all `DELETED` control + /// bytes to `EMPTY`, i.e. performs the following conversion: + /// + /// - `EMPTY` control bytes -> `EMPTY`; + /// - `DELETED` control bytes -> `EMPTY`; + /// - `FULL` control bytes -> `DELETED`. + /// + /// This function does not make any changes to the `data` parts of the table, + /// or any changes to the the `items` or `growth_left` field of the table. + /// + /// # Safety + /// + /// You must observe the following safety rules when calling this function: + /// + /// * The [`RawTableInner`] has already been allocated; + /// + /// * The caller of this function must convert the `DELETED` bytes back to `FULL` + /// bytes when re-inserting them into their ideal position (which was impossible + /// to do during the first insert due to tombstones). If the caller does not do + /// this, then calling this function may result in a memory leak. + /// + /// * The [`RawTableInner`] must have properly initialized control bytes otherwise + /// calling this function results in [`undefined behavior`]. + /// + /// Calling this function on a table that has not been allocated results in + /// [`undefined behavior`]. + /// + /// See also [`Bucket::as_ptr`] method, for more information about of properly removing + /// or saving `data element` from / into the [`RawTable`] / [`RawTableInner`]. + /// + /// [`Bucket::as_ptr`]: Bucket::as_ptr + /// [`undefined behavior`]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html + #[allow(clippy::mut_mut)] + #[inline] + unsafe fn prepare_rehash_in_place(&mut self) { + // Bulk convert all full control bytes to DELETED, and all DELETED control bytes to EMPTY. + // This effectively frees up all buckets containing a DELETED entry. + // + // SAFETY: + // 1. `i` is guaranteed to be within bounds since we are iterating from zero to `buckets - 1`; + // 2. Even if `i` will be `i == self.bucket_mask`, it is safe to call `Group::load_aligned` + // due to the extended control bytes range, which is `self.bucket_mask + 1 + Group::WIDTH`; + // 3. The caller of this function guarantees that [`RawTableInner`] has already been allocated; + // 4. We can use `Group::load_aligned` and `Group::store_aligned` here since we start from 0 + // and go to the end with a step equal to `Group::WIDTH` (see TableLayout::calculate_layout_for). + for i in (0..self.buckets()).step_by(Group::WIDTH) { + let group = Group::load_aligned(self.ctrl(i)); + let group = group.convert_special_to_empty_and_full_to_deleted(); + group.store_aligned(self.ctrl(i)); + } + + // Fix up the trailing control bytes. See the comments in set_ctrl + // for the handling of tables smaller than the group width. + // + // SAFETY: The caller of this function guarantees that [`RawTableInner`] + // has already been allocated + if unlikely(self.buckets() < Group::WIDTH) { + // SAFETY: We have `self.bucket_mask + 1 + Group::WIDTH` number of control bytes, + // so copying `self.buckets() == self.bucket_mask + 1` bytes with offset equal to + // `Group::WIDTH` is safe + self.ctrl(0) + .copy_to(self.ctrl(Group::WIDTH), self.buckets()); + } else { + // SAFETY: We have `self.bucket_mask + 1 + Group::WIDTH` number of + // control bytes,so copying `Group::WIDTH` bytes with offset equal + // to `self.buckets() == self.bucket_mask + 1` is safe + self.ctrl(0) + .copy_to(self.ctrl(self.buckets()), Group::WIDTH); + } + } + + /// Returns an iterator over every element in the table. + /// + /// # Safety + /// + /// If any of the following conditions are violated, the result + /// is [`undefined behavior`]: + /// + /// * The caller has to ensure that the `RawTableInner` outlives the + /// `RawIter`. Because we cannot make the `next` method unsafe on + /// the `RawIter` struct, we have to make the `iter` method unsafe. + /// + /// * The [`RawTableInner`] must have properly initialized control bytes. + /// + /// [`undefined behavior`]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html + #[inline] + unsafe fn iter<T>(&self) -> RawIter<T> { + // SAFETY: + // 1. Since the caller of this function ensures that the control bytes + // are properly initialized and `self.data_end()` points to the start + // of the array of control bytes, therefore: `ctrl` is valid for reads, + // properly aligned to `Group::WIDTH` and points to the properly initialized + // control bytes. + // 2. `data` bucket index in the table is equal to the `ctrl` index (i.e. + // equal to zero). + // 3. We pass the exact value of buckets of the table to the function. + // + // `ctrl` points here (to the start + // of the first control byte `CT0`) + // ∨ + // [Pad], T_n, ..., T1, T0, |CT0, CT1, ..., CT_n|, Group::WIDTH + // \________ ________/ + // \/ + // `n = buckets - 1`, i.e. `RawIndexTableInner::buckets() - 1` + // + // where: T0...T_n - our stored data; + // CT0...CT_n - control bytes or metadata for `data`. + let data = Bucket::from_base_index(self.data_end(), 0); + RawIter { + // SAFETY: See explanation above + iter: RawIterRange::new(self.ctrl.as_ptr(), data, self.buckets()), + items: self.items, + } + } + + /// Executes the destructors (if any) of the values stored in the table. + /// + /// # Note + /// + /// This function does not erase the control bytes of the table and does + /// not make any changes to the `items` or `growth_left` fields of the + /// table. If necessary, the caller of this function must manually set + /// up these table fields, for example using the [`clear_no_drop`] function. + /// + /// Be careful during calling this function, because drop function of + /// the elements can panic, and this can leave table in an inconsistent + /// state. + /// + /// # Safety + /// + /// If `T` is a type that should be dropped and **the table is not empty**, + /// calling this function more than once results in [`undefined behavior`]. + /// + /// If `T` is not [`Copy`], attempting to use values stored in the table after + /// calling this function may result in [`undefined behavior`]. + /// + /// It is safe to call this function on a table that has not been allocated, + /// on a table with uninitialized control bytes, and on a table with no actual + /// data but with `Full` control bytes if `self.items == 0`. + /// + /// See also [`Bucket::drop`] / [`Bucket::as_ptr`] methods, for more information + /// about of properly removing or saving `element` from / into the [`RawTable`] / + /// [`RawTableInner`]. + /// + /// [`Bucket::drop`]: Bucket::drop + /// [`Bucket::as_ptr`]: Bucket::as_ptr + /// [`clear_no_drop`]: RawTableInner::clear_no_drop + /// [`undefined behavior`]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html + unsafe fn drop_elements<T>(&mut self) { + // Check that `self.items != 0`. Protects against the possibility + // of creating an iterator on an table with uninitialized control bytes. + if T::NEEDS_DROP && self.items != 0 { + // SAFETY: We know for sure that RawTableInner will outlive the + // returned `RawIter` iterator, and the caller of this function + // must uphold the safety contract for `drop_elements` method. + for item in self.iter::<T>() { + // SAFETY: The caller must uphold the safety contract for + // `drop_elements` method. + item.drop(); + } + } + } + + /// Executes the destructors (if any) of the values stored in the table and than + /// deallocates the table. + /// + /// # Note + /// + /// Calling this function automatically makes invalid (dangling) all instances of + /// buckets ([`Bucket`]) and makes invalid (dangling) the `ctrl` field of the table. + /// + /// This function does not make any changes to the `bucket_mask`, `items` or `growth_left` + /// fields of the table. If necessary, the caller of this function must manually set + /// up these table fields. + /// + /// # Safety + /// + /// If any of the following conditions are violated, the result is [`undefined behavior`]: + /// + /// * Calling this function more than once; + /// + /// * The `alloc` must be the same [`Allocator`] as the `Allocator` that was used + /// to allocate this table. + /// + /// * The `table_layout` must be the same [`TableLayout`] as the `TableLayout` that + /// was used to allocate this table. + /// + /// The caller of this function should pay attention to the possibility of the + /// elements' drop function panicking, because this: + /// + /// * May leave the table in an inconsistent state; + /// + /// * Memory is never deallocated, so a memory leak may occur. + /// + /// Attempt to use the `ctrl` field of the table (dereference) after calling this + /// function results in [`undefined behavior`]. + /// + /// It is safe to call this function on a table that has not been allocated, + /// on a table with uninitialized control bytes, and on a table with no actual + /// data but with `Full` control bytes if `self.items == 0`. + /// + /// See also [`RawTableInner::drop_elements`] or [`RawTableInner::free_buckets`] + /// for more information. + /// + /// [`RawTableInner::drop_elements`]: RawTableInner::drop_elements + /// [`RawTableInner::free_buckets`]: RawTableInner::free_buckets + /// [`undefined behavior`]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html + unsafe fn drop_inner_table<T, A: Allocator>(&mut self, alloc: &A, table_layout: TableLayout) { + if !self.is_empty_singleton() { + unsafe { + // SAFETY: The caller must uphold the safety contract for `drop_inner_table` method. + self.drop_elements::<T>(); + // SAFETY: + // 1. We have checked that our table is allocated. + // 2. The caller must uphold the safety contract for `drop_inner_table` method. + self.free_buckets(alloc, table_layout); + } + } + } + + #[inline] + unsafe fn bucket<T>(&self, index: usize) -> Bucket<T> { + debug_assert_ne!(self.bucket_mask, 0); + debug_assert!(index < self.buckets()); + Bucket::from_base_index(self.data_end(), index) + } + + #[inline] + unsafe fn bucket_ptr(&self, index: usize, size_of: usize) -> *mut u8 { + debug_assert_ne!(self.bucket_mask, 0); + debug_assert!(index < self.buckets()); + let base: *mut u8 = self.data_end().as_ptr(); + base.sub((index + 1) * size_of) + } + + #[inline] + unsafe fn data_end<T>(&self) -> NonNull<T> { + NonNull::new_unchecked(self.ctrl.as_ptr().cast()) + } + + /// Returns an iterator-like object for a probe sequence on the table. + /// + /// This iterator never terminates, but is guaranteed to visit each bucket + /// group exactly once. The loop using `probe_seq` must terminate upon + /// reaching a group containing an empty bucket. + #[inline] + fn probe_seq(&self, hash: u64) -> ProbeSeq { + ProbeSeq { + // This is the same as `hash as usize % self.buckets()` because the number + // of buckets is a power of two, and `self.bucket_mask = self.buckets() - 1`. + pos: h1(hash) & self.bucket_mask, + stride: 0, + } + } + + /// Returns the index of a bucket for which a value must be inserted if there is enough rooom + /// in the table, otherwise returns error + #[cfg(feature = "raw")] + #[inline] + unsafe fn prepare_insert_no_grow(&mut self, hash: u64) -> Result<usize, ()> { + let index = self.find_insert_slot(hash).index; + let old_ctrl = *self.ctrl(index); + if unlikely(self.growth_left == 0 && special_is_empty(old_ctrl)) { + Err(()) + } else { + self.record_item_insert_at(index, old_ctrl, hash); + Ok(index) + } + } + + #[inline] + unsafe fn record_item_insert_at(&mut self, index: usize, old_ctrl: u8, hash: u64) { + self.growth_left -= usize::from(special_is_empty(old_ctrl)); + self.set_ctrl_h2(index, hash); + self.items += 1; + } + + #[inline] + fn is_in_same_group(&self, i: usize, new_i: usize, hash: u64) -> bool { + let probe_seq_pos = self.probe_seq(hash).pos; + let probe_index = + |pos: usize| (pos.wrapping_sub(probe_seq_pos) & self.bucket_mask) / Group::WIDTH; + probe_index(i) == probe_index(new_i) + } + + /// Sets a control byte to the hash, and possibly also the replicated control byte at + /// the end of the array. + /// + /// This function does not make any changes to the `data` parts of the table, + /// or any changes to the the `items` or `growth_left` field of the table. + /// + /// # Safety + /// + /// The safety rules are directly derived from the safety rules for [`RawTableInner::set_ctrl`] + /// method. Thus, in order to uphold the safety contracts for the method, you must observe the + /// following rules when calling this function: + /// + /// * The [`RawTableInner`] has already been allocated; + /// + /// * The `index` must not be greater than the `RawTableInner.bucket_mask`, i.e. + /// `index <= RawTableInner.bucket_mask` or, in other words, `(index + 1)` must + /// be no greater than the number returned by the function [`RawTableInner::buckets`]. + /// + /// Calling this function on a table that has not been allocated results in [`undefined behavior`]. + /// + /// See also [`Bucket::as_ptr`] method, for more information about of properly removing + /// or saving `data element` from / into the [`RawTable`] / [`RawTableInner`]. + /// + /// [`RawTableInner::set_ctrl`]: RawTableInner::set_ctrl + /// [`RawTableInner::buckets`]: RawTableInner::buckets + /// [`Bucket::as_ptr`]: Bucket::as_ptr + /// [`undefined behavior`]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html + #[inline] + unsafe fn set_ctrl_h2(&mut self, index: usize, hash: u64) { + // SAFETY: The caller must uphold the safety rules for the [`RawTableInner::set_ctrl_h2`] + self.set_ctrl(index, h2(hash)); + } + + /// Replaces the hash in the control byte at the given index with the provided one, + /// and possibly also replicates the new control byte at the end of the array of control + /// bytes, returning the old control byte. + /// + /// This function does not make any changes to the `data` parts of the table, + /// or any changes to the the `items` or `growth_left` field of the table. + /// + /// # Safety + /// + /// The safety rules are directly derived from the safety rules for [`RawTableInner::set_ctrl_h2`] + /// and [`RawTableInner::ctrl`] methods. Thus, in order to uphold the safety contracts for both + /// methods, you must observe the following rules when calling this function: + /// + /// * The [`RawTableInner`] has already been allocated; + /// + /// * The `index` must not be greater than the `RawTableInner.bucket_mask`, i.e. + /// `index <= RawTableInner.bucket_mask` or, in other words, `(index + 1)` must + /// be no greater than the number returned by the function [`RawTableInner::buckets`]. + /// + /// Calling this function on a table that has not been allocated results in [`undefined behavior`]. + /// + /// See also [`Bucket::as_ptr`] method, for more information about of properly removing + /// or saving `data element` from / into the [`RawTable`] / [`RawTableInner`]. + /// + /// [`RawTableInner::set_ctrl_h2`]: RawTableInner::set_ctrl_h2 + /// [`RawTableInner::buckets`]: RawTableInner::buckets + /// [`Bucket::as_ptr`]: Bucket::as_ptr + /// [`undefined behavior`]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html + #[inline] + unsafe fn replace_ctrl_h2(&mut self, index: usize, hash: u64) -> u8 { + // SAFETY: The caller must uphold the safety rules for the [`RawTableInner::replace_ctrl_h2`] + let prev_ctrl = *self.ctrl(index); + self.set_ctrl_h2(index, hash); + prev_ctrl + } + + /// Sets a control byte, and possibly also the replicated control byte at + /// the end of the array. + /// + /// This function does not make any changes to the `data` parts of the table, + /// or any changes to the the `items` or `growth_left` field of the table. + /// + /// # Safety + /// + /// You must observe the following safety rules when calling this function: + /// + /// * The [`RawTableInner`] has already been allocated; + /// + /// * The `index` must not be greater than the `RawTableInner.bucket_mask`, i.e. + /// `index <= RawTableInner.bucket_mask` or, in other words, `(index + 1)` must + /// be no greater than the number returned by the function [`RawTableInner::buckets`]. + /// + /// Calling this function on a table that has not been allocated results in [`undefined behavior`]. + /// + /// See also [`Bucket::as_ptr`] method, for more information about of properly removing + /// or saving `data element` from / into the [`RawTable`] / [`RawTableInner`]. + /// + /// [`RawTableInner::buckets`]: RawTableInner::buckets + /// [`Bucket::as_ptr`]: Bucket::as_ptr + /// [`undefined behavior`]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html + #[inline] + unsafe fn set_ctrl(&mut self, index: usize, ctrl: u8) { + // Replicate the first Group::WIDTH control bytes at the end of + // the array without using a branch. If the tables smaller than + // the group width (self.buckets() < Group::WIDTH), + // `index2 = Group::WIDTH + index`, otherwise `index2` is: + // + // - If index >= Group::WIDTH then index == index2. + // - Otherwise index2 == self.bucket_mask + 1 + index. + // + // The very last replicated control byte is never actually read because + // we mask the initial index for unaligned loads, but we write it + // anyways because it makes the set_ctrl implementation simpler. + // + // If there are fewer buckets than Group::WIDTH then this code will + // replicate the buckets at the end of the trailing group. For example + // with 2 buckets and a group size of 4, the control bytes will look + // like this: + // + // Real | Replicated + // --------------------------------------------- + // | [A] | [B] | [EMPTY] | [EMPTY] | [A] | [B] | + // --------------------------------------------- + + // This is the same as `(index.wrapping_sub(Group::WIDTH)) % self.buckets() + Group::WIDTH` + // because the number of buckets is a power of two, and `self.bucket_mask = self.buckets() - 1`. + let index2 = ((index.wrapping_sub(Group::WIDTH)) & self.bucket_mask) + Group::WIDTH; + + // SAFETY: The caller must uphold the safety rules for the [`RawTableInner::set_ctrl`] + *self.ctrl(index) = ctrl; + *self.ctrl(index2) = ctrl; + } + + /// Returns a pointer to a control byte. + /// + /// # Safety + /// + /// For the allocated [`RawTableInner`], the result is [`Undefined Behavior`], + /// if the `index` is greater than the `self.bucket_mask + 1 + Group::WIDTH`. + /// In that case, calling this function with `index == self.bucket_mask + 1 + Group::WIDTH` + /// will return a pointer to the end of the allocated table and it is useless on its own. + /// + /// Calling this function with `index >= self.bucket_mask + 1 + Group::WIDTH` on a + /// table that has not been allocated results in [`Undefined Behavior`]. + /// + /// So to satisfy both requirements you should always follow the rule that + /// `index < self.bucket_mask + 1 + Group::WIDTH` + /// + /// Calling this function on [`RawTableInner`] that are not already allocated is safe + /// for read-only purpose. + /// + /// See also [`Bucket::as_ptr()`] method, for more information about of properly removing + /// or saving `data element` from / into the [`RawTable`] / [`RawTableInner`]. + /// + /// [`Bucket::as_ptr()`]: Bucket::as_ptr() + /// [`Undefined Behavior`]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html + #[inline] + unsafe fn ctrl(&self, index: usize) -> *mut u8 { + debug_assert!(index < self.num_ctrl_bytes()); + // SAFETY: The caller must uphold the safety rules for the [`RawTableInner::ctrl`] + self.ctrl.as_ptr().add(index) + } + + #[inline] + fn buckets(&self) -> usize { + self.bucket_mask + 1 + } + + /// Checks whether the bucket at `index` is full. + /// + /// # Safety + /// + /// The caller must ensure `index` is less than the number of buckets. + #[inline] + unsafe fn is_bucket_full(&self, index: usize) -> bool { + debug_assert!(index < self.buckets()); + is_full(*self.ctrl(index)) + } + + #[inline] + fn num_ctrl_bytes(&self) -> usize { + self.bucket_mask + 1 + Group::WIDTH + } + + #[inline] + fn is_empty_singleton(&self) -> bool { + self.bucket_mask == 0 + } + + /// Attempts to allocate a new hash table with at least enough capacity + /// for inserting the given number of elements without reallocating, + /// and return it inside ScopeGuard to protect against panic in the hash + /// function. + /// + /// # Note + /// + /// It is recommended (but not required): + /// + /// * That the new table's `capacity` be greater than or equal to `self.items`. + /// + /// * The `alloc` is the same [`Allocator`] as the `Allocator` used + /// to allocate this table. + /// + /// * The `table_layout` is the same [`TableLayout`] as the `TableLayout` used + /// to allocate this table. + /// + /// If `table_layout` does not match the `TableLayout` that was used to allocate + /// this table, then using `mem::swap` with the `self` and the new table returned + /// by this function results in [`undefined behavior`]. + /// + /// [`undefined behavior`]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html + #[allow(clippy::mut_mut)] + #[inline] + fn prepare_resize<'a, A>( + &self, + alloc: &'a A, + table_layout: TableLayout, + capacity: usize, + fallibility: Fallibility, + ) -> Result<crate::scopeguard::ScopeGuard<Self, impl FnMut(&mut Self) + 'a>, TryReserveError> + where + A: Allocator, + { + debug_assert!(self.items <= capacity); + + // Allocate and initialize the new table. + let new_table = + RawTableInner::fallible_with_capacity(alloc, table_layout, capacity, fallibility)?; + + // The hash function may panic, in which case we simply free the new + // table without dropping any elements that may have been copied into + // it. + // + // This guard is also used to free the old table on success, see + // the comment at the bottom of this function. + Ok(guard(new_table, move |self_| { + if !self_.is_empty_singleton() { + // SAFETY: + // 1. We have checked that our table is allocated. + // 2. We know for sure that the `alloc` and `table_layout` matches the + // [`Allocator`] and [`TableLayout`] used to allocate this table. + unsafe { self_.free_buckets(alloc, table_layout) }; + } + })) + } + + /// Reserves or rehashes to make room for `additional` more elements. + /// + /// This uses dynamic dispatch to reduce the amount of + /// code generated, but it is eliminated by LLVM optimizations when inlined. + /// + /// # Safety + /// + /// If any of the following conditions are violated, the result is + /// [`undefined behavior`]: + /// + /// * The `alloc` must be the same [`Allocator`] as the `Allocator` used + /// to allocate this table. + /// + /// * The `layout` must be the same [`TableLayout`] as the `TableLayout` + /// used to allocate this table. + /// + /// * The `drop` function (`fn(*mut u8)`) must be the actual drop function of + /// the elements stored in the table. + /// + /// * The [`RawTableInner`] must have properly initialized control bytes. + /// + /// [`undefined behavior`]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html + #[allow(clippy::inline_always)] + #[inline(always)] + unsafe fn reserve_rehash_inner<A>( + &mut self, + alloc: &A, + additional: usize, + hasher: &dyn Fn(&mut Self, usize) -> u64, + fallibility: Fallibility, + layout: TableLayout, + drop: Option<fn(*mut u8)>, + ) -> Result<(), TryReserveError> + where + A: Allocator, + { + // Avoid `Option::ok_or_else` because it bloats LLVM IR. + let new_items = match self.items.checked_add(additional) { + Some(new_items) => new_items, + None => return Err(fallibility.capacity_overflow()), + }; + let full_capacity = bucket_mask_to_capacity(self.bucket_mask); + if new_items <= full_capacity / 2 { + // Rehash in-place without re-allocating if we have plenty of spare + // capacity that is locked up due to DELETED entries. + + // SAFETY: + // 1. We know for sure that `[`RawTableInner`]` has already been allocated + // (since new_items <= full_capacity / 2); + // 2. The caller ensures that `drop` function is the actual drop function of + // the elements stored in the table. + // 3. The caller ensures that `layout` matches the [`TableLayout`] that was + // used to allocate this table. + // 4. The caller ensures that the control bytes of the `RawTableInner` + // are already initialized. + self.rehash_in_place(hasher, layout.size, drop); + Ok(()) + } else { + // Otherwise, conservatively resize to at least the next size up + // to avoid churning deletes into frequent rehashes. + // + // SAFETY: + // 1. We know for sure that `capacity >= self.items`. + // 2. The caller ensures that `alloc` and `layout` matches the [`Allocator`] and + // [`TableLayout`] that were used to allocate this table. + // 3. The caller ensures that the control bytes of the `RawTableInner` + // are already initialized. + self.resize_inner( + alloc, + usize::max(new_items, full_capacity + 1), + hasher, + fallibility, + layout, + ) + } + } + + /// Returns an iterator over full buckets indices in the table. + /// + /// # Safety + /// + /// Behavior is undefined if any of the following conditions are violated: + /// + /// * The caller has to ensure that the `RawTableInner` outlives the + /// `FullBucketsIndices`. Because we cannot make the `next` method + /// unsafe on the `FullBucketsIndices` struct, we have to make the + /// `full_buckets_indices` method unsafe. + /// + /// * The [`RawTableInner`] must have properly initialized control bytes. + #[inline(always)] + unsafe fn full_buckets_indices(&self) -> FullBucketsIndices { + // SAFETY: + // 1. Since the caller of this function ensures that the control bytes + // are properly initialized and `self.ctrl(0)` points to the start + // of the array of control bytes, therefore: `ctrl` is valid for reads, + // properly aligned to `Group::WIDTH` and points to the properly initialized + // control bytes. + // 2. The value of `items` is equal to the amount of data (values) added + // to the table. + // + // `ctrl` points here (to the start + // of the first control byte `CT0`) + // ∨ + // [Pad], T_n, ..., T1, T0, |CT0, CT1, ..., CT_n|, Group::WIDTH + // \________ ________/ + // \/ + // `n = buckets - 1`, i.e. `RawIndexTableInner::buckets() - 1` + // + // where: T0...T_n - our stored data; + // CT0...CT_n - control bytes or metadata for `data`. + let ctrl = NonNull::new_unchecked(self.ctrl(0)); + + FullBucketsIndices { + // Load the first group + // SAFETY: See explanation above. + current_group: Group::load_aligned(ctrl.as_ptr()).match_full().into_iter(), + group_first_index: 0, + ctrl, + items: self.items, + } + } + + /// Allocates a new table of a different size and moves the contents of the + /// current table into it. + /// + /// This uses dynamic dispatch to reduce the amount of + /// code generated, but it is eliminated by LLVM optimizations when inlined. + /// + /// # Safety + /// + /// If any of the following conditions are violated, the result is + /// [`undefined behavior`]: + /// + /// * The `alloc` must be the same [`Allocator`] as the `Allocator` used + /// to allocate this table; + /// + /// * The `layout` must be the same [`TableLayout`] as the `TableLayout` + /// used to allocate this table; + /// + /// * The [`RawTableInner`] must have properly initialized control bytes. + /// + /// The caller of this function must ensure that `capacity >= self.items` + /// otherwise: + /// + /// * If `self.items != 0`, calling of this function with `capacity == 0` + /// results in [`undefined behavior`]. + /// + /// * If `capacity_to_buckets(capacity) < Group::WIDTH` and + /// `self.items > capacity_to_buckets(capacity)` calling this function + /// results in [`undefined behavior`]. + /// + /// * If `capacity_to_buckets(capacity) >= Group::WIDTH` and + /// `self.items > capacity_to_buckets(capacity)` calling this function + /// are never return (will go into an infinite loop). + /// + /// Note: It is recommended (but not required) that the new table's `capacity` + /// be greater than or equal to `self.items`. In case if `capacity <= self.items` + /// this function can never return. See [`RawTableInner::find_insert_slot`] for + /// more information. + /// + /// [`RawTableInner::find_insert_slot`]: RawTableInner::find_insert_slot + /// [`undefined behavior`]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html + #[allow(clippy::inline_always)] + #[inline(always)] + unsafe fn resize_inner<A>( + &mut self, + alloc: &A, + capacity: usize, + hasher: &dyn Fn(&mut Self, usize) -> u64, + fallibility: Fallibility, + layout: TableLayout, + ) -> Result<(), TryReserveError> + where + A: Allocator, + { + // SAFETY: We know for sure that `alloc` and `layout` matches the [`Allocator`] and [`TableLayout`] + // that were used to allocate this table. + let mut new_table = self.prepare_resize(alloc, layout, capacity, fallibility)?; + + // SAFETY: We know for sure that RawTableInner will outlive the + // returned `FullBucketsIndices` iterator, and the caller of this + // function ensures that the control bytes are properly initialized. + for full_byte_index in self.full_buckets_indices() { + // This may panic. + let hash = hasher(self, full_byte_index); + + // SAFETY: + // We can use a simpler version of insert() here since: + // 1. There are no DELETED entries. + // 2. We know there is enough space in the table. + // 3. All elements are unique. + // 4. The caller of this function guarantees that `capacity > 0` + // so `new_table` must already have some allocated memory. + // 5. We set `growth_left` and `items` fields of the new table + // after the loop. + // 6. We insert into the table, at the returned index, the data + // matching the given hash immediately after calling this function. + let (new_index, _) = new_table.prepare_insert_slot(hash); + + // SAFETY: + // + // * `src` is valid for reads of `layout.size` bytes, since the + // table is alive and the `full_byte_index` is guaranteed to be + // within bounds (see `FullBucketsIndices::next_impl`); + // + // * `dst` is valid for writes of `layout.size` bytes, since the + // caller ensures that `table_layout` matches the [`TableLayout`] + // that was used to allocate old table and we have the `new_index` + // returned by `prepare_insert_slot`. + // + // * Both `src` and `dst` are properly aligned. + // + // * Both `src` and `dst` point to different region of memory. + ptr::copy_nonoverlapping( + self.bucket_ptr(full_byte_index, layout.size), + new_table.bucket_ptr(new_index, layout.size), + layout.size, + ); + } + + // The hash function didn't panic, so we can safely set the + // `growth_left` and `items` fields of the new table. + new_table.growth_left -= self.items; + new_table.items = self.items; + + // We successfully copied all elements without panicking. Now replace + // self with the new table. The old table will have its memory freed but + // the items will not be dropped (since they have been moved into the + // new table). + // SAFETY: The caller ensures that `table_layout` matches the [`TableLayout`] + // that was used to allocate this table. + mem::swap(self, &mut new_table); + + Ok(()) + } + + /// Rehashes the contents of the table in place (i.e. without changing the + /// allocation). + /// + /// If `hasher` panics then some the table's contents may be lost. + /// + /// This uses dynamic dispatch to reduce the amount of + /// code generated, but it is eliminated by LLVM optimizations when inlined. + /// + /// # Safety + /// + /// If any of the following conditions are violated, the result is [`undefined behavior`]: + /// + /// * The `size_of` must be equal to the size of the elements stored in the table; + /// + /// * The `drop` function (`fn(*mut u8)`) must be the actual drop function of + /// the elements stored in the table. + /// + /// * The [`RawTableInner`] has already been allocated; + /// + /// * The [`RawTableInner`] must have properly initialized control bytes. + /// + /// [`undefined behavior`]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html + #[allow(clippy::inline_always)] + #[cfg_attr(feature = "inline-more", inline(always))] + #[cfg_attr(not(feature = "inline-more"), inline)] + unsafe fn rehash_in_place( + &mut self, + hasher: &dyn Fn(&mut Self, usize) -> u64, + size_of: usize, + drop: Option<fn(*mut u8)>, + ) { + // If the hash function panics then properly clean up any elements + // that we haven't rehashed yet. We unfortunately can't preserve the + // element since we lost their hash and have no way of recovering it + // without risking another panic. + self.prepare_rehash_in_place(); + + let mut guard = guard(self, move |self_| { + if let Some(drop) = drop { + for i in 0..self_.buckets() { + if *self_.ctrl(i) == DELETED { + self_.set_ctrl(i, EMPTY); + drop(self_.bucket_ptr(i, size_of)); + self_.items -= 1; + } + } + } + self_.growth_left = bucket_mask_to_capacity(self_.bucket_mask) - self_.items; + }); + + // At this point, DELETED elements are elements that we haven't + // rehashed yet. Find them and re-insert them at their ideal + // position. + 'outer: for i in 0..guard.buckets() { + if *guard.ctrl(i) != DELETED { + continue; + } + + let i_p = guard.bucket_ptr(i, size_of); + + 'inner: loop { + // Hash the current item + let hash = hasher(*guard, i); + + // Search for a suitable place to put it + // + // SAFETY: Caller of this function ensures that the control bytes + // are properly initialized. + let new_i = guard.find_insert_slot(hash).index; + + // Probing works by scanning through all of the control + // bytes in groups, which may not be aligned to the group + // size. If both the new and old position fall within the + // same unaligned group, then there is no benefit in moving + // it and we can just continue to the next item. + if likely(guard.is_in_same_group(i, new_i, hash)) { + guard.set_ctrl_h2(i, hash); + continue 'outer; + } + + let new_i_p = guard.bucket_ptr(new_i, size_of); + + // We are moving the current item to a new position. Write + // our H2 to the control byte of the new position. + let prev_ctrl = guard.replace_ctrl_h2(new_i, hash); + if prev_ctrl == EMPTY { + guard.set_ctrl(i, EMPTY); + // If the target slot is empty, simply move the current + // element into the new slot and clear the old control + // byte. + ptr::copy_nonoverlapping(i_p, new_i_p, size_of); + continue 'outer; + } else { + // If the target slot is occupied, swap the two elements + // and then continue processing the element that we just + // swapped into the old slot. + debug_assert_eq!(prev_ctrl, DELETED); + ptr::swap_nonoverlapping(i_p, new_i_p, size_of); + continue 'inner; + } + } + } + + guard.growth_left = bucket_mask_to_capacity(guard.bucket_mask) - guard.items; + + mem::forget(guard); + } + + /// Deallocates the table without dropping any entries. + /// + /// # Note + /// + /// This function must be called only after [`drop_elements`](RawTable::drop_elements), + /// else it can lead to leaking of memory. Also calling this function automatically + /// makes invalid (dangling) all instances of buckets ([`Bucket`]) and makes invalid + /// (dangling) the `ctrl` field of the table. + /// + /// # Safety + /// + /// If any of the following conditions are violated, the result is [`Undefined Behavior`]: + /// + /// * The [`RawTableInner`] has already been allocated; + /// + /// * The `alloc` must be the same [`Allocator`] as the `Allocator` that was used + /// to allocate this table. + /// + /// * The `table_layout` must be the same [`TableLayout`] as the `TableLayout` that was used + /// to allocate this table. + /// + /// See also [`GlobalAlloc::dealloc`] or [`Allocator::deallocate`] for more information. + /// + /// [`Undefined Behavior`]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html + /// [`GlobalAlloc::dealloc`]: https://doc.rust-lang.org/alloc/alloc/trait.GlobalAlloc.html#tymethod.dealloc + /// [`Allocator::deallocate`]: https://doc.rust-lang.org/alloc/alloc/trait.Allocator.html#tymethod.deallocate + #[inline] + unsafe fn free_buckets<A>(&mut self, alloc: &A, table_layout: TableLayout) + where + A: Allocator, + { + // SAFETY: The caller must uphold the safety contract for `free_buckets` + // method. + let (ptr, layout) = self.allocation_info(table_layout); + alloc.deallocate(ptr, layout); + } + + /// Returns a pointer to the allocated memory and the layout that was used to + /// allocate the table. + /// + /// # Safety + /// + /// Caller of this function must observe the following safety rules: + /// + /// * The [`RawTableInner`] has already been allocated, otherwise + /// calling this function results in [`undefined behavior`] + /// + /// * The `table_layout` must be the same [`TableLayout`] as the `TableLayout` + /// that was used to allocate this table. Failure to comply with this condition + /// may result in [`undefined behavior`]. + /// + /// See also [`GlobalAlloc::dealloc`] or [`Allocator::deallocate`] for more information. + /// + /// [`undefined behavior`]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html + /// [`GlobalAlloc::dealloc`]: https://doc.rust-lang.org/alloc/alloc/trait.GlobalAlloc.html#tymethod.dealloc + /// [`Allocator::deallocate`]: https://doc.rust-lang.org/alloc/alloc/trait.Allocator.html#tymethod.deallocate + #[inline] + unsafe fn allocation_info(&self, table_layout: TableLayout) -> (NonNull<u8>, Layout) { + debug_assert!( + !self.is_empty_singleton(), + "this function can only be called on non-empty tables" + ); + + // Avoid `Option::unwrap_or_else` because it bloats LLVM IR. + let (layout, ctrl_offset) = match table_layout.calculate_layout_for(self.buckets()) { + Some(lco) => lco, + None => unsafe { hint::unreachable_unchecked() }, + }; + ( + // SAFETY: The caller must uphold the safety contract for `allocation_info` method. + unsafe { NonNull::new_unchecked(self.ctrl.as_ptr().sub(ctrl_offset)) }, + layout, + ) + } + + /// Returns a pointer to the allocated memory and the layout that was used to + /// allocate the table. If [`RawTableInner`] has not been allocated, this + /// function return `dangling` pointer and `()` (unit) layout. + /// + /// # Safety + /// + /// The `table_layout` must be the same [`TableLayout`] as the `TableLayout` + /// that was used to allocate this table. Failure to comply with this condition + /// may result in [`undefined behavior`]. + /// + /// See also [`GlobalAlloc::dealloc`] or [`Allocator::deallocate`] for more information. + /// + /// [`undefined behavior`]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html + /// [`GlobalAlloc::dealloc`]: https://doc.rust-lang.org/alloc/alloc/trait.GlobalAlloc.html#tymethod.dealloc + /// [`Allocator::deallocate`]: https://doc.rust-lang.org/alloc/alloc/trait.Allocator.html#tymethod.deallocate + #[cfg(feature = "raw")] + unsafe fn allocation_info_or_zero(&self, table_layout: TableLayout) -> (NonNull<u8>, Layout) { + if self.is_empty_singleton() { + (NonNull::dangling(), Layout::new::<()>()) + } else { + // SAFETY: + // 1. We have checked that our table is allocated. + // 2. The caller ensures that `table_layout` matches the [`TableLayout`] + // that was used to allocate this table. + unsafe { self.allocation_info(table_layout) } + } + } + + /// Marks all table buckets as empty without dropping their contents. + #[inline] + fn clear_no_drop(&mut self) { + if !self.is_empty_singleton() { + unsafe { + self.ctrl(0).write_bytes(EMPTY, self.num_ctrl_bytes()); + } + } + self.items = 0; + self.growth_left = bucket_mask_to_capacity(self.bucket_mask); + } + + /// Erases the [`Bucket`]'s control byte at the given index so that it does not + /// triggered as full, decreases the `items` of the table and, if it can be done, + /// increases `self.growth_left`. + /// + /// This function does not actually erase / drop the [`Bucket`] itself, i.e. it + /// does not make any changes to the `data` parts of the table. The caller of this + /// function must take care to properly drop the `data`, otherwise calling this + /// function may result in a memory leak. + /// + /// # Safety + /// + /// You must observe the following safety rules when calling this function: + /// + /// * The [`RawTableInner`] has already been allocated; + /// + /// * It must be the full control byte at the given position; + /// + /// * The `index` must not be greater than the `RawTableInner.bucket_mask`, i.e. + /// `index <= RawTableInner.bucket_mask` or, in other words, `(index + 1)` must + /// be no greater than the number returned by the function [`RawTableInner::buckets`]. + /// + /// Calling this function on a table that has not been allocated results in [`undefined behavior`]. + /// + /// Calling this function on a table with no elements is unspecified, but calling subsequent + /// functions is likely to result in [`undefined behavior`] due to overflow subtraction + /// (`self.items -= 1 cause overflow when self.items == 0`). + /// + /// See also [`Bucket::as_ptr`] method, for more information about of properly removing + /// or saving `data element` from / into the [`RawTable`] / [`RawTableInner`]. + /// + /// [`RawTableInner::buckets`]: RawTableInner::buckets + /// [`Bucket::as_ptr`]: Bucket::as_ptr + /// [`undefined behavior`]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html + #[inline] + unsafe fn erase(&mut self, index: usize) { + debug_assert!(self.is_bucket_full(index)); + + // This is the same as `index.wrapping_sub(Group::WIDTH) % self.buckets()` because + // the number of buckets is a power of two, and `self.bucket_mask = self.buckets() - 1`. + let index_before = index.wrapping_sub(Group::WIDTH) & self.bucket_mask; + // SAFETY: + // - The caller must uphold the safety contract for `erase` method; + // - `index_before` is guaranteed to be in range due to masking with `self.bucket_mask` + let empty_before = Group::load(self.ctrl(index_before)).match_empty(); + let empty_after = Group::load(self.ctrl(index)).match_empty(); + + // Inserting and searching in the map is performed by two key functions: + // + // - The `find_insert_slot` function that looks up the index of any `EMPTY` or `DELETED` + // slot in a group to be able to insert. If it doesn't find an `EMPTY` or `DELETED` + // slot immediately in the first group, it jumps to the next `Group` looking for it, + // and so on until it has gone through all the groups in the control bytes. + // + // - The `find_inner` function that looks for the index of the desired element by looking + // at all the `FULL` bytes in the group. If it did not find the element right away, and + // there is no `EMPTY` byte in the group, then this means that the `find_insert_slot` + // function may have found a suitable slot in the next group. Therefore, `find_inner` + // jumps further, and if it does not find the desired element and again there is no `EMPTY` + // byte, then it jumps further, and so on. The search stops only if `find_inner` function + // finds the desired element or hits an `EMPTY` slot/byte. + // + // Accordingly, this leads to two consequences: + // + // - The map must have `EMPTY` slots (bytes); + // + // - You can't just mark the byte to be erased as `EMPTY`, because otherwise the `find_inner` + // function may stumble upon an `EMPTY` byte before finding the desired element and stop + // searching. + // + // Thus it is necessary to check all bytes after and before the erased element. If we are in + // a contiguous `Group` of `FULL` or `DELETED` bytes (the number of `FULL` or `DELETED` bytes + // before and after is greater than or equal to `Group::WIDTH`), then we must mark our byte as + // `DELETED` in order for the `find_inner` function to go further. On the other hand, if there + // is at least one `EMPTY` slot in the `Group`, then the `find_inner` function will still stumble + // upon an `EMPTY` byte, so we can safely mark our erased byte as `EMPTY` as well. + // + // Finally, since `index_before == (index.wrapping_sub(Group::WIDTH) & self.bucket_mask) == index` + // and given all of the above, tables smaller than the group width (self.buckets() < Group::WIDTH) + // cannot have `DELETED` bytes. + // + // Note that in this context `leading_zeros` refers to the bytes at the end of a group, while + // `trailing_zeros` refers to the bytes at the beginning of a group. + let ctrl = if empty_before.leading_zeros() + empty_after.trailing_zeros() >= Group::WIDTH { + DELETED + } else { + self.growth_left += 1; + EMPTY + }; + // SAFETY: the caller must uphold the safety contract for `erase` method. + self.set_ctrl(index, ctrl); + self.items -= 1; + } +} + +impl<T: Clone, A: Allocator + Clone> Clone for RawTable<T, A> { + fn clone(&self) -> Self { + if self.table.is_empty_singleton() { + Self::new_in(self.alloc.clone()) + } else { + unsafe { + // Avoid `Result::ok_or_else` because it bloats LLVM IR. + // + // SAFETY: This is safe as we are taking the size of an already allocated table + // and therefore сapacity overflow cannot occur, `self.table.buckets()` is power + // of two and all allocator errors will be caught inside `RawTableInner::new_uninitialized`. + let mut new_table = match Self::new_uninitialized( + self.alloc.clone(), + self.table.buckets(), + Fallibility::Infallible, + ) { + Ok(table) => table, + Err(_) => hint::unreachable_unchecked(), + }; + + // Cloning elements may fail (the clone function may panic). But we don't + // need to worry about uninitialized control bits, since: + // 1. The number of items (elements) in the table is zero, which means that + // the control bits will not be readed by Drop function. + // 2. The `clone_from_spec` method will first copy all control bits from + // `self` (thus initializing them). But this will not affect the `Drop` + // function, since the `clone_from_spec` function sets `items` only after + // successfully clonning all elements. + new_table.clone_from_spec(self); + new_table + } + } + } + + fn clone_from(&mut self, source: &Self) { + if source.table.is_empty_singleton() { + let mut old_inner = mem::replace(&mut self.table, RawTableInner::NEW); + unsafe { + // SAFETY: + // 1. We call the function only once; + // 2. We know for sure that `alloc` and `table_layout` matches the [`Allocator`] + // and [`TableLayout`] that were used to allocate this table. + // 3. If any elements' drop function panics, then there will only be a memory leak, + // because we have replaced the inner table with a new one. + old_inner.drop_inner_table::<T, _>(&self.alloc, Self::TABLE_LAYOUT); + } + } else { + unsafe { + // Make sure that if any panics occurs, we clear the table and + // leave it in an empty state. + let mut self_ = guard(self, |self_| { + self_.clear_no_drop(); + }); + + // First, drop all our elements without clearing the control + // bytes. If this panics then the scope guard will clear the + // table, leaking any elements that were not dropped yet. + // + // This leak is unavoidable: we can't try dropping more elements + // since this could lead to another panic and abort the process. + // + // SAFETY: If something gets wrong we clear our table right after + // dropping the elements, so there is no double drop, since `items` + // will be equal to zero. + self_.table.drop_elements::<T>(); + + // If necessary, resize our table to match the source. + if self_.buckets() != source.buckets() { + let new_inner = match RawTableInner::new_uninitialized( + &self_.alloc, + Self::TABLE_LAYOUT, + source.buckets(), + Fallibility::Infallible, + ) { + Ok(table) => table, + Err(_) => hint::unreachable_unchecked(), + }; + // Replace the old inner with new uninitialized one. It's ok, since if something gets + // wrong `ScopeGuard` will initialize all control bytes and leave empty table. + let mut old_inner = mem::replace(&mut self_.table, new_inner); + if !old_inner.is_empty_singleton() { + // SAFETY: + // 1. We have checked that our table is allocated. + // 2. We know for sure that `alloc` and `table_layout` matches + // the [`Allocator`] and [`TableLayout`] that were used to allocate this table. + old_inner.free_buckets(&self_.alloc, Self::TABLE_LAYOUT); + } + } + + // Cloning elements may fail (the clone function may panic), but the `ScopeGuard` + // inside the `clone_from_impl` function will take care of that, dropping all + // cloned elements if necessary. Our `ScopeGuard` will clear the table. + self_.clone_from_spec(source); + + // Disarm the scope guard if cloning was successful. + ScopeGuard::into_inner(self_); + } + } + } +} + +/// Specialization of `clone_from` for `Copy` types +trait RawTableClone { + unsafe fn clone_from_spec(&mut self, source: &Self); +} +impl<T: Clone, A: Allocator + Clone> RawTableClone for RawTable<T, A> { + default_fn! { + #[cfg_attr(feature = "inline-more", inline)] + unsafe fn clone_from_spec(&mut self, source: &Self) { + self.clone_from_impl(source); + } + } +} +#[cfg(feature = "nightly")] +impl<T: Copy, A: Allocator + Clone> RawTableClone for RawTable<T, A> { + #[cfg_attr(feature = "inline-more", inline)] + unsafe fn clone_from_spec(&mut self, source: &Self) { + source + .table + .ctrl(0) + .copy_to_nonoverlapping(self.table.ctrl(0), self.table.num_ctrl_bytes()); + source + .data_start() + .as_ptr() + .copy_to_nonoverlapping(self.data_start().as_ptr(), self.table.buckets()); + + self.table.items = source.table.items; + self.table.growth_left = source.table.growth_left; + } +} + +impl<T: Clone, A: Allocator + Clone> RawTable<T, A> { + /// Common code for clone and clone_from. Assumes: + /// - `self.buckets() == source.buckets()`. + /// - Any existing elements have been dropped. + /// - The control bytes are not initialized yet. + #[cfg_attr(feature = "inline-more", inline)] + unsafe fn clone_from_impl(&mut self, source: &Self) { + // Copy the control bytes unchanged. We do this in a single pass + source + .table + .ctrl(0) + .copy_to_nonoverlapping(self.table.ctrl(0), self.table.num_ctrl_bytes()); + + // The cloning of elements may panic, in which case we need + // to make sure we drop only the elements that have been + // cloned so far. + let mut guard = guard((0, &mut *self), |(index, self_)| { + if T::NEEDS_DROP { + for i in 0..=*index { + if self_.is_bucket_full(i) { + self_.bucket(i).drop(); + } + } + } + }); + + for from in source.iter() { + let index = source.bucket_index(&from); + let to = guard.1.bucket(index); + to.write(from.as_ref().clone()); + + // Update the index in case we need to unwind. + guard.0 = index; + } + + // Successfully cloned all items, no need to clean up. + mem::forget(guard); + + self.table.items = source.table.items; + self.table.growth_left = source.table.growth_left; + } + + /// Variant of `clone_from` to use when a hasher is available. + #[cfg(feature = "raw")] + pub fn clone_from_with_hasher(&mut self, source: &Self, hasher: impl Fn(&T) -> u64) { + // If we have enough capacity in the table, just clear it and insert + // elements one by one. We don't do this if we have the same number of + // buckets as the source since we can just copy the contents directly + // in that case. + if self.table.buckets() != source.table.buckets() + && bucket_mask_to_capacity(self.table.bucket_mask) >= source.len() + { + self.clear(); + + let mut guard_self = guard(&mut *self, |self_| { + // Clear the partially copied table if a panic occurs, otherwise + // items and growth_left will be out of sync with the contents + // of the table. + self_.clear(); + }); + + unsafe { + for item in source.iter() { + // This may panic. + let item = item.as_ref().clone(); + let hash = hasher(&item); + + // We can use a simpler version of insert() here since: + // - there are no DELETED entries. + // - we know there is enough space in the table. + // - all elements are unique. + let (index, _) = guard_self.table.prepare_insert_slot(hash); + guard_self.bucket(index).write(item); + } + } + + // Successfully cloned all items, no need to clean up. + mem::forget(guard_self); + + self.table.items = source.table.items; + self.table.growth_left -= source.table.items; + } else { + self.clone_from(source); + } + } +} + +impl<T, A: Allocator + Default> Default for RawTable<T, A> { + #[inline] + fn default() -> Self { + Self::new_in(Default::default()) + } +} + +#[cfg(feature = "nightly")] +unsafe impl<#[may_dangle] T, A: Allocator> Drop for RawTable<T, A> { + #[cfg_attr(feature = "inline-more", inline)] + fn drop(&mut self) { + unsafe { + // SAFETY: + // 1. We call the function only once; + // 2. We know for sure that `alloc` and `table_layout` matches the [`Allocator`] + // and [`TableLayout`] that were used to allocate this table. + // 3. If the drop function of any elements fails, then only a memory leak will occur, + // and we don't care because we are inside the `Drop` function of the `RawTable`, + // so there won't be any table left in an inconsistent state. + self.table + .drop_inner_table::<T, _>(&self.alloc, Self::TABLE_LAYOUT); + } + } +} +#[cfg(not(feature = "nightly"))] +impl<T, A: Allocator> Drop for RawTable<T, A> { + #[cfg_attr(feature = "inline-more", inline)] + fn drop(&mut self) { + unsafe { + // SAFETY: + // 1. We call the function only once; + // 2. We know for sure that `alloc` and `table_layout` matches the [`Allocator`] + // and [`TableLayout`] that were used to allocate this table. + // 3. If the drop function of any elements fails, then only a memory leak will occur, + // and we don't care because we are inside the `Drop` function of the `RawTable`, + // so there won't be any table left in an inconsistent state. + self.table + .drop_inner_table::<T, _>(&self.alloc, Self::TABLE_LAYOUT); + } + } +} + +impl<T, A: Allocator> IntoIterator for RawTable<T, A> { + type Item = T; + type IntoIter = RawIntoIter<T, A>; + + #[cfg_attr(feature = "inline-more", inline)] + fn into_iter(self) -> RawIntoIter<T, A> { + unsafe { + let iter = self.iter(); + self.into_iter_from(iter) + } + } +} + +/// Iterator over a sub-range of a table. Unlike `RawIter` this iterator does +/// not track an item count. +pub(crate) struct RawIterRange<T> { + // Mask of full buckets in the current group. Bits are cleared from this + // mask as each element is processed. + current_group: BitMaskIter, + + // Pointer to the buckets for the current group. + data: Bucket<T>, + + // Pointer to the next group of control bytes, + // Must be aligned to the group size. + next_ctrl: *const u8, + + // Pointer one past the last control byte of this range. + end: *const u8, +} + +impl<T> RawIterRange<T> { + /// Returns a `RawIterRange` covering a subset of a table. + /// + /// # Safety + /// + /// If any of the following conditions are violated, the result is + /// [`undefined behavior`]: + /// + /// * `ctrl` must be [valid] for reads, i.e. table outlives the `RawIterRange`; + /// + /// * `ctrl` must be properly aligned to the group size (Group::WIDTH); + /// + /// * `ctrl` must point to the array of properly initialized control bytes; + /// + /// * `data` must be the [`Bucket`] at the `ctrl` index in the table; + /// + /// * the value of `len` must be less than or equal to the number of table buckets, + /// and the returned value of `ctrl.as_ptr().add(len).offset_from(ctrl.as_ptr())` + /// must be positive. + /// + /// * The `ctrl.add(len)` pointer must be either in bounds or one + /// byte past the end of the same [allocated table]. + /// + /// * The `len` must be a power of two. + /// + /// [`undefined behavior`]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html + #[cfg_attr(feature = "inline-more", inline)] + unsafe fn new(ctrl: *const u8, data: Bucket<T>, len: usize) -> Self { + debug_assert_ne!(len, 0); + debug_assert_eq!(ctrl as usize % Group::WIDTH, 0); + // SAFETY: The caller must uphold the safety rules for the [`RawIterRange::new`] + let end = ctrl.add(len); + + // Load the first group and advance ctrl to point to the next group + // SAFETY: The caller must uphold the safety rules for the [`RawIterRange::new`] + let current_group = Group::load_aligned(ctrl).match_full(); + let next_ctrl = ctrl.add(Group::WIDTH); + + Self { + current_group: current_group.into_iter(), + data, + next_ctrl, + end, + } + } + + /// Splits a `RawIterRange` into two halves. + /// + /// Returns `None` if the remaining range is smaller than or equal to the + /// group width. + #[cfg_attr(feature = "inline-more", inline)] + #[cfg(feature = "rayon")] + pub(crate) fn split(mut self) -> (Self, Option<RawIterRange<T>>) { + unsafe { + if self.end <= self.next_ctrl { + // Nothing to split if the group that we are current processing + // is the last one. + (self, None) + } else { + // len is the remaining number of elements after the group that + // we are currently processing. It must be a multiple of the + // group size (small tables are caught by the check above). + let len = offset_from(self.end, self.next_ctrl); + debug_assert_eq!(len % Group::WIDTH, 0); + + // Split the remaining elements into two halves, but round the + // midpoint down in case there is an odd number of groups + // remaining. This ensures that: + // - The tail is at least 1 group long. + // - The split is roughly even considering we still have the + // current group to process. + let mid = (len / 2) & !(Group::WIDTH - 1); + + let tail = Self::new( + self.next_ctrl.add(mid), + self.data.next_n(Group::WIDTH).next_n(mid), + len - mid, + ); + debug_assert_eq!( + self.data.next_n(Group::WIDTH).next_n(mid).ptr, + tail.data.ptr + ); + debug_assert_eq!(self.end, tail.end); + self.end = self.next_ctrl.add(mid); + debug_assert_eq!(self.end.add(Group::WIDTH), tail.next_ctrl); + (self, Some(tail)) + } + } + } + + /// # Safety + /// If DO_CHECK_PTR_RANGE is false, caller must ensure that we never try to iterate + /// after yielding all elements. + #[cfg_attr(feature = "inline-more", inline)] + unsafe fn next_impl<const DO_CHECK_PTR_RANGE: bool>(&mut self) -> Option<Bucket<T>> { + loop { + if let Some(index) = self.current_group.next() { + return Some(self.data.next_n(index)); + } + + if DO_CHECK_PTR_RANGE && self.next_ctrl >= self.end { + return None; + } + + // We might read past self.end up to the next group boundary, + // but this is fine because it only occurs on tables smaller + // than the group size where the trailing control bytes are all + // EMPTY. On larger tables self.end is guaranteed to be aligned + // to the group size (since tables are power-of-two sized). + self.current_group = Group::load_aligned(self.next_ctrl).match_full().into_iter(); + self.data = self.data.next_n(Group::WIDTH); + self.next_ctrl = self.next_ctrl.add(Group::WIDTH); + } + } +} + +// We make raw iterators unconditionally Send and Sync, and let the PhantomData +// in the actual iterator implementations determine the real Send/Sync bounds. +unsafe impl<T> Send for RawIterRange<T> {} +unsafe impl<T> Sync for RawIterRange<T> {} + +impl<T> Clone for RawIterRange<T> { + #[cfg_attr(feature = "inline-more", inline)] + fn clone(&self) -> Self { + Self { + data: self.data.clone(), + next_ctrl: self.next_ctrl, + current_group: self.current_group, + end: self.end, + } + } +} + +impl<T> Iterator for RawIterRange<T> { + type Item = Bucket<T>; + + #[cfg_attr(feature = "inline-more", inline)] + fn next(&mut self) -> Option<Bucket<T>> { + unsafe { + // SAFETY: We set checker flag to true. + self.next_impl::<true>() + } + } + + #[inline] + fn size_hint(&self) -> (usize, Option<usize>) { + // We don't have an item count, so just guess based on the range size. + let remaining_buckets = if self.end > self.next_ctrl { + unsafe { offset_from(self.end, self.next_ctrl) } + } else { + 0 + }; + + // Add a group width to include the group we are currently processing. + (0, Some(Group::WIDTH + remaining_buckets)) + } +} + +impl<T> FusedIterator for RawIterRange<T> {} + +/// Iterator which returns a raw pointer to every full bucket in the table. +/// +/// For maximum flexibility this iterator is not bound by a lifetime, but you +/// must observe several rules when using it: +/// - You must not free the hash table while iterating (including via growing/shrinking). +/// - It is fine to erase a bucket that has been yielded by the iterator. +/// - Erasing a bucket that has not yet been yielded by the iterator may still +/// result in the iterator yielding that bucket (unless `reflect_remove` is called). +/// - It is unspecified whether an element inserted after the iterator was +/// created will be yielded by that iterator (unless `reflect_insert` is called). +/// - The order in which the iterator yields bucket is unspecified and may +/// change in the future. +pub struct RawIter<T> { + pub(crate) iter: RawIterRange<T>, + items: usize, +} + +impl<T> RawIter<T> { + /// Refresh the iterator so that it reflects a removal from the given bucket. + /// + /// For the iterator to remain valid, this method must be called once + /// for each removed bucket before `next` is called again. + /// + /// This method should be called _before_ the removal is made. It is not necessary to call this + /// method if you are removing an item that this iterator yielded in the past. + #[cfg(feature = "raw")] + pub unsafe fn reflect_remove(&mut self, b: &Bucket<T>) { + self.reflect_toggle_full(b, false); + } + + /// Refresh the iterator so that it reflects an insertion into the given bucket. + /// + /// For the iterator to remain valid, this method must be called once + /// for each insert before `next` is called again. + /// + /// This method does not guarantee that an insertion of a bucket with a greater + /// index than the last one yielded will be reflected in the iterator. + /// + /// This method should be called _after_ the given insert is made. + #[cfg(feature = "raw")] + pub unsafe fn reflect_insert(&mut self, b: &Bucket<T>) { + self.reflect_toggle_full(b, true); + } + + /// Refresh the iterator so that it reflects a change to the state of the given bucket. + #[cfg(feature = "raw")] + unsafe fn reflect_toggle_full(&mut self, b: &Bucket<T>, is_insert: bool) { + if b.as_ptr() > self.iter.data.as_ptr() { + // The iterator has already passed the bucket's group. + // So the toggle isn't relevant to this iterator. + return; + } + + if self.iter.next_ctrl < self.iter.end + && b.as_ptr() <= self.iter.data.next_n(Group::WIDTH).as_ptr() + { + // The iterator has not yet reached the bucket's group. + // We don't need to reload anything, but we do need to adjust the item count. + + if cfg!(debug_assertions) { + // Double-check that the user isn't lying to us by checking the bucket state. + // To do that, we need to find its control byte. We know that self.iter.data is + // at self.iter.next_ctrl - Group::WIDTH, so we work from there: + let offset = offset_from(self.iter.data.as_ptr(), b.as_ptr()); + let ctrl = self.iter.next_ctrl.sub(Group::WIDTH).add(offset); + // This method should be called _before_ a removal, or _after_ an insert, + // so in both cases the ctrl byte should indicate that the bucket is full. + assert!(is_full(*ctrl)); + } + + if is_insert { + self.items += 1; + } else { + self.items -= 1; + } + + return; + } + + // The iterator is at the bucket group that the toggled bucket is in. + // We need to do two things: + // + // - Determine if the iterator already yielded the toggled bucket. + // If it did, we're done. + // - Otherwise, update the iterator cached group so that it won't + // yield a to-be-removed bucket, or _will_ yield a to-be-added bucket. + // We'll also need to update the item count accordingly. + if let Some(index) = self.iter.current_group.0.lowest_set_bit() { + let next_bucket = self.iter.data.next_n(index); + if b.as_ptr() > next_bucket.as_ptr() { + // The toggled bucket is "before" the bucket the iterator would yield next. We + // therefore don't need to do anything --- the iterator has already passed the + // bucket in question. + // + // The item count must already be correct, since a removal or insert "prior" to + // the iterator's position wouldn't affect the item count. + } else { + // The removed bucket is an upcoming bucket. We need to make sure it does _not_ + // get yielded, and also that it's no longer included in the item count. + // + // NOTE: We can't just reload the group here, both since that might reflect + // inserts we've already passed, and because that might inadvertently unset the + // bits for _other_ removals. If we do that, we'd have to also decrement the + // item count for those other bits that we unset. But the presumably subsequent + // call to reflect for those buckets might _also_ decrement the item count. + // Instead, we _just_ flip the bit for the particular bucket the caller asked + // us to reflect. + let our_bit = offset_from(self.iter.data.as_ptr(), b.as_ptr()); + let was_full = self.iter.current_group.flip(our_bit); + debug_assert_ne!(was_full, is_insert); + + if is_insert { + self.items += 1; + } else { + self.items -= 1; + } + + if cfg!(debug_assertions) { + if b.as_ptr() == next_bucket.as_ptr() { + // The removed bucket should no longer be next + debug_assert_ne!(self.iter.current_group.0.lowest_set_bit(), Some(index)); + } else { + // We should not have changed what bucket comes next. + debug_assert_eq!(self.iter.current_group.0.lowest_set_bit(), Some(index)); + } + } + } + } else { + // We must have already iterated past the removed item. + } + } + + unsafe fn drop_elements(&mut self) { + if T::NEEDS_DROP && self.items != 0 { + for item in self { + item.drop(); + } + } + } +} + +impl<T> Clone for RawIter<T> { + #[cfg_attr(feature = "inline-more", inline)] + fn clone(&self) -> Self { + Self { + iter: self.iter.clone(), + items: self.items, + } + } +} + +impl<T> Iterator for RawIter<T> { + type Item = Bucket<T>; + + #[cfg_attr(feature = "inline-more", inline)] + fn next(&mut self) -> Option<Bucket<T>> { + // Inner iterator iterates over buckets + // so it can do unnecessary work if we already yielded all items. + if self.items == 0 { + return None; + } + + let nxt = unsafe { + // SAFETY: We check number of items to yield using `items` field. + self.iter.next_impl::<false>() + }; + + debug_assert!(nxt.is_some()); + self.items -= 1; + + nxt + } + + #[inline] + fn size_hint(&self) -> (usize, Option<usize>) { + (self.items, Some(self.items)) + } +} + +impl<T> ExactSizeIterator for RawIter<T> {} +impl<T> FusedIterator for RawIter<T> {} + +/// Iterator which returns an index of every full bucket in the table. +/// +/// For maximum flexibility this iterator is not bound by a lifetime, but you +/// must observe several rules when using it: +/// - You must not free the hash table while iterating (including via growing/shrinking). +/// - It is fine to erase a bucket that has been yielded by the iterator. +/// - Erasing a bucket that has not yet been yielded by the iterator may still +/// result in the iterator yielding index of that bucket. +/// - It is unspecified whether an element inserted after the iterator was +/// created will be yielded by that iterator. +/// - The order in which the iterator yields indices of the buckets is unspecified +/// and may change in the future. +pub(crate) struct FullBucketsIndices { + // Mask of full buckets in the current group. Bits are cleared from this + // mask as each element is processed. + current_group: BitMaskIter, + + // Initial value of the bytes' indices of the current group (relative + // to the start of the control bytes). + group_first_index: usize, + + // Pointer to the current group of control bytes, + // Must be aligned to the group size (Group::WIDTH). + ctrl: NonNull<u8>, + + // Number of elements in the table. + items: usize, +} + +impl FullBucketsIndices { + /// Advances the iterator and returns the next value. + /// + /// # Safety + /// + /// If any of the following conditions are violated, the result is + /// [`Undefined Behavior`]: + /// + /// * The [`RawTableInner`] / [`RawTable`] must be alive and not moved, + /// i.e. table outlives the `FullBucketsIndices`; + /// + /// * It never tries to iterate after getting all elements. + /// + /// [`Undefined Behavior`]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html + #[inline(always)] + unsafe fn next_impl(&mut self) -> Option<usize> { + loop { + if let Some(index) = self.current_group.next() { + // The returned `self.group_first_index + index` will always + // be in the range `0..self.buckets()`. See explanation below. + return Some(self.group_first_index + index); + } + + // SAFETY: The caller of this function ensures that: + // + // 1. It never tries to iterate after getting all the elements; + // 2. The table is alive and did not moved; + // 3. The first `self.ctrl` pointed to the start of the array of control bytes. + // + // Taking the above into account, we always stay within the bounds, because: + // + // 1. For tables smaller than the group width (self.buckets() <= Group::WIDTH), + // we will never end up in the given branch, since we should have already + // yielded all the elements of the table. + // + // 2. For tables larger than the group width. The the number of buckets is a + // power of two (2 ^ n), Group::WIDTH is also power of two (2 ^ k). Sinse + // `(2 ^ n) > (2 ^ k)`, than `(2 ^ n) % (2 ^ k) = 0`. As we start from the + // the start of the array of control bytes, and never try to iterate after + // getting all the elements, the last `self.ctrl` will be equal to + // the `self.buckets() - Group::WIDTH`, so `self.current_group.next()` + // will always contains indices within the range `0..Group::WIDTH`, + // and subsequent `self.group_first_index + index` will always return a + // number less than `self.buckets()`. + self.ctrl = NonNull::new_unchecked(self.ctrl.as_ptr().add(Group::WIDTH)); + + // SAFETY: See explanation above. + self.current_group = Group::load_aligned(self.ctrl.as_ptr()) + .match_full() + .into_iter(); + self.group_first_index += Group::WIDTH; + } + } +} + +impl Iterator for FullBucketsIndices { + type Item = usize; + + /// Advances the iterator and returns the next value. It is up to + /// the caller to ensure that the `RawTable` outlives the `FullBucketsIndices`, + /// because we cannot make the `next` method unsafe. + #[inline(always)] + fn next(&mut self) -> Option<usize> { + // Return if we already yielded all items. + if self.items == 0 { + return None; + } + + let nxt = unsafe { + // SAFETY: + // 1. We check number of items to yield using `items` field. + // 2. The caller ensures that the table is alive and has not moved. + self.next_impl() + }; + + debug_assert!(nxt.is_some()); + self.items -= 1; + + nxt + } + + #[inline(always)] + fn size_hint(&self) -> (usize, Option<usize>) { + (self.items, Some(self.items)) + } +} + +impl ExactSizeIterator for FullBucketsIndices {} +impl FusedIterator for FullBucketsIndices {} + +/// Iterator which consumes a table and returns elements. +pub struct RawIntoIter<T, A: Allocator = Global> { + iter: RawIter<T>, + allocation: Option<(NonNull<u8>, Layout, A)>, + marker: PhantomData<T>, +} + +impl<T, A: Allocator> RawIntoIter<T, A> { + #[cfg_attr(feature = "inline-more", inline)] + pub fn iter(&self) -> RawIter<T> { + self.iter.clone() + } +} + +unsafe impl<T, A: Allocator> Send for RawIntoIter<T, A> +where + T: Send, + A: Send, +{ +} +unsafe impl<T, A: Allocator> Sync for RawIntoIter<T, A> +where + T: Sync, + A: Sync, +{ +} + +#[cfg(feature = "nightly")] +unsafe impl<#[may_dangle] T, A: Allocator> Drop for RawIntoIter<T, A> { + #[cfg_attr(feature = "inline-more", inline)] + fn drop(&mut self) { + unsafe { + // Drop all remaining elements + self.iter.drop_elements(); + + // Free the table + if let Some((ptr, layout, ref alloc)) = self.allocation { + alloc.deallocate(ptr, layout); + } + } + } +} +#[cfg(not(feature = "nightly"))] +impl<T, A: Allocator> Drop for RawIntoIter<T, A> { + #[cfg_attr(feature = "inline-more", inline)] + fn drop(&mut self) { + unsafe { + // Drop all remaining elements + self.iter.drop_elements(); + + // Free the table + if let Some((ptr, layout, ref alloc)) = self.allocation { + alloc.deallocate(ptr, layout); + } + } + } +} + +impl<T, A: Allocator> Iterator for RawIntoIter<T, A> { + type Item = T; + + #[cfg_attr(feature = "inline-more", inline)] + fn next(&mut self) -> Option<T> { + unsafe { Some(self.iter.next()?.read()) } + } + + #[inline] + fn size_hint(&self) -> (usize, Option<usize>) { + self.iter.size_hint() + } +} + +impl<T, A: Allocator> ExactSizeIterator for RawIntoIter<T, A> {} +impl<T, A: Allocator> FusedIterator for RawIntoIter<T, A> {} + +/// Iterator which consumes elements without freeing the table storage. +pub struct RawDrain<'a, T, A: Allocator = Global> { + iter: RawIter<T>, + + // The table is moved into the iterator for the duration of the drain. This + // ensures that an empty table is left if the drain iterator is leaked + // without dropping. + table: RawTableInner, + orig_table: NonNull<RawTableInner>, + + // We don't use a &'a mut RawTable<T> because we want RawDrain to be + // covariant over T. + marker: PhantomData<&'a RawTable<T, A>>, +} + +impl<T, A: Allocator> RawDrain<'_, T, A> { + #[cfg_attr(feature = "inline-more", inline)] + pub fn iter(&self) -> RawIter<T> { + self.iter.clone() + } +} + +unsafe impl<T, A: Allocator> Send for RawDrain<'_, T, A> +where + T: Send, + A: Send, +{ +} +unsafe impl<T, A: Allocator> Sync for RawDrain<'_, T, A> +where + T: Sync, + A: Sync, +{ +} + +impl<T, A: Allocator> Drop for RawDrain<'_, T, A> { + #[cfg_attr(feature = "inline-more", inline)] + fn drop(&mut self) { + unsafe { + // Drop all remaining elements. Note that this may panic. + self.iter.drop_elements(); + + // Reset the contents of the table now that all elements have been + // dropped. + self.table.clear_no_drop(); + + // Move the now empty table back to its original location. + self.orig_table + .as_ptr() + .copy_from_nonoverlapping(&self.table, 1); + } + } +} + +impl<T, A: Allocator> Iterator for RawDrain<'_, T, A> { + type Item = T; + + #[cfg_attr(feature = "inline-more", inline)] + fn next(&mut self) -> Option<T> { + unsafe { + let item = self.iter.next()?; + Some(item.read()) + } + } + + #[inline] + fn size_hint(&self) -> (usize, Option<usize>) { + self.iter.size_hint() + } +} + +impl<T, A: Allocator> ExactSizeIterator for RawDrain<'_, T, A> {} +impl<T, A: Allocator> FusedIterator for RawDrain<'_, T, A> {} + +/// Iterator over occupied buckets that could match a given hash. +/// +/// `RawTable` only stores 7 bits of the hash value, so this iterator may return +/// items that have a hash value different than the one provided. You should +/// always validate the returned values before using them. +/// +/// For maximum flexibility this iterator is not bound by a lifetime, but you +/// must observe several rules when using it: +/// - You must not free the hash table while iterating (including via growing/shrinking). +/// - It is fine to erase a bucket that has been yielded by the iterator. +/// - Erasing a bucket that has not yet been yielded by the iterator may still +/// result in the iterator yielding that bucket. +/// - It is unspecified whether an element inserted after the iterator was +/// created will be yielded by that iterator. +/// - The order in which the iterator yields buckets is unspecified and may +/// change in the future. +pub struct RawIterHash<T> { + inner: RawIterHashInner, + _marker: PhantomData<T>, +} + +struct RawIterHashInner { + // See `RawTableInner`'s corresponding fields for details. + // We can't store a `*const RawTableInner` as it would get + // invalidated by the user calling `&mut` methods on `RawTable`. + bucket_mask: usize, + ctrl: NonNull<u8>, + + // The top 7 bits of the hash. + h2_hash: u8, + + // The sequence of groups to probe in the search. + probe_seq: ProbeSeq, + + group: Group, + + // The elements within the group with a matching h2-hash. + bitmask: BitMaskIter, +} + +impl<T> RawIterHash<T> { + #[cfg_attr(feature = "inline-more", inline)] + #[cfg(feature = "raw")] + unsafe fn new<A: Allocator>(table: &RawTable<T, A>, hash: u64) -> Self { + RawIterHash { + inner: RawIterHashInner::new(&table.table, hash), + _marker: PhantomData, + } + } +} +impl RawIterHashInner { + #[cfg_attr(feature = "inline-more", inline)] + #[cfg(feature = "raw")] + unsafe fn new(table: &RawTableInner, hash: u64) -> Self { + let h2_hash = h2(hash); + let probe_seq = table.probe_seq(hash); + let group = Group::load(table.ctrl(probe_seq.pos)); + let bitmask = group.match_byte(h2_hash).into_iter(); + + RawIterHashInner { + bucket_mask: table.bucket_mask, + ctrl: table.ctrl, + h2_hash, + probe_seq, + group, + bitmask, + } + } +} + +impl<T> Iterator for RawIterHash<T> { + type Item = Bucket<T>; + + fn next(&mut self) -> Option<Bucket<T>> { + unsafe { + match self.inner.next() { + Some(index) => { + // Can't use `RawTable::bucket` here as we don't have + // an actual `RawTable` reference to use. + debug_assert!(index <= self.inner.bucket_mask); + let bucket = Bucket::from_base_index(self.inner.ctrl.cast(), index); + Some(bucket) + } + None => None, + } + } + } +} + +impl Iterator for RawIterHashInner { + type Item = usize; + + fn next(&mut self) -> Option<Self::Item> { + unsafe { + loop { + if let Some(bit) = self.bitmask.next() { + let index = (self.probe_seq.pos + bit) & self.bucket_mask; + return Some(index); + } + if likely(self.group.match_empty().any_bit_set()) { + return None; + } + self.probe_seq.move_next(self.bucket_mask); + + // Can't use `RawTableInner::ctrl` here as we don't have + // an actual `RawTableInner` reference to use. + let index = self.probe_seq.pos; + debug_assert!(index < self.bucket_mask + 1 + Group::WIDTH); + let group_ctrl = self.ctrl.as_ptr().add(index); + + self.group = Group::load(group_ctrl); + self.bitmask = self.group.match_byte(self.h2_hash).into_iter(); + } + } + } +} + +#[cfg(test)] +mod test_map { + use super::*; + + fn rehash_in_place<T>(table: &mut RawTable<T>, hasher: impl Fn(&T) -> u64) { + unsafe { + table.table.rehash_in_place( + &|table, index| hasher(table.bucket::<T>(index).as_ref()), + mem::size_of::<T>(), + if mem::needs_drop::<T>() { + Some(mem::transmute(ptr::drop_in_place::<T> as unsafe fn(*mut T))) + } else { + None + }, + ); + } + } + + #[test] + fn rehash() { + let mut table = RawTable::new(); + let hasher = |i: &u64| *i; + for i in 0..100 { + table.insert(i, i, hasher); + } + + for i in 0..100 { + unsafe { + assert_eq!(table.find(i, |x| *x == i).map(|b| b.read()), Some(i)); + } + assert!(table.find(i + 100, |x| *x == i + 100).is_none()); + } + + rehash_in_place(&mut table, hasher); + + for i in 0..100 { + unsafe { + assert_eq!(table.find(i, |x| *x == i).map(|b| b.read()), Some(i)); + } + assert!(table.find(i + 100, |x| *x == i + 100).is_none()); + } + } + + /// CHECKING THAT WE ARE NOT TRYING TO READ THE MEMORY OF + /// AN UNINITIALIZED TABLE DURING THE DROP + #[test] + fn test_drop_uninitialized() { + use ::alloc::vec::Vec; + + let table = unsafe { + // SAFETY: The `buckets` is power of two and we're not + // trying to actually use the returned RawTable. + RawTable::<(u64, Vec<i32>)>::new_uninitialized(Global, 8, Fallibility::Infallible) + .unwrap() + }; + drop(table); + } + + /// CHECKING THAT WE DON'T TRY TO DROP DATA IF THE `ITEMS` + /// ARE ZERO, EVEN IF WE HAVE `FULL` CONTROL BYTES. + #[test] + fn test_drop_zero_items() { + use ::alloc::vec::Vec; + unsafe { + // SAFETY: The `buckets` is power of two and we're not + // trying to actually use the returned RawTable. + let table = + RawTable::<(u64, Vec<i32>)>::new_uninitialized(Global, 8, Fallibility::Infallible) + .unwrap(); + + // WE SIMULATE, AS IT WERE, A FULL TABLE. + + // SAFETY: We checked that the table is allocated and therefore the table already has + // `self.bucket_mask + 1 + Group::WIDTH` number of control bytes (see TableLayout::calculate_layout_for) + // so writing `table.table.num_ctrl_bytes() == bucket_mask + 1 + Group::WIDTH` bytes is safe. + table + .table + .ctrl(0) + .write_bytes(EMPTY, table.table.num_ctrl_bytes()); + + // SAFETY: table.capacity() is guaranteed to be smaller than table.buckets() + table.table.ctrl(0).write_bytes(0, table.capacity()); + + // Fix up the trailing control bytes. See the comments in set_ctrl + // for the handling of tables smaller than the group width. + if table.buckets() < Group::WIDTH { + // SAFETY: We have `self.bucket_mask + 1 + Group::WIDTH` number of control bytes, + // so copying `self.buckets() == self.bucket_mask + 1` bytes with offset equal to + // `Group::WIDTH` is safe + table + .table + .ctrl(0) + .copy_to(table.table.ctrl(Group::WIDTH), table.table.buckets()); + } else { + // SAFETY: We have `self.bucket_mask + 1 + Group::WIDTH` number of + // control bytes,so copying `Group::WIDTH` bytes with offset equal + // to `self.buckets() == self.bucket_mask + 1` is safe + table + .table + .ctrl(0) + .copy_to(table.table.ctrl(table.table.buckets()), Group::WIDTH); + } + drop(table); + } + } + + /// CHECKING THAT WE DON'T TRY TO DROP DATA IF THE `ITEMS` + /// ARE ZERO, EVEN IF WE HAVE `FULL` CONTROL BYTES. + #[test] + fn test_catch_panic_clone_from() { + use ::alloc::sync::Arc; + use ::alloc::vec::Vec; + use allocator_api2::alloc::{AllocError, Allocator, Global}; + use core::sync::atomic::{AtomicI8, Ordering}; + use std::thread; + + struct MyAllocInner { + drop_count: Arc<AtomicI8>, + } + + #[derive(Clone)] + struct MyAlloc { + _inner: Arc<MyAllocInner>, + } + + impl Drop for MyAllocInner { + fn drop(&mut self) { + println!("MyAlloc freed."); + self.drop_count.fetch_sub(1, Ordering::SeqCst); + } + } + + unsafe impl Allocator for MyAlloc { + fn allocate(&self, layout: Layout) -> std::result::Result<NonNull<[u8]>, AllocError> { + let g = Global; + g.allocate(layout) + } + + unsafe fn deallocate(&self, ptr: NonNull<u8>, layout: Layout) { + let g = Global; + g.deallocate(ptr, layout) + } + } + + const DISARMED: bool = false; + const ARMED: bool = true; + + struct CheckedCloneDrop { + panic_in_clone: bool, + dropped: bool, + need_drop: Vec<u64>, + } + + impl Clone for CheckedCloneDrop { + fn clone(&self) -> Self { + if self.panic_in_clone { + panic!("panic in clone") + } + Self { + panic_in_clone: self.panic_in_clone, + dropped: self.dropped, + need_drop: self.need_drop.clone(), + } + } + } + + impl Drop for CheckedCloneDrop { + fn drop(&mut self) { + if self.dropped { + panic!("double drop"); + } + self.dropped = true; + } + } + + let dropped: Arc<AtomicI8> = Arc::new(AtomicI8::new(2)); + + let mut table = RawTable::new_in(MyAlloc { + _inner: Arc::new(MyAllocInner { + drop_count: dropped.clone(), + }), + }); + + for (idx, panic_in_clone) in core::iter::repeat(DISARMED).take(7).enumerate() { + let idx = idx as u64; + table.insert( + idx, + ( + idx, + CheckedCloneDrop { + panic_in_clone, + dropped: false, + need_drop: vec![idx], + }, + ), + |(k, _)| *k, + ); + } + + assert_eq!(table.len(), 7); + + thread::scope(|s| { + let result = s.spawn(|| { + let armed_flags = [ + DISARMED, DISARMED, ARMED, DISARMED, DISARMED, DISARMED, DISARMED, + ]; + let mut scope_table = RawTable::new_in(MyAlloc { + _inner: Arc::new(MyAllocInner { + drop_count: dropped.clone(), + }), + }); + for (idx, &panic_in_clone) in armed_flags.iter().enumerate() { + let idx = idx as u64; + scope_table.insert( + idx, + ( + idx, + CheckedCloneDrop { + panic_in_clone, + dropped: false, + need_drop: vec![idx + 100], + }, + ), + |(k, _)| *k, + ); + } + table.clone_from(&scope_table); + }); + assert!(result.join().is_err()); + }); + + // Let's check that all iterators work fine and do not return elements + // (especially `RawIterRange`, which does not depend on the number of + // elements in the table, but looks directly at the control bytes) + // + // SAFETY: We know for sure that `RawTable` will outlive + // the returned `RawIter / RawIterRange` iterator. + assert_eq!(table.len(), 0); + assert_eq!(unsafe { table.iter().count() }, 0); + assert_eq!(unsafe { table.iter().iter.count() }, 0); + + for idx in 0..table.buckets() { + let idx = idx as u64; + assert!( + table.find(idx, |(k, _)| *k == idx).is_none(), + "Index: {idx}" + ); + } + + // All allocator clones should already be dropped. + assert_eq!(dropped.load(Ordering::SeqCst), 1); + } +} diff --git a/vendor/hashbrown/src/raw/neon.rs b/vendor/hashbrown/src/raw/neon.rs new file mode 100644 index 0000000..44e82d5 --- /dev/null +++ b/vendor/hashbrown/src/raw/neon.rs @@ -0,0 +1,124 @@ +use super::bitmask::BitMask; +use super::EMPTY; +use core::arch::aarch64 as neon; +use core::mem; +use core::num::NonZeroU64; + +pub(crate) type BitMaskWord = u64; +pub(crate) type NonZeroBitMaskWord = NonZeroU64; +pub(crate) const BITMASK_STRIDE: usize = 8; +pub(crate) const BITMASK_MASK: BitMaskWord = !0; +pub(crate) const BITMASK_ITER_MASK: BitMaskWord = 0x8080_8080_8080_8080; + +/// Abstraction over a group of control bytes which can be scanned in +/// parallel. +/// +/// This implementation uses a 64-bit NEON value. +#[derive(Copy, Clone)] +pub(crate) struct Group(neon::uint8x8_t); + +#[allow(clippy::use_self)] +impl Group { + /// Number of bytes in the group. + pub(crate) const WIDTH: usize = mem::size_of::<Self>(); + + /// Returns a full group of empty bytes, suitable for use as the initial + /// value for an empty hash table. + /// + /// This is guaranteed to be aligned to the group size. + #[inline] + pub(crate) const fn static_empty() -> &'static [u8; Group::WIDTH] { + #[repr(C)] + struct AlignedBytes { + _align: [Group; 0], + bytes: [u8; Group::WIDTH], + } + const ALIGNED_BYTES: AlignedBytes = AlignedBytes { + _align: [], + bytes: [EMPTY; Group::WIDTH], + }; + &ALIGNED_BYTES.bytes + } + + /// Loads a group of bytes starting at the given address. + #[inline] + #[allow(clippy::cast_ptr_alignment)] // unaligned load + pub(crate) unsafe fn load(ptr: *const u8) -> Self { + Group(neon::vld1_u8(ptr)) + } + + /// Loads a group of bytes starting at the given address, which must be + /// aligned to `mem::align_of::<Group>()`. + #[inline] + #[allow(clippy::cast_ptr_alignment)] + pub(crate) unsafe fn load_aligned(ptr: *const u8) -> Self { + // FIXME: use align_offset once it stabilizes + debug_assert_eq!(ptr as usize & (mem::align_of::<Self>() - 1), 0); + Group(neon::vld1_u8(ptr)) + } + + /// Stores the group of bytes to the given address, which must be + /// aligned to `mem::align_of::<Group>()`. + #[inline] + #[allow(clippy::cast_ptr_alignment)] + pub(crate) unsafe fn store_aligned(self, ptr: *mut u8) { + // FIXME: use align_offset once it stabilizes + debug_assert_eq!(ptr as usize & (mem::align_of::<Self>() - 1), 0); + neon::vst1_u8(ptr, self.0); + } + + /// Returns a `BitMask` indicating all bytes in the group which *may* + /// have the given value. + #[inline] + pub(crate) fn match_byte(self, byte: u8) -> BitMask { + unsafe { + let cmp = neon::vceq_u8(self.0, neon::vdup_n_u8(byte)); + BitMask(neon::vget_lane_u64(neon::vreinterpret_u64_u8(cmp), 0)) + } + } + + /// Returns a `BitMask` indicating all bytes in the group which are + /// `EMPTY`. + #[inline] + pub(crate) fn match_empty(self) -> BitMask { + self.match_byte(EMPTY) + } + + /// Returns a `BitMask` indicating all bytes in the group which are + /// `EMPTY` or `DELETED`. + #[inline] + pub(crate) fn match_empty_or_deleted(self) -> BitMask { + unsafe { + let cmp = neon::vcltz_s8(neon::vreinterpret_s8_u8(self.0)); + BitMask(neon::vget_lane_u64(neon::vreinterpret_u64_u8(cmp), 0)) + } + } + + /// Returns a `BitMask` indicating all bytes in the group which are full. + #[inline] + pub(crate) fn match_full(self) -> BitMask { + unsafe { + let cmp = neon::vcgez_s8(neon::vreinterpret_s8_u8(self.0)); + BitMask(neon::vget_lane_u64(neon::vreinterpret_u64_u8(cmp), 0)) + } + } + + /// Performs the following transformation on all bytes in the group: + /// - `EMPTY => EMPTY` + /// - `DELETED => EMPTY` + /// - `FULL => DELETED` + #[inline] + pub(crate) fn convert_special_to_empty_and_full_to_deleted(self) -> Self { + // Map high_bit = 1 (EMPTY or DELETED) to 1111_1111 + // and high_bit = 0 (FULL) to 1000_0000 + // + // Here's this logic expanded to concrete values: + // let special = 0 > byte = 1111_1111 (true) or 0000_0000 (false) + // 1111_1111 | 1000_0000 = 1111_1111 + // 0000_0000 | 1000_0000 = 1000_0000 + unsafe { + let special = neon::vcltz_s8(neon::vreinterpret_s8_u8(self.0)); + Group(neon::vorr_u8(special, neon::vdup_n_u8(0x80))) + } + } +} diff --git a/vendor/hashbrown/src/raw/sse2.rs b/vendor/hashbrown/src/raw/sse2.rs new file mode 100644 index 0000000..956ba5d --- /dev/null +++ b/vendor/hashbrown/src/raw/sse2.rs @@ -0,0 +1,149 @@ +use super::bitmask::BitMask; +use super::EMPTY; +use core::mem; +use core::num::NonZeroU16; + +#[cfg(target_arch = "x86")] +use core::arch::x86; +#[cfg(target_arch = "x86_64")] +use core::arch::x86_64 as x86; + +pub(crate) type BitMaskWord = u16; +pub(crate) type NonZeroBitMaskWord = NonZeroU16; +pub(crate) const BITMASK_STRIDE: usize = 1; +pub(crate) const BITMASK_MASK: BitMaskWord = 0xffff; +pub(crate) const BITMASK_ITER_MASK: BitMaskWord = !0; + +/// Abstraction over a group of control bytes which can be scanned in +/// parallel. +/// +/// This implementation uses a 128-bit SSE value. +#[derive(Copy, Clone)] +pub(crate) struct Group(x86::__m128i); + +// FIXME: https://github.com/rust-lang/rust-clippy/issues/3859 +#[allow(clippy::use_self)] +impl Group { + /// Number of bytes in the group. + pub(crate) const WIDTH: usize = mem::size_of::<Self>(); + + /// Returns a full group of empty bytes, suitable for use as the initial + /// value for an empty hash table. + /// + /// This is guaranteed to be aligned to the group size. + #[inline] + #[allow(clippy::items_after_statements)] + pub(crate) const fn static_empty() -> &'static [u8; Group::WIDTH] { + #[repr(C)] + struct AlignedBytes { + _align: [Group; 0], + bytes: [u8; Group::WIDTH], + } + const ALIGNED_BYTES: AlignedBytes = AlignedBytes { + _align: [], + bytes: [EMPTY; Group::WIDTH], + }; + &ALIGNED_BYTES.bytes + } + + /// Loads a group of bytes starting at the given address. + #[inline] + #[allow(clippy::cast_ptr_alignment)] // unaligned load + pub(crate) unsafe fn load(ptr: *const u8) -> Self { + Group(x86::_mm_loadu_si128(ptr.cast())) + } + + /// Loads a group of bytes starting at the given address, which must be + /// aligned to `mem::align_of::<Group>()`. + #[inline] + #[allow(clippy::cast_ptr_alignment)] + pub(crate) unsafe fn load_aligned(ptr: *const u8) -> Self { + // FIXME: use align_offset once it stabilizes + debug_assert_eq!(ptr as usize & (mem::align_of::<Self>() - 1), 0); + Group(x86::_mm_load_si128(ptr.cast())) + } + + /// Stores the group of bytes to the given address, which must be + /// aligned to `mem::align_of::<Group>()`. + #[inline] + #[allow(clippy::cast_ptr_alignment)] + pub(crate) unsafe fn store_aligned(self, ptr: *mut u8) { + // FIXME: use align_offset once it stabilizes + debug_assert_eq!(ptr as usize & (mem::align_of::<Self>() - 1), 0); + x86::_mm_store_si128(ptr.cast(), self.0); + } + + /// Returns a `BitMask` indicating all bytes in the group which have + /// the given value. + #[inline] + pub(crate) fn match_byte(self, byte: u8) -> BitMask { + #[allow( + clippy::cast_possible_wrap, // byte: u8 as i8 + // byte: i32 as u16 + // note: _mm_movemask_epi8 returns a 16-bit mask in a i32, the + // upper 16-bits of the i32 are zeroed: + clippy::cast_sign_loss, + clippy::cast_possible_truncation + )] + unsafe { + let cmp = x86::_mm_cmpeq_epi8(self.0, x86::_mm_set1_epi8(byte as i8)); + BitMask(x86::_mm_movemask_epi8(cmp) as u16) + } + } + + /// Returns a `BitMask` indicating all bytes in the group which are + /// `EMPTY`. + #[inline] + pub(crate) fn match_empty(self) -> BitMask { + self.match_byte(EMPTY) + } + + /// Returns a `BitMask` indicating all bytes in the group which are + /// `EMPTY` or `DELETED`. + #[inline] + pub(crate) fn match_empty_or_deleted(self) -> BitMask { + #[allow( + // byte: i32 as u16 + // note: _mm_movemask_epi8 returns a 16-bit mask in a i32, the + // upper 16-bits of the i32 are zeroed: + clippy::cast_sign_loss, + clippy::cast_possible_truncation + )] + unsafe { + // A byte is EMPTY or DELETED iff the high bit is set + BitMask(x86::_mm_movemask_epi8(self.0) as u16) + } + } + + /// Returns a `BitMask` indicating all bytes in the group which are full. + #[inline] + pub(crate) fn match_full(&self) -> BitMask { + self.match_empty_or_deleted().invert() + } + + /// Performs the following transformation on all bytes in the group: + /// - `EMPTY => EMPTY` + /// - `DELETED => EMPTY` + /// - `FULL => DELETED` + #[inline] + pub(crate) fn convert_special_to_empty_and_full_to_deleted(self) -> Self { + // Map high_bit = 1 (EMPTY or DELETED) to 1111_1111 + // and high_bit = 0 (FULL) to 1000_0000 + // + // Here's this logic expanded to concrete values: + // let special = 0 > byte = 1111_1111 (true) or 0000_0000 (false) + // 1111_1111 | 1000_0000 = 1111_1111 + // 0000_0000 | 1000_0000 = 1000_0000 + #[allow( + clippy::cast_possible_wrap, // byte: 0x80_u8 as i8 + )] + unsafe { + let zero = x86::_mm_setzero_si128(); + let special = x86::_mm_cmpgt_epi8(zero, self.0); + Group(x86::_mm_or_si128( + special, + x86::_mm_set1_epi8(0x80_u8 as i8), + )) + } + } +} diff --git a/vendor/hashbrown/src/rustc_entry.rs b/vendor/hashbrown/src/rustc_entry.rs new file mode 100644 index 0000000..defbd4b --- /dev/null +++ b/vendor/hashbrown/src/rustc_entry.rs @@ -0,0 +1,630 @@ +use self::RustcEntry::*; +use crate::map::{make_hash, Drain, HashMap, IntoIter, Iter, IterMut}; +use crate::raw::{Allocator, Bucket, Global, RawTable}; +use core::fmt::{self, Debug}; +use core::hash::{BuildHasher, Hash}; +use core::mem; + +impl<K, V, S, A> HashMap<K, V, S, A> +where + K: Eq + Hash, + S: BuildHasher, + A: Allocator, +{ + /// Gets the given key's corresponding entry in the map for in-place manipulation. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let mut letters = HashMap::new(); + /// + /// for ch in "a short treatise on fungi".chars() { + /// let counter = letters.rustc_entry(ch).or_insert(0); + /// *counter += 1; + /// } + /// + /// assert_eq!(letters[&'s'], 2); + /// assert_eq!(letters[&'t'], 3); + /// assert_eq!(letters[&'u'], 1); + /// assert_eq!(letters.get(&'y'), None); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn rustc_entry(&mut self, key: K) -> RustcEntry<'_, K, V, A> { + let hash = make_hash(&self.hash_builder, &key); + if let Some(elem) = self.table.find(hash, |q| q.0.eq(&key)) { + RustcEntry::Occupied(RustcOccupiedEntry { + key: Some(key), + elem, + table: &mut self.table, + }) + } else { + // Ideally we would put this in VacantEntry::insert, but Entry is not + // generic over the BuildHasher and adding a generic parameter would be + // a breaking change. + self.reserve(1); + + RustcEntry::Vacant(RustcVacantEntry { + hash, + key, + table: &mut self.table, + }) + } + } +} + +/// A view into a single entry in a map, which may either be vacant or occupied. +/// +/// This `enum` is constructed from the [`rustc_entry`] method on [`HashMap`]. +/// +/// [`HashMap`]: struct.HashMap.html +/// [`rustc_entry`]: struct.HashMap.html#method.rustc_entry +pub enum RustcEntry<'a, K, V, A = Global> +where + A: Allocator, +{ + /// An occupied entry. + Occupied(RustcOccupiedEntry<'a, K, V, A>), + + /// A vacant entry. + Vacant(RustcVacantEntry<'a, K, V, A>), +} + +impl<K: Debug, V: Debug, A: Allocator> Debug for RustcEntry<'_, K, V, A> { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + match *self { + Vacant(ref v) => f.debug_tuple("Entry").field(v).finish(), + Occupied(ref o) => f.debug_tuple("Entry").field(o).finish(), + } + } +} + +/// A view into an occupied entry in a `HashMap`. +/// It is part of the [`RustcEntry`] enum. +/// +/// [`RustcEntry`]: enum.RustcEntry.html +pub struct RustcOccupiedEntry<'a, K, V, A = Global> +where + A: Allocator, +{ + key: Option<K>, + elem: Bucket<(K, V)>, + table: &'a mut RawTable<(K, V), A>, +} + +unsafe impl<K, V, A> Send for RustcOccupiedEntry<'_, K, V, A> +where + K: Send, + V: Send, + A: Allocator + Send, +{ +} +unsafe impl<K, V, A> Sync for RustcOccupiedEntry<'_, K, V, A> +where + K: Sync, + V: Sync, + A: Allocator + Sync, +{ +} + +impl<K: Debug, V: Debug, A: Allocator> Debug for RustcOccupiedEntry<'_, K, V, A> { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + f.debug_struct("OccupiedEntry") + .field("key", self.key()) + .field("value", self.get()) + .finish() + } +} + +/// A view into a vacant entry in a `HashMap`. +/// It is part of the [`RustcEntry`] enum. +/// +/// [`RustcEntry`]: enum.RustcEntry.html +pub struct RustcVacantEntry<'a, K, V, A = Global> +where + A: Allocator, +{ + hash: u64, + key: K, + table: &'a mut RawTable<(K, V), A>, +} + +impl<K: Debug, V, A: Allocator> Debug for RustcVacantEntry<'_, K, V, A> { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + f.debug_tuple("VacantEntry").field(self.key()).finish() + } +} + +impl<'a, K, V, A: Allocator> RustcEntry<'a, K, V, A> { + /// Sets the value of the entry, and returns a RustcOccupiedEntry. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let mut map: HashMap<&str, u32> = HashMap::new(); + /// let entry = map.rustc_entry("horseyland").insert(37); + /// + /// assert_eq!(entry.key(), &"horseyland"); + /// ``` + pub fn insert(self, value: V) -> RustcOccupiedEntry<'a, K, V, A> { + match self { + Vacant(entry) => entry.insert_entry(value), + Occupied(mut entry) => { + entry.insert(value); + entry + } + } + } + + /// Ensures a value is in the entry by inserting the default if empty, and returns + /// a mutable reference to the value in the entry. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let mut map: HashMap<&str, u32> = HashMap::new(); + /// + /// map.rustc_entry("poneyland").or_insert(3); + /// assert_eq!(map["poneyland"], 3); + /// + /// *map.rustc_entry("poneyland").or_insert(10) *= 2; + /// assert_eq!(map["poneyland"], 6); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn or_insert(self, default: V) -> &'a mut V + where + K: Hash, + { + match self { + Occupied(entry) => entry.into_mut(), + Vacant(entry) => entry.insert(default), + } + } + + /// Ensures a value is in the entry by inserting the result of the default function if empty, + /// and returns a mutable reference to the value in the entry. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let mut map: HashMap<&str, String> = HashMap::new(); + /// let s = "hoho".to_string(); + /// + /// map.rustc_entry("poneyland").or_insert_with(|| s); + /// + /// assert_eq!(map["poneyland"], "hoho".to_string()); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn or_insert_with<F: FnOnce() -> V>(self, default: F) -> &'a mut V + where + K: Hash, + { + match self { + Occupied(entry) => entry.into_mut(), + Vacant(entry) => entry.insert(default()), + } + } + + /// Returns a reference to this entry's key. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let mut map: HashMap<&str, u32> = HashMap::new(); + /// assert_eq!(map.rustc_entry("poneyland").key(), &"poneyland"); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn key(&self) -> &K { + match *self { + Occupied(ref entry) => entry.key(), + Vacant(ref entry) => entry.key(), + } + } + + /// Provides in-place mutable access to an occupied entry before any + /// potential inserts into the map. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let mut map: HashMap<&str, u32> = HashMap::new(); + /// + /// map.rustc_entry("poneyland") + /// .and_modify(|e| { *e += 1 }) + /// .or_insert(42); + /// assert_eq!(map["poneyland"], 42); + /// + /// map.rustc_entry("poneyland") + /// .and_modify(|e| { *e += 1 }) + /// .or_insert(42); + /// assert_eq!(map["poneyland"], 43); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn and_modify<F>(self, f: F) -> Self + where + F: FnOnce(&mut V), + { + match self { + Occupied(mut entry) => { + f(entry.get_mut()); + Occupied(entry) + } + Vacant(entry) => Vacant(entry), + } + } +} + +impl<'a, K, V: Default, A: Allocator> RustcEntry<'a, K, V, A> { + /// Ensures a value is in the entry by inserting the default value if empty, + /// and returns a mutable reference to the value in the entry. + /// + /// # Examples + /// + /// ``` + /// # fn main() { + /// use hashbrown::HashMap; + /// + /// let mut map: HashMap<&str, Option<u32>> = HashMap::new(); + /// map.rustc_entry("poneyland").or_default(); + /// + /// assert_eq!(map["poneyland"], None); + /// # } + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn or_default(self) -> &'a mut V + where + K: Hash, + { + match self { + Occupied(entry) => entry.into_mut(), + Vacant(entry) => entry.insert(Default::default()), + } + } +} + +impl<'a, K, V, A: Allocator> RustcOccupiedEntry<'a, K, V, A> { + /// Gets a reference to the key in the entry. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let mut map: HashMap<&str, u32> = HashMap::new(); + /// map.rustc_entry("poneyland").or_insert(12); + /// assert_eq!(map.rustc_entry("poneyland").key(), &"poneyland"); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn key(&self) -> &K { + unsafe { &self.elem.as_ref().0 } + } + + /// Take the ownership of the key and value from the map. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// use hashbrown::hash_map::RustcEntry; + /// + /// let mut map: HashMap<&str, u32> = HashMap::new(); + /// map.rustc_entry("poneyland").or_insert(12); + /// + /// if let RustcEntry::Occupied(o) = map.rustc_entry("poneyland") { + /// // We delete the entry from the map. + /// o.remove_entry(); + /// } + /// + /// assert_eq!(map.contains_key("poneyland"), false); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn remove_entry(self) -> (K, V) { + unsafe { self.table.remove(self.elem).0 } + } + + /// Gets a reference to the value in the entry. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// use hashbrown::hash_map::RustcEntry; + /// + /// let mut map: HashMap<&str, u32> = HashMap::new(); + /// map.rustc_entry("poneyland").or_insert(12); + /// + /// if let RustcEntry::Occupied(o) = map.rustc_entry("poneyland") { + /// assert_eq!(o.get(), &12); + /// } + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn get(&self) -> &V { + unsafe { &self.elem.as_ref().1 } + } + + /// Gets a mutable reference to the value in the entry. + /// + /// If you need a reference to the `RustcOccupiedEntry` which may outlive the + /// destruction of the `RustcEntry` value, see [`into_mut`]. + /// + /// [`into_mut`]: #method.into_mut + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// use hashbrown::hash_map::RustcEntry; + /// + /// let mut map: HashMap<&str, u32> = HashMap::new(); + /// map.rustc_entry("poneyland").or_insert(12); + /// + /// assert_eq!(map["poneyland"], 12); + /// if let RustcEntry::Occupied(mut o) = map.rustc_entry("poneyland") { + /// *o.get_mut() += 10; + /// assert_eq!(*o.get(), 22); + /// + /// // We can use the same RustcEntry multiple times. + /// *o.get_mut() += 2; + /// } + /// + /// assert_eq!(map["poneyland"], 24); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn get_mut(&mut self) -> &mut V { + unsafe { &mut self.elem.as_mut().1 } + } + + /// Converts the RustcOccupiedEntry into a mutable reference to the value in the entry + /// with a lifetime bound to the map itself. + /// + /// If you need multiple references to the `RustcOccupiedEntry`, see [`get_mut`]. + /// + /// [`get_mut`]: #method.get_mut + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// use hashbrown::hash_map::RustcEntry; + /// + /// let mut map: HashMap<&str, u32> = HashMap::new(); + /// map.rustc_entry("poneyland").or_insert(12); + /// + /// assert_eq!(map["poneyland"], 12); + /// if let RustcEntry::Occupied(o) = map.rustc_entry("poneyland") { + /// *o.into_mut() += 10; + /// } + /// + /// assert_eq!(map["poneyland"], 22); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn into_mut(self) -> &'a mut V { + unsafe { &mut self.elem.as_mut().1 } + } + + /// Sets the value of the entry, and returns the entry's old value. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// use hashbrown::hash_map::RustcEntry; + /// + /// let mut map: HashMap<&str, u32> = HashMap::new(); + /// map.rustc_entry("poneyland").or_insert(12); + /// + /// if let RustcEntry::Occupied(mut o) = map.rustc_entry("poneyland") { + /// assert_eq!(o.insert(15), 12); + /// } + /// + /// assert_eq!(map["poneyland"], 15); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn insert(&mut self, value: V) -> V { + mem::replace(self.get_mut(), value) + } + + /// Takes the value out of the entry, and returns it. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// use hashbrown::hash_map::RustcEntry; + /// + /// let mut map: HashMap<&str, u32> = HashMap::new(); + /// map.rustc_entry("poneyland").or_insert(12); + /// + /// if let RustcEntry::Occupied(o) = map.rustc_entry("poneyland") { + /// assert_eq!(o.remove(), 12); + /// } + /// + /// assert_eq!(map.contains_key("poneyland"), false); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn remove(self) -> V { + self.remove_entry().1 + } + + /// Replaces the entry, returning the old key and value. The new key in the hash map will be + /// the key used to create this entry. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::hash_map::{RustcEntry, HashMap}; + /// use std::rc::Rc; + /// + /// let mut map: HashMap<Rc<String>, u32> = HashMap::new(); + /// map.insert(Rc::new("Stringthing".to_string()), 15); + /// + /// let my_key = Rc::new("Stringthing".to_string()); + /// + /// if let RustcEntry::Occupied(entry) = map.rustc_entry(my_key) { + /// // Also replace the key with a handle to our other key. + /// let (old_key, old_value): (Rc<String>, u32) = entry.replace_entry(16); + /// } + /// + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn replace_entry(self, value: V) -> (K, V) { + let entry = unsafe { self.elem.as_mut() }; + + let old_key = mem::replace(&mut entry.0, self.key.unwrap()); + let old_value = mem::replace(&mut entry.1, value); + + (old_key, old_value) + } + + /// Replaces the key in the hash map with the key used to create this entry. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::hash_map::{RustcEntry, HashMap}; + /// use std::rc::Rc; + /// + /// let mut map: HashMap<Rc<String>, u32> = HashMap::new(); + /// let mut known_strings: Vec<Rc<String>> = Vec::new(); + /// + /// // Initialise known strings, run program, etc. + /// + /// reclaim_memory(&mut map, &known_strings); + /// + /// fn reclaim_memory(map: &mut HashMap<Rc<String>, u32>, known_strings: &[Rc<String>] ) { + /// for s in known_strings { + /// if let RustcEntry::Occupied(entry) = map.rustc_entry(s.clone()) { + /// // Replaces the entry's key with our version of it in `known_strings`. + /// entry.replace_key(); + /// } + /// } + /// } + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn replace_key(self) -> K { + let entry = unsafe { self.elem.as_mut() }; + mem::replace(&mut entry.0, self.key.unwrap()) + } +} + +impl<'a, K, V, A: Allocator> RustcVacantEntry<'a, K, V, A> { + /// Gets a reference to the key that would be used when inserting a value + /// through the `RustcVacantEntry`. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// + /// let mut map: HashMap<&str, u32> = HashMap::new(); + /// assert_eq!(map.rustc_entry("poneyland").key(), &"poneyland"); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn key(&self) -> &K { + &self.key + } + + /// Take ownership of the key. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// use hashbrown::hash_map::RustcEntry; + /// + /// let mut map: HashMap<&str, u32> = HashMap::new(); + /// + /// if let RustcEntry::Vacant(v) = map.rustc_entry("poneyland") { + /// v.into_key(); + /// } + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn into_key(self) -> K { + self.key + } + + /// Sets the value of the entry with the RustcVacantEntry's key, + /// and returns a mutable reference to it. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// use hashbrown::hash_map::RustcEntry; + /// + /// let mut map: HashMap<&str, u32> = HashMap::new(); + /// + /// if let RustcEntry::Vacant(o) = map.rustc_entry("poneyland") { + /// o.insert(37); + /// } + /// assert_eq!(map["poneyland"], 37); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn insert(self, value: V) -> &'a mut V { + unsafe { + let bucket = self.table.insert_no_grow(self.hash, (self.key, value)); + &mut bucket.as_mut().1 + } + } + + /// Sets the value of the entry with the RustcVacantEntry's key, + /// and returns a RustcOccupiedEntry. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashMap; + /// use hashbrown::hash_map::RustcEntry; + /// + /// let mut map: HashMap<&str, u32> = HashMap::new(); + /// + /// if let RustcEntry::Vacant(v) = map.rustc_entry("poneyland") { + /// let o = v.insert_entry(37); + /// assert_eq!(o.get(), &37); + /// } + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn insert_entry(self, value: V) -> RustcOccupiedEntry<'a, K, V, A> { + let bucket = unsafe { self.table.insert_no_grow(self.hash, (self.key, value)) }; + RustcOccupiedEntry { + key: None, + elem: bucket, + table: self.table, + } + } +} + +impl<K, V> IterMut<'_, K, V> { + /// Returns a iterator of references over the remaining items. + #[cfg_attr(feature = "inline-more", inline)] + pub fn rustc_iter(&self) -> Iter<'_, K, V> { + self.iter() + } +} + +impl<K, V> IntoIter<K, V> { + /// Returns a iterator of references over the remaining items. + #[cfg_attr(feature = "inline-more", inline)] + pub fn rustc_iter(&self) -> Iter<'_, K, V> { + self.iter() + } +} + +impl<K, V> Drain<'_, K, V> { + /// Returns a iterator of references over the remaining items. + #[cfg_attr(feature = "inline-more", inline)] + pub fn rustc_iter(&self) -> Iter<'_, K, V> { + self.iter() + } +} diff --git a/vendor/hashbrown/src/scopeguard.rs b/vendor/hashbrown/src/scopeguard.rs new file mode 100644 index 0000000..382d060 --- /dev/null +++ b/vendor/hashbrown/src/scopeguard.rs @@ -0,0 +1,72 @@ +// Extracted from the scopeguard crate +use core::{ + mem::ManuallyDrop, + ops::{Deref, DerefMut}, + ptr, +}; + +pub struct ScopeGuard<T, F> +where + F: FnMut(&mut T), +{ + dropfn: F, + value: T, +} + +#[inline] +pub fn guard<T, F>(value: T, dropfn: F) -> ScopeGuard<T, F> +where + F: FnMut(&mut T), +{ + ScopeGuard { dropfn, value } +} + +impl<T, F> ScopeGuard<T, F> +where + F: FnMut(&mut T), +{ + #[inline] + pub fn into_inner(guard: Self) -> T { + // Cannot move out of Drop-implementing types, so + // ptr::read the value out of a ManuallyDrop<Self> + // Don't use mem::forget as that might invalidate value + let guard = ManuallyDrop::new(guard); + unsafe { + let value = ptr::read(&guard.value); + // read the closure so that it is dropped + let _ = ptr::read(&guard.dropfn); + value + } + } +} + +impl<T, F> Deref for ScopeGuard<T, F> +where + F: FnMut(&mut T), +{ + type Target = T; + #[inline] + fn deref(&self) -> &T { + &self.value + } +} + +impl<T, F> DerefMut for ScopeGuard<T, F> +where + F: FnMut(&mut T), +{ + #[inline] + fn deref_mut(&mut self) -> &mut T { + &mut self.value + } +} + +impl<T, F> Drop for ScopeGuard<T, F> +where + F: FnMut(&mut T), +{ + #[inline] + fn drop(&mut self) { + (self.dropfn)(&mut self.value); + } +} diff --git a/vendor/hashbrown/src/set.rs b/vendor/hashbrown/src/set.rs new file mode 100644 index 0000000..0b8d984 --- /dev/null +++ b/vendor/hashbrown/src/set.rs @@ -0,0 +1,2899 @@ +#[cfg(feature = "raw")] +use crate::raw::RawTable; +use crate::{Equivalent, TryReserveError}; +use alloc::borrow::ToOwned; +use core::fmt; +use core::hash::{BuildHasher, Hash}; +use core::iter::{Chain, FromIterator, FusedIterator}; +use core::ops::{BitAnd, BitOr, BitXor, Sub}; + +use super::map::{self, DefaultHashBuilder, ExtractIfInner, HashMap, Keys}; +use crate::raw::{Allocator, Global}; + +// Future Optimization (FIXME!) +// ============================= +// +// Iteration over zero sized values is a noop. There is no need +// for `bucket.val` in the case of HashSet. I suppose we would need HKT +// to get rid of it properly. + +/// A hash set implemented as a `HashMap` where the value is `()`. +/// +/// As with the [`HashMap`] type, a `HashSet` requires that the elements +/// implement the [`Eq`] and [`Hash`] traits. This can frequently be achieved by +/// using `#[derive(PartialEq, Eq, Hash)]`. If you implement these yourself, +/// it is important that the following property holds: +/// +/// ```text +/// k1 == k2 -> hash(k1) == hash(k2) +/// ``` +/// +/// In other words, if two keys are equal, their hashes must be equal. +/// +/// +/// It is a logic error for an item to be modified in such a way that the +/// item's hash, as determined by the [`Hash`] trait, or its equality, as +/// determined by the [`Eq`] trait, changes while it is in the set. This is +/// normally only possible through [`Cell`], [`RefCell`], global state, I/O, or +/// unsafe code. +/// +/// It is also a logic error for the [`Hash`] implementation of a key to panic. +/// This is generally only possible if the trait is implemented manually. If a +/// panic does occur then the contents of the `HashSet` may become corrupted and +/// some items may be dropped from the table. +/// +/// # Examples +/// +/// ``` +/// use hashbrown::HashSet; +/// // Type inference lets us omit an explicit type signature (which +/// // would be `HashSet<String>` in this example). +/// let mut books = HashSet::new(); +/// +/// // Add some books. +/// books.insert("A Dance With Dragons".to_string()); +/// books.insert("To Kill a Mockingbird".to_string()); +/// books.insert("The Odyssey".to_string()); +/// books.insert("The Great Gatsby".to_string()); +/// +/// // Check for a specific one. +/// if !books.contains("The Winds of Winter") { +/// println!("We have {} books, but The Winds of Winter ain't one.", +/// books.len()); +/// } +/// +/// // Remove a book. +/// books.remove("The Odyssey"); +/// +/// // Iterate over everything. +/// for book in &books { +/// println!("{}", book); +/// } +/// ``` +/// +/// The easiest way to use `HashSet` with a custom type is to derive +/// [`Eq`] and [`Hash`]. We must also derive [`PartialEq`]. This will in the +/// future be implied by [`Eq`]. +/// +/// ``` +/// use hashbrown::HashSet; +/// #[derive(Hash, Eq, PartialEq, Debug)] +/// struct Viking { +/// name: String, +/// power: usize, +/// } +/// +/// let mut vikings = HashSet::new(); +/// +/// vikings.insert(Viking { name: "Einar".to_string(), power: 9 }); +/// vikings.insert(Viking { name: "Einar".to_string(), power: 9 }); +/// vikings.insert(Viking { name: "Olaf".to_string(), power: 4 }); +/// vikings.insert(Viking { name: "Harald".to_string(), power: 8 }); +/// +/// // Use derived implementation to print the vikings. +/// for x in &vikings { +/// println!("{:?}", x); +/// } +/// ``` +/// +/// A `HashSet` with fixed list of elements can be initialized from an array: +/// +/// ``` +/// use hashbrown::HashSet; +/// +/// let viking_names: HashSet<&'static str> = +/// [ "Einar", "Olaf", "Harald" ].iter().cloned().collect(); +/// // use the values stored in the set +/// ``` +/// +/// [`Cell`]: https://doc.rust-lang.org/std/cell/struct.Cell.html +/// [`Eq`]: https://doc.rust-lang.org/std/cmp/trait.Eq.html +/// [`Hash`]: https://doc.rust-lang.org/std/hash/trait.Hash.html +/// [`HashMap`]: struct.HashMap.html +/// [`PartialEq`]: https://doc.rust-lang.org/std/cmp/trait.PartialEq.html +/// [`RefCell`]: https://doc.rust-lang.org/std/cell/struct.RefCell.html +pub struct HashSet<T, S = DefaultHashBuilder, A: Allocator = Global> { + pub(crate) map: HashMap<T, (), S, A>, +} + +impl<T: Clone, S: Clone, A: Allocator + Clone> Clone for HashSet<T, S, A> { + fn clone(&self) -> Self { + HashSet { + map: self.map.clone(), + } + } + + fn clone_from(&mut self, source: &Self) { + self.map.clone_from(&source.map); + } +} + +#[cfg(feature = "ahash")] +impl<T> HashSet<T, DefaultHashBuilder> { + /// Creates an empty `HashSet`. + /// + /// The hash set is initially created with a capacity of 0, so it will not allocate until it + /// is first inserted into. + /// + /// # HashDoS resistance + /// + /// The `hash_builder` normally use a fixed key by default and that does + /// not allow the `HashSet` to be protected against attacks such as [`HashDoS`]. + /// Users who require HashDoS resistance should explicitly use + /// [`ahash::RandomState`] or [`std::collections::hash_map::RandomState`] + /// as the hasher when creating a [`HashSet`], for example with + /// [`with_hasher`](HashSet::with_hasher) method. + /// + /// [`HashDoS`]: https://en.wikipedia.org/wiki/Collision_attack + /// [`std::collections::hash_map::RandomState`]: https://doc.rust-lang.org/std/collections/hash_map/struct.RandomState.html + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashSet; + /// let set: HashSet<i32> = HashSet::new(); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn new() -> Self { + Self { + map: HashMap::new(), + } + } + + /// Creates an empty `HashSet` with the specified capacity. + /// + /// The hash set will be able to hold at least `capacity` elements without + /// reallocating. If `capacity` is 0, the hash set will not allocate. + /// + /// # HashDoS resistance + /// + /// The `hash_builder` normally use a fixed key by default and that does + /// not allow the `HashSet` to be protected against attacks such as [`HashDoS`]. + /// Users who require HashDoS resistance should explicitly use + /// [`ahash::RandomState`] or [`std::collections::hash_map::RandomState`] + /// as the hasher when creating a [`HashSet`], for example with + /// [`with_capacity_and_hasher`](HashSet::with_capacity_and_hasher) method. + /// + /// [`HashDoS`]: https://en.wikipedia.org/wiki/Collision_attack + /// [`std::collections::hash_map::RandomState`]: https://doc.rust-lang.org/std/collections/hash_map/struct.RandomState.html + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashSet; + /// let set: HashSet<i32> = HashSet::with_capacity(10); + /// assert!(set.capacity() >= 10); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn with_capacity(capacity: usize) -> Self { + Self { + map: HashMap::with_capacity(capacity), + } + } +} + +#[cfg(feature = "ahash")] +impl<T: Hash + Eq, A: Allocator> HashSet<T, DefaultHashBuilder, A> { + /// Creates an empty `HashSet`. + /// + /// The hash set is initially created with a capacity of 0, so it will not allocate until it + /// is first inserted into. + /// + /// # HashDoS resistance + /// + /// The `hash_builder` normally use a fixed key by default and that does + /// not allow the `HashSet` to be protected against attacks such as [`HashDoS`]. + /// Users who require HashDoS resistance should explicitly use + /// [`ahash::RandomState`] or [`std::collections::hash_map::RandomState`] + /// as the hasher when creating a [`HashSet`], for example with + /// [`with_hasher_in`](HashSet::with_hasher_in) method. + /// + /// [`HashDoS`]: https://en.wikipedia.org/wiki/Collision_attack + /// [`std::collections::hash_map::RandomState`]: https://doc.rust-lang.org/std/collections/hash_map/struct.RandomState.html + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashSet; + /// let set: HashSet<i32> = HashSet::new(); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn new_in(alloc: A) -> Self { + Self { + map: HashMap::new_in(alloc), + } + } + + /// Creates an empty `HashSet` with the specified capacity. + /// + /// The hash set will be able to hold at least `capacity` elements without + /// reallocating. If `capacity` is 0, the hash set will not allocate. + /// + /// # HashDoS resistance + /// + /// The `hash_builder` normally use a fixed key by default and that does + /// not allow the `HashSet` to be protected against attacks such as [`HashDoS`]. + /// Users who require HashDoS resistance should explicitly use + /// [`ahash::RandomState`] or [`std::collections::hash_map::RandomState`] + /// as the hasher when creating a [`HashSet`], for example with + /// [`with_capacity_and_hasher_in`](HashSet::with_capacity_and_hasher_in) method. + /// + /// [`HashDoS`]: https://en.wikipedia.org/wiki/Collision_attack + /// [`std::collections::hash_map::RandomState`]: https://doc.rust-lang.org/std/collections/hash_map/struct.RandomState.html + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashSet; + /// let set: HashSet<i32> = HashSet::with_capacity(10); + /// assert!(set.capacity() >= 10); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn with_capacity_in(capacity: usize, alloc: A) -> Self { + Self { + map: HashMap::with_capacity_in(capacity, alloc), + } + } +} + +impl<T, S, A: Allocator> HashSet<T, S, A> { + /// Returns the number of elements the set can hold without reallocating. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashSet; + /// let set: HashSet<i32> = HashSet::with_capacity(100); + /// assert!(set.capacity() >= 100); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn capacity(&self) -> usize { + self.map.capacity() + } + + /// An iterator visiting all elements in arbitrary order. + /// The iterator element type is `&'a T`. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashSet; + /// let mut set = HashSet::new(); + /// set.insert("a"); + /// set.insert("b"); + /// + /// // Will print in an arbitrary order. + /// for x in set.iter() { + /// println!("{}", x); + /// } + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn iter(&self) -> Iter<'_, T> { + Iter { + iter: self.map.keys(), + } + } + + /// Returns the number of elements in the set. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashSet; + /// + /// let mut v = HashSet::new(); + /// assert_eq!(v.len(), 0); + /// v.insert(1); + /// assert_eq!(v.len(), 1); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn len(&self) -> usize { + self.map.len() + } + + /// Returns `true` if the set contains no elements. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashSet; + /// + /// let mut v = HashSet::new(); + /// assert!(v.is_empty()); + /// v.insert(1); + /// assert!(!v.is_empty()); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn is_empty(&self) -> bool { + self.map.is_empty() + } + + /// Clears the set, returning all elements in an iterator. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashSet; + /// + /// let mut set: HashSet<_> = [1, 2, 3].iter().cloned().collect(); + /// assert!(!set.is_empty()); + /// + /// // print 1, 2, 3 in an arbitrary order + /// for i in set.drain() { + /// println!("{}", i); + /// } + /// + /// assert!(set.is_empty()); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn drain(&mut self) -> Drain<'_, T, A> { + Drain { + iter: self.map.drain(), + } + } + + /// Retains only the elements specified by the predicate. + /// + /// In other words, remove all elements `e` such that `f(&e)` returns `false`. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashSet; + /// + /// let xs = [1,2,3,4,5,6]; + /// let mut set: HashSet<i32> = xs.iter().cloned().collect(); + /// set.retain(|&k| k % 2 == 0); + /// assert_eq!(set.len(), 3); + /// ``` + pub fn retain<F>(&mut self, mut f: F) + where + F: FnMut(&T) -> bool, + { + self.map.retain(|k, _| f(k)); + } + + /// Drains elements which are true under the given predicate, + /// and returns an iterator over the removed items. + /// + /// In other words, move all elements `e` such that `f(&e)` returns `true` out + /// into another iterator. + /// + /// If the returned `ExtractIf` is not exhausted, e.g. because it is dropped without iterating + /// or the iteration short-circuits, then the remaining elements will be retained. + /// Use [`retain()`] with a negated predicate if you do not need the returned iterator. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashSet; + /// + /// let mut set: HashSet<i32> = (0..8).collect(); + /// let drained: HashSet<i32> = set.extract_if(|v| v % 2 == 0).collect(); + /// + /// let mut evens = drained.into_iter().collect::<Vec<_>>(); + /// let mut odds = set.into_iter().collect::<Vec<_>>(); + /// evens.sort(); + /// odds.sort(); + /// + /// assert_eq!(evens, vec![0, 2, 4, 6]); + /// assert_eq!(odds, vec![1, 3, 5, 7]); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn extract_if<F>(&mut self, f: F) -> ExtractIf<'_, T, F, A> + where + F: FnMut(&T) -> bool, + { + ExtractIf { + f, + inner: ExtractIfInner { + iter: unsafe { self.map.table.iter() }, + table: &mut self.map.table, + }, + } + } + + /// Clears the set, removing all values. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashSet; + /// + /// let mut v = HashSet::new(); + /// v.insert(1); + /// v.clear(); + /// assert!(v.is_empty()); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn clear(&mut self) { + self.map.clear(); + } +} + +impl<T, S> HashSet<T, S, Global> { + /// Creates a new empty hash set which will use the given hasher to hash + /// keys. + /// + /// The hash set is initially created with a capacity of 0, so it will not + /// allocate until it is first inserted into. + /// + /// # HashDoS resistance + /// + /// The `hash_builder` normally use a fixed key by default and that does + /// not allow the `HashSet` to be protected against attacks such as [`HashDoS`]. + /// Users who require HashDoS resistance should explicitly use + /// [`ahash::RandomState`] or [`std::collections::hash_map::RandomState`] + /// as the hasher when creating a [`HashSet`]. + /// + /// The `hash_builder` passed should implement the [`BuildHasher`] trait for + /// the HashSet to be useful, see its documentation for details. + /// + /// [`HashDoS`]: https://en.wikipedia.org/wiki/Collision_attack + /// [`std::collections::hash_map::RandomState`]: https://doc.rust-lang.org/std/collections/hash_map/struct.RandomState.html + /// [`BuildHasher`]: https://doc.rust-lang.org/std/hash/trait.BuildHasher.html + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashSet; + /// use hashbrown::hash_map::DefaultHashBuilder; + /// + /// let s = DefaultHashBuilder::default(); + /// let mut set = HashSet::with_hasher(s); + /// set.insert(2); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub const fn with_hasher(hasher: S) -> Self { + Self { + map: HashMap::with_hasher(hasher), + } + } + + /// Creates an empty `HashSet` with the specified capacity, using + /// `hasher` to hash the keys. + /// + /// The hash set will be able to hold at least `capacity` elements without + /// reallocating. If `capacity` is 0, the hash set will not allocate. + /// + /// # HashDoS resistance + /// + /// The `hash_builder` normally use a fixed key by default and that does + /// not allow the `HashSet` to be protected against attacks such as [`HashDoS`]. + /// Users who require HashDoS resistance should explicitly use + /// [`ahash::RandomState`] or [`std::collections::hash_map::RandomState`] + /// as the hasher when creating a [`HashSet`]. + /// + /// The `hash_builder` passed should implement the [`BuildHasher`] trait for + /// the HashSet to be useful, see its documentation for details. + /// + /// [`HashDoS`]: https://en.wikipedia.org/wiki/Collision_attack + /// [`std::collections::hash_map::RandomState`]: https://doc.rust-lang.org/std/collections/hash_map/struct.RandomState.html + /// [`BuildHasher`]: https://doc.rust-lang.org/std/hash/trait.BuildHasher.html + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashSet; + /// use hashbrown::hash_map::DefaultHashBuilder; + /// + /// let s = DefaultHashBuilder::default(); + /// let mut set = HashSet::with_capacity_and_hasher(10, s); + /// set.insert(1); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn with_capacity_and_hasher(capacity: usize, hasher: S) -> Self { + Self { + map: HashMap::with_capacity_and_hasher(capacity, hasher), + } + } +} + +impl<T, S, A> HashSet<T, S, A> +where + A: Allocator, +{ + /// Returns a reference to the underlying allocator. + #[inline] + pub fn allocator(&self) -> &A { + self.map.allocator() + } + + /// Creates a new empty hash set which will use the given hasher to hash + /// keys. + /// + /// The hash set is initially created with a capacity of 0, so it will not + /// allocate until it is first inserted into. + /// + /// # HashDoS resistance + /// + /// The `hash_builder` normally use a fixed key by default and that does + /// not allow the `HashSet` to be protected against attacks such as [`HashDoS`]. + /// Users who require HashDoS resistance should explicitly use + /// [`ahash::RandomState`] or [`std::collections::hash_map::RandomState`] + /// as the hasher when creating a [`HashSet`]. + /// + /// The `hash_builder` passed should implement the [`BuildHasher`] trait for + /// the HashSet to be useful, see its documentation for details. + /// + /// [`HashDoS`]: https://en.wikipedia.org/wiki/Collision_attack + /// [`std::collections::hash_map::RandomState`]: https://doc.rust-lang.org/std/collections/hash_map/struct.RandomState.html + /// [`BuildHasher`]: https://doc.rust-lang.org/std/hash/trait.BuildHasher.html + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashSet; + /// use hashbrown::hash_map::DefaultHashBuilder; + /// + /// let s = DefaultHashBuilder::default(); + /// let mut set = HashSet::with_hasher(s); + /// set.insert(2); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub const fn with_hasher_in(hasher: S, alloc: A) -> Self { + Self { + map: HashMap::with_hasher_in(hasher, alloc), + } + } + + /// Creates an empty `HashSet` with the specified capacity, using + /// `hasher` to hash the keys. + /// + /// The hash set will be able to hold at least `capacity` elements without + /// reallocating. If `capacity` is 0, the hash set will not allocate. + /// + /// # HashDoS resistance + /// + /// The `hash_builder` normally use a fixed key by default and that does + /// not allow the `HashSet` to be protected against attacks such as [`HashDoS`]. + /// Users who require HashDoS resistance should explicitly use + /// [`ahash::RandomState`] or [`std::collections::hash_map::RandomState`] + /// as the hasher when creating a [`HashSet`]. + /// + /// The `hash_builder` passed should implement the [`BuildHasher`] trait for + /// the HashSet to be useful, see its documentation for details. + /// + /// [`HashDoS`]: https://en.wikipedia.org/wiki/Collision_attack + /// [`std::collections::hash_map::RandomState`]: https://doc.rust-lang.org/std/collections/hash_map/struct.RandomState.html + /// [`BuildHasher`]: https://doc.rust-lang.org/std/hash/trait.BuildHasher.html + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashSet; + /// use hashbrown::hash_map::DefaultHashBuilder; + /// + /// let s = DefaultHashBuilder::default(); + /// let mut set = HashSet::with_capacity_and_hasher(10, s); + /// set.insert(1); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn with_capacity_and_hasher_in(capacity: usize, hasher: S, alloc: A) -> Self { + Self { + map: HashMap::with_capacity_and_hasher_in(capacity, hasher, alloc), + } + } + + /// Returns a reference to the set's [`BuildHasher`]. + /// + /// [`BuildHasher`]: https://doc.rust-lang.org/std/hash/trait.BuildHasher.html + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashSet; + /// use hashbrown::hash_map::DefaultHashBuilder; + /// + /// let hasher = DefaultHashBuilder::default(); + /// let set: HashSet<i32> = HashSet::with_hasher(hasher); + /// let hasher: &DefaultHashBuilder = set.hasher(); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn hasher(&self) -> &S { + self.map.hasher() + } +} + +impl<T, S, A> HashSet<T, S, A> +where + T: Eq + Hash, + S: BuildHasher, + A: Allocator, +{ + /// Reserves capacity for at least `additional` more elements to be inserted + /// in the `HashSet`. The collection may reserve more space to avoid + /// frequent reallocations. + /// + /// # Panics + /// + /// Panics if the new capacity exceeds [`isize::MAX`] bytes and [`abort`] the program + /// in case of allocation error. Use [`try_reserve`](HashSet::try_reserve) instead + /// if you want to handle memory allocation failure. + /// + /// [`isize::MAX`]: https://doc.rust-lang.org/std/primitive.isize.html + /// [`abort`]: https://doc.rust-lang.org/alloc/alloc/fn.handle_alloc_error.html + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashSet; + /// let mut set: HashSet<i32> = HashSet::new(); + /// set.reserve(10); + /// assert!(set.capacity() >= 10); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn reserve(&mut self, additional: usize) { + self.map.reserve(additional); + } + + /// Tries to reserve capacity for at least `additional` more elements to be inserted + /// in the given `HashSet<K,V>`. The collection may reserve more space to avoid + /// frequent reallocations. + /// + /// # Errors + /// + /// If the capacity overflows, or the allocator reports a failure, then an error + /// is returned. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashSet; + /// let mut set: HashSet<i32> = HashSet::new(); + /// set.try_reserve(10).expect("why is the test harness OOMing on 10 bytes?"); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn try_reserve(&mut self, additional: usize) -> Result<(), TryReserveError> { + self.map.try_reserve(additional) + } + + /// Shrinks the capacity of the set as much as possible. It will drop + /// down as much as possible while maintaining the internal rules + /// and possibly leaving some space in accordance with the resize policy. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashSet; + /// + /// let mut set = HashSet::with_capacity(100); + /// set.insert(1); + /// set.insert(2); + /// assert!(set.capacity() >= 100); + /// set.shrink_to_fit(); + /// assert!(set.capacity() >= 2); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn shrink_to_fit(&mut self) { + self.map.shrink_to_fit(); + } + + /// Shrinks the capacity of the set with a lower limit. It will drop + /// down no lower than the supplied limit while maintaining the internal rules + /// and possibly leaving some space in accordance with the resize policy. + /// + /// Panics if the current capacity is smaller than the supplied + /// minimum capacity. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashSet; + /// + /// let mut set = HashSet::with_capacity(100); + /// set.insert(1); + /// set.insert(2); + /// assert!(set.capacity() >= 100); + /// set.shrink_to(10); + /// assert!(set.capacity() >= 10); + /// set.shrink_to(0); + /// assert!(set.capacity() >= 2); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn shrink_to(&mut self, min_capacity: usize) { + self.map.shrink_to(min_capacity); + } + + /// Visits the values representing the difference, + /// i.e., the values that are in `self` but not in `other`. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashSet; + /// let a: HashSet<_> = [1, 2, 3].iter().cloned().collect(); + /// let b: HashSet<_> = [4, 2, 3, 4].iter().cloned().collect(); + /// + /// // Can be seen as `a - b`. + /// for x in a.difference(&b) { + /// println!("{}", x); // Print 1 + /// } + /// + /// let diff: HashSet<_> = a.difference(&b).collect(); + /// assert_eq!(diff, [1].iter().collect()); + /// + /// // Note that difference is not symmetric, + /// // and `b - a` means something else: + /// let diff: HashSet<_> = b.difference(&a).collect(); + /// assert_eq!(diff, [4].iter().collect()); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn difference<'a>(&'a self, other: &'a Self) -> Difference<'a, T, S, A> { + Difference { + iter: self.iter(), + other, + } + } + + /// Visits the values representing the symmetric difference, + /// i.e., the values that are in `self` or in `other` but not in both. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashSet; + /// let a: HashSet<_> = [1, 2, 3].iter().cloned().collect(); + /// let b: HashSet<_> = [4, 2, 3, 4].iter().cloned().collect(); + /// + /// // Print 1, 4 in arbitrary order. + /// for x in a.symmetric_difference(&b) { + /// println!("{}", x); + /// } + /// + /// let diff1: HashSet<_> = a.symmetric_difference(&b).collect(); + /// let diff2: HashSet<_> = b.symmetric_difference(&a).collect(); + /// + /// assert_eq!(diff1, diff2); + /// assert_eq!(diff1, [1, 4].iter().collect()); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn symmetric_difference<'a>(&'a self, other: &'a Self) -> SymmetricDifference<'a, T, S, A> { + SymmetricDifference { + iter: self.difference(other).chain(other.difference(self)), + } + } + + /// Visits the values representing the intersection, + /// i.e., the values that are both in `self` and `other`. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashSet; + /// let a: HashSet<_> = [1, 2, 3].iter().cloned().collect(); + /// let b: HashSet<_> = [4, 2, 3, 4].iter().cloned().collect(); + /// + /// // Print 2, 3 in arbitrary order. + /// for x in a.intersection(&b) { + /// println!("{}", x); + /// } + /// + /// let intersection: HashSet<_> = a.intersection(&b).collect(); + /// assert_eq!(intersection, [2, 3].iter().collect()); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn intersection<'a>(&'a self, other: &'a Self) -> Intersection<'a, T, S, A> { + let (smaller, larger) = if self.len() <= other.len() { + (self, other) + } else { + (other, self) + }; + Intersection { + iter: smaller.iter(), + other: larger, + } + } + + /// Visits the values representing the union, + /// i.e., all the values in `self` or `other`, without duplicates. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashSet; + /// let a: HashSet<_> = [1, 2, 3].iter().cloned().collect(); + /// let b: HashSet<_> = [4, 2, 3, 4].iter().cloned().collect(); + /// + /// // Print 1, 2, 3, 4 in arbitrary order. + /// for x in a.union(&b) { + /// println!("{}", x); + /// } + /// + /// let union: HashSet<_> = a.union(&b).collect(); + /// assert_eq!(union, [1, 2, 3, 4].iter().collect()); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn union<'a>(&'a self, other: &'a Self) -> Union<'a, T, S, A> { + // We'll iterate one set in full, and only the remaining difference from the other. + // Use the smaller set for the difference in order to reduce hash lookups. + let (smaller, larger) = if self.len() <= other.len() { + (self, other) + } else { + (other, self) + }; + Union { + iter: larger.iter().chain(smaller.difference(larger)), + } + } + + /// Returns `true` if the set contains a value. + /// + /// The value may be any borrowed form of the set's value type, but + /// [`Hash`] and [`Eq`] on the borrowed form *must* match those for + /// the value type. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashSet; + /// + /// let set: HashSet<_> = [1, 2, 3].iter().cloned().collect(); + /// assert_eq!(set.contains(&1), true); + /// assert_eq!(set.contains(&4), false); + /// ``` + /// + /// [`Eq`]: https://doc.rust-lang.org/std/cmp/trait.Eq.html + /// [`Hash`]: https://doc.rust-lang.org/std/hash/trait.Hash.html + #[cfg_attr(feature = "inline-more", inline)] + pub fn contains<Q: ?Sized>(&self, value: &Q) -> bool + where + Q: Hash + Equivalent<T>, + { + self.map.contains_key(value) + } + + /// Returns a reference to the value in the set, if any, that is equal to the given value. + /// + /// The value may be any borrowed form of the set's value type, but + /// [`Hash`] and [`Eq`] on the borrowed form *must* match those for + /// the value type. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashSet; + /// + /// let set: HashSet<_> = [1, 2, 3].iter().cloned().collect(); + /// assert_eq!(set.get(&2), Some(&2)); + /// assert_eq!(set.get(&4), None); + /// ``` + /// + /// [`Eq`]: https://doc.rust-lang.org/std/cmp/trait.Eq.html + /// [`Hash`]: https://doc.rust-lang.org/std/hash/trait.Hash.html + #[cfg_attr(feature = "inline-more", inline)] + pub fn get<Q: ?Sized>(&self, value: &Q) -> Option<&T> + where + Q: Hash + Equivalent<T>, + { + // Avoid `Option::map` because it bloats LLVM IR. + match self.map.get_key_value(value) { + Some((k, _)) => Some(k), + None => None, + } + } + + /// Inserts the given `value` into the set if it is not present, then + /// returns a reference to the value in the set. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashSet; + /// + /// let mut set: HashSet<_> = [1, 2, 3].iter().cloned().collect(); + /// assert_eq!(set.len(), 3); + /// assert_eq!(set.get_or_insert(2), &2); + /// assert_eq!(set.get_or_insert(100), &100); + /// assert_eq!(set.len(), 4); // 100 was inserted + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn get_or_insert(&mut self, value: T) -> &T { + // Although the raw entry gives us `&mut T`, we only return `&T` to be consistent with + // `get`. Key mutation is "raw" because you're not supposed to affect `Eq` or `Hash`. + self.map + .raw_entry_mut() + .from_key(&value) + .or_insert(value, ()) + .0 + } + + /// Inserts an owned copy of the given `value` into the set if it is not + /// present, then returns a reference to the value in the set. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashSet; + /// + /// let mut set: HashSet<String> = ["cat", "dog", "horse"] + /// .iter().map(|&pet| pet.to_owned()).collect(); + /// + /// assert_eq!(set.len(), 3); + /// for &pet in &["cat", "dog", "fish"] { + /// let value = set.get_or_insert_owned(pet); + /// assert_eq!(value, pet); + /// } + /// assert_eq!(set.len(), 4); // a new "fish" was inserted + /// ``` + #[inline] + pub fn get_or_insert_owned<Q: ?Sized>(&mut self, value: &Q) -> &T + where + Q: Hash + Equivalent<T> + ToOwned<Owned = T>, + { + // Although the raw entry gives us `&mut T`, we only return `&T` to be consistent with + // `get`. Key mutation is "raw" because you're not supposed to affect `Eq` or `Hash`. + self.map + .raw_entry_mut() + .from_key(value) + .or_insert_with(|| (value.to_owned(), ())) + .0 + } + + /// Inserts a value computed from `f` into the set if the given `value` is + /// not present, then returns a reference to the value in the set. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashSet; + /// + /// let mut set: HashSet<String> = ["cat", "dog", "horse"] + /// .iter().map(|&pet| pet.to_owned()).collect(); + /// + /// assert_eq!(set.len(), 3); + /// for &pet in &["cat", "dog", "fish"] { + /// let value = set.get_or_insert_with(pet, str::to_owned); + /// assert_eq!(value, pet); + /// } + /// assert_eq!(set.len(), 4); // a new "fish" was inserted + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn get_or_insert_with<Q: ?Sized, F>(&mut self, value: &Q, f: F) -> &T + where + Q: Hash + Equivalent<T>, + F: FnOnce(&Q) -> T, + { + // Although the raw entry gives us `&mut T`, we only return `&T` to be consistent with + // `get`. Key mutation is "raw" because you're not supposed to affect `Eq` or `Hash`. + self.map + .raw_entry_mut() + .from_key(value) + .or_insert_with(|| (f(value), ())) + .0 + } + + /// Gets the given value's corresponding entry in the set for in-place manipulation. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashSet; + /// use hashbrown::hash_set::Entry::*; + /// + /// let mut singles = HashSet::new(); + /// let mut dupes = HashSet::new(); + /// + /// for ch in "a short treatise on fungi".chars() { + /// if let Vacant(dupe_entry) = dupes.entry(ch) { + /// // We haven't already seen a duplicate, so + /// // check if we've at least seen it once. + /// match singles.entry(ch) { + /// Vacant(single_entry) => { + /// // We found a new character for the first time. + /// single_entry.insert() + /// } + /// Occupied(single_entry) => { + /// // We've already seen this once, "move" it to dupes. + /// single_entry.remove(); + /// dupe_entry.insert(); + /// } + /// } + /// } + /// } + /// + /// assert!(!singles.contains(&'t') && dupes.contains(&'t')); + /// assert!(singles.contains(&'u') && !dupes.contains(&'u')); + /// assert!(!singles.contains(&'v') && !dupes.contains(&'v')); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn entry(&mut self, value: T) -> Entry<'_, T, S, A> { + match self.map.entry(value) { + map::Entry::Occupied(entry) => Entry::Occupied(OccupiedEntry { inner: entry }), + map::Entry::Vacant(entry) => Entry::Vacant(VacantEntry { inner: entry }), + } + } + + /// Returns `true` if `self` has no elements in common with `other`. + /// This is equivalent to checking for an empty intersection. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashSet; + /// + /// let a: HashSet<_> = [1, 2, 3].iter().cloned().collect(); + /// let mut b = HashSet::new(); + /// + /// assert_eq!(a.is_disjoint(&b), true); + /// b.insert(4); + /// assert_eq!(a.is_disjoint(&b), true); + /// b.insert(1); + /// assert_eq!(a.is_disjoint(&b), false); + /// ``` + pub fn is_disjoint(&self, other: &Self) -> bool { + self.iter().all(|v| !other.contains(v)) + } + + /// Returns `true` if the set is a subset of another, + /// i.e., `other` contains at least all the values in `self`. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashSet; + /// + /// let sup: HashSet<_> = [1, 2, 3].iter().cloned().collect(); + /// let mut set = HashSet::new(); + /// + /// assert_eq!(set.is_subset(&sup), true); + /// set.insert(2); + /// assert_eq!(set.is_subset(&sup), true); + /// set.insert(4); + /// assert_eq!(set.is_subset(&sup), false); + /// ``` + pub fn is_subset(&self, other: &Self) -> bool { + self.len() <= other.len() && self.iter().all(|v| other.contains(v)) + } + + /// Returns `true` if the set is a superset of another, + /// i.e., `self` contains at least all the values in `other`. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashSet; + /// + /// let sub: HashSet<_> = [1, 2].iter().cloned().collect(); + /// let mut set = HashSet::new(); + /// + /// assert_eq!(set.is_superset(&sub), false); + /// + /// set.insert(0); + /// set.insert(1); + /// assert_eq!(set.is_superset(&sub), false); + /// + /// set.insert(2); + /// assert_eq!(set.is_superset(&sub), true); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn is_superset(&self, other: &Self) -> bool { + other.is_subset(self) + } + + /// Adds a value to the set. + /// + /// If the set did not have this value present, `true` is returned. + /// + /// If the set did have this value present, `false` is returned. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashSet; + /// + /// let mut set = HashSet::new(); + /// + /// assert_eq!(set.insert(2), true); + /// assert_eq!(set.insert(2), false); + /// assert_eq!(set.len(), 1); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn insert(&mut self, value: T) -> bool { + self.map.insert(value, ()).is_none() + } + + /// Insert a value the set without checking if the value already exists in the set. + /// + /// Returns a reference to the value just inserted. + /// + /// This operation is safe if a value does not exist in the set. + /// + /// However, if a value exists in the set already, the behavior is unspecified: + /// this operation may panic, loop forever, or any following operation with the set + /// may panic, loop forever or return arbitrary result. + /// + /// That said, this operation (and following operations) are guaranteed to + /// not violate memory safety. + /// + /// This operation is faster than regular insert, because it does not perform + /// lookup before insertion. + /// + /// This operation is useful during initial population of the set. + /// For example, when constructing a set from another set, we know + /// that values are unique. + #[cfg_attr(feature = "inline-more", inline)] + pub fn insert_unique_unchecked(&mut self, value: T) -> &T { + self.map.insert_unique_unchecked(value, ()).0 + } + + /// Adds a value to the set, replacing the existing value, if any, that is equal to the given + /// one. Returns the replaced value. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashSet; + /// + /// let mut set = HashSet::new(); + /// set.insert(Vec::<i32>::new()); + /// + /// assert_eq!(set.get(&[][..]).unwrap().capacity(), 0); + /// set.replace(Vec::with_capacity(10)); + /// assert_eq!(set.get(&[][..]).unwrap().capacity(), 10); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn replace(&mut self, value: T) -> Option<T> { + match self.map.entry(value) { + map::Entry::Occupied(occupied) => Some(occupied.replace_key()), + map::Entry::Vacant(vacant) => { + vacant.insert(()); + None + } + } + } + + /// Removes a value from the set. Returns whether the value was + /// present in the set. + /// + /// The value may be any borrowed form of the set's value type, but + /// [`Hash`] and [`Eq`] on the borrowed form *must* match those for + /// the value type. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashSet; + /// + /// let mut set = HashSet::new(); + /// + /// set.insert(2); + /// assert_eq!(set.remove(&2), true); + /// assert_eq!(set.remove(&2), false); + /// ``` + /// + /// [`Eq`]: https://doc.rust-lang.org/std/cmp/trait.Eq.html + /// [`Hash`]: https://doc.rust-lang.org/std/hash/trait.Hash.html + #[cfg_attr(feature = "inline-more", inline)] + pub fn remove<Q: ?Sized>(&mut self, value: &Q) -> bool + where + Q: Hash + Equivalent<T>, + { + self.map.remove(value).is_some() + } + + /// Removes and returns the value in the set, if any, that is equal to the given one. + /// + /// The value may be any borrowed form of the set's value type, but + /// [`Hash`] and [`Eq`] on the borrowed form *must* match those for + /// the value type. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashSet; + /// + /// let mut set: HashSet<_> = [1, 2, 3].iter().cloned().collect(); + /// assert_eq!(set.take(&2), Some(2)); + /// assert_eq!(set.take(&2), None); + /// ``` + /// + /// [`Eq`]: https://doc.rust-lang.org/std/cmp/trait.Eq.html + /// [`Hash`]: https://doc.rust-lang.org/std/hash/trait.Hash.html + #[cfg_attr(feature = "inline-more", inline)] + pub fn take<Q: ?Sized>(&mut self, value: &Q) -> Option<T> + where + Q: Hash + Equivalent<T>, + { + // Avoid `Option::map` because it bloats LLVM IR. + match self.map.remove_entry(value) { + Some((k, _)) => Some(k), + None => None, + } + } +} + +impl<T, S, A: Allocator> HashSet<T, S, A> { + /// Returns a reference to the [`RawTable`] used underneath [`HashSet`]. + /// This function is only available if the `raw` feature of the crate is enabled. + /// + /// # Note + /// + /// Calling this function is safe, but using the raw hash table API may require + /// unsafe functions or blocks. + /// + /// `RawTable` API gives the lowest level of control under the set that can be useful + /// for extending the HashSet's API, but may lead to *[undefined behavior]*. + /// + /// [`HashSet`]: struct.HashSet.html + /// [`RawTable`]: crate::raw::RawTable + /// [undefined behavior]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html + #[cfg(feature = "raw")] + #[cfg_attr(feature = "inline-more", inline)] + pub fn raw_table(&self) -> &RawTable<(T, ()), A> { + self.map.raw_table() + } + + /// Returns a mutable reference to the [`RawTable`] used underneath [`HashSet`]. + /// This function is only available if the `raw` feature of the crate is enabled. + /// + /// # Note + /// + /// Calling this function is safe, but using the raw hash table API may require + /// unsafe functions or blocks. + /// + /// `RawTable` API gives the lowest level of control under the set that can be useful + /// for extending the HashSet's API, but may lead to *[undefined behavior]*. + /// + /// [`HashSet`]: struct.HashSet.html + /// [`RawTable`]: crate::raw::RawTable + /// [undefined behavior]: https://doc.rust-lang.org/reference/behavior-considered-undefined.html + #[cfg(feature = "raw")] + #[cfg_attr(feature = "inline-more", inline)] + pub fn raw_table_mut(&mut self) -> &mut RawTable<(T, ()), A> { + self.map.raw_table_mut() + } +} + +impl<T, S, A> PartialEq for HashSet<T, S, A> +where + T: Eq + Hash, + S: BuildHasher, + A: Allocator, +{ + fn eq(&self, other: &Self) -> bool { + if self.len() != other.len() { + return false; + } + + self.iter().all(|key| other.contains(key)) + } +} + +impl<T, S, A> Eq for HashSet<T, S, A> +where + T: Eq + Hash, + S: BuildHasher, + A: Allocator, +{ +} + +impl<T, S, A> fmt::Debug for HashSet<T, S, A> +where + T: fmt::Debug, + A: Allocator, +{ + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + f.debug_set().entries(self.iter()).finish() + } +} + +impl<T, S, A> From<HashMap<T, (), S, A>> for HashSet<T, S, A> +where + A: Allocator, +{ + fn from(map: HashMap<T, (), S, A>) -> Self { + Self { map } + } +} + +impl<T, S, A> FromIterator<T> for HashSet<T, S, A> +where + T: Eq + Hash, + S: BuildHasher + Default, + A: Default + Allocator, +{ + #[cfg_attr(feature = "inline-more", inline)] + fn from_iter<I: IntoIterator<Item = T>>(iter: I) -> Self { + let mut set = Self::with_hasher_in(Default::default(), Default::default()); + set.extend(iter); + set + } +} + +// The default hasher is used to match the std implementation signature +#[cfg(feature = "ahash")] +impl<T, A, const N: usize> From<[T; N]> for HashSet<T, DefaultHashBuilder, A> +where + T: Eq + Hash, + A: Default + Allocator, +{ + /// # Examples + /// + /// ``` + /// use hashbrown::HashSet; + /// + /// let set1 = HashSet::from([1, 2, 3, 4]); + /// let set2: HashSet<_> = [1, 2, 3, 4].into(); + /// assert_eq!(set1, set2); + /// ``` + fn from(arr: [T; N]) -> Self { + arr.into_iter().collect() + } +} + +impl<T, S, A> Extend<T> for HashSet<T, S, A> +where + T: Eq + Hash, + S: BuildHasher, + A: Allocator, +{ + #[cfg_attr(feature = "inline-more", inline)] + fn extend<I: IntoIterator<Item = T>>(&mut self, iter: I) { + self.map.extend(iter.into_iter().map(|k| (k, ()))); + } + + #[inline] + #[cfg(feature = "nightly")] + fn extend_one(&mut self, k: T) { + self.map.insert(k, ()); + } + + #[inline] + #[cfg(feature = "nightly")] + fn extend_reserve(&mut self, additional: usize) { + Extend::<(T, ())>::extend_reserve(&mut self.map, additional); + } +} + +impl<'a, T, S, A> Extend<&'a T> for HashSet<T, S, A> +where + T: 'a + Eq + Hash + Copy, + S: BuildHasher, + A: Allocator, +{ + #[cfg_attr(feature = "inline-more", inline)] + fn extend<I: IntoIterator<Item = &'a T>>(&mut self, iter: I) { + self.extend(iter.into_iter().copied()); + } + + #[inline] + #[cfg(feature = "nightly")] + fn extend_one(&mut self, k: &'a T) { + self.map.insert(*k, ()); + } + + #[inline] + #[cfg(feature = "nightly")] + fn extend_reserve(&mut self, additional: usize) { + Extend::<(T, ())>::extend_reserve(&mut self.map, additional); + } +} + +impl<T, S, A> Default for HashSet<T, S, A> +where + S: Default, + A: Default + Allocator, +{ + /// Creates an empty `HashSet<T, S>` with the `Default` value for the hasher. + #[cfg_attr(feature = "inline-more", inline)] + fn default() -> Self { + Self { + map: HashMap::default(), + } + } +} + +impl<T, S, A> BitOr<&HashSet<T, S, A>> for &HashSet<T, S, A> +where + T: Eq + Hash + Clone, + S: BuildHasher + Default, + A: Allocator, +{ + type Output = HashSet<T, S>; + + /// Returns the union of `self` and `rhs` as a new `HashSet<T, S>`. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashSet; + /// + /// let a: HashSet<_> = vec![1, 2, 3].into_iter().collect(); + /// let b: HashSet<_> = vec![3, 4, 5].into_iter().collect(); + /// + /// let set = &a | &b; + /// + /// let mut i = 0; + /// let expected = [1, 2, 3, 4, 5]; + /// for x in &set { + /// assert!(expected.contains(x)); + /// i += 1; + /// } + /// assert_eq!(i, expected.len()); + /// ``` + fn bitor(self, rhs: &HashSet<T, S, A>) -> HashSet<T, S> { + self.union(rhs).cloned().collect() + } +} + +impl<T, S, A> BitAnd<&HashSet<T, S, A>> for &HashSet<T, S, A> +where + T: Eq + Hash + Clone, + S: BuildHasher + Default, + A: Allocator, +{ + type Output = HashSet<T, S>; + + /// Returns the intersection of `self` and `rhs` as a new `HashSet<T, S>`. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashSet; + /// + /// let a: HashSet<_> = vec![1, 2, 3].into_iter().collect(); + /// let b: HashSet<_> = vec![2, 3, 4].into_iter().collect(); + /// + /// let set = &a & &b; + /// + /// let mut i = 0; + /// let expected = [2, 3]; + /// for x in &set { + /// assert!(expected.contains(x)); + /// i += 1; + /// } + /// assert_eq!(i, expected.len()); + /// ``` + fn bitand(self, rhs: &HashSet<T, S, A>) -> HashSet<T, S> { + self.intersection(rhs).cloned().collect() + } +} + +impl<T, S> BitXor<&HashSet<T, S>> for &HashSet<T, S> +where + T: Eq + Hash + Clone, + S: BuildHasher + Default, +{ + type Output = HashSet<T, S>; + + /// Returns the symmetric difference of `self` and `rhs` as a new `HashSet<T, S>`. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashSet; + /// + /// let a: HashSet<_> = vec![1, 2, 3].into_iter().collect(); + /// let b: HashSet<_> = vec![3, 4, 5].into_iter().collect(); + /// + /// let set = &a ^ &b; + /// + /// let mut i = 0; + /// let expected = [1, 2, 4, 5]; + /// for x in &set { + /// assert!(expected.contains(x)); + /// i += 1; + /// } + /// assert_eq!(i, expected.len()); + /// ``` + fn bitxor(self, rhs: &HashSet<T, S>) -> HashSet<T, S> { + self.symmetric_difference(rhs).cloned().collect() + } +} + +impl<T, S> Sub<&HashSet<T, S>> for &HashSet<T, S> +where + T: Eq + Hash + Clone, + S: BuildHasher + Default, +{ + type Output = HashSet<T, S>; + + /// Returns the difference of `self` and `rhs` as a new `HashSet<T, S>`. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashSet; + /// + /// let a: HashSet<_> = vec![1, 2, 3].into_iter().collect(); + /// let b: HashSet<_> = vec![3, 4, 5].into_iter().collect(); + /// + /// let set = &a - &b; + /// + /// let mut i = 0; + /// let expected = [1, 2]; + /// for x in &set { + /// assert!(expected.contains(x)); + /// i += 1; + /// } + /// assert_eq!(i, expected.len()); + /// ``` + fn sub(self, rhs: &HashSet<T, S>) -> HashSet<T, S> { + self.difference(rhs).cloned().collect() + } +} + +/// An iterator over the items of a `HashSet`. +/// +/// This `struct` is created by the [`iter`] method on [`HashSet`]. +/// See its documentation for more. +/// +/// [`HashSet`]: struct.HashSet.html +/// [`iter`]: struct.HashSet.html#method.iter +pub struct Iter<'a, K> { + iter: Keys<'a, K, ()>, +} + +/// An owning iterator over the items of a `HashSet`. +/// +/// This `struct` is created by the [`into_iter`] method on [`HashSet`] +/// (provided by the `IntoIterator` trait). See its documentation for more. +/// +/// [`HashSet`]: struct.HashSet.html +/// [`into_iter`]: struct.HashSet.html#method.into_iter +pub struct IntoIter<K, A: Allocator = Global> { + iter: map::IntoIter<K, (), A>, +} + +/// A draining iterator over the items of a `HashSet`. +/// +/// This `struct` is created by the [`drain`] method on [`HashSet`]. +/// See its documentation for more. +/// +/// [`HashSet`]: struct.HashSet.html +/// [`drain`]: struct.HashSet.html#method.drain +pub struct Drain<'a, K, A: Allocator = Global> { + iter: map::Drain<'a, K, (), A>, +} + +/// A draining iterator over entries of a `HashSet` which don't satisfy the predicate `f`. +/// +/// This `struct` is created by the [`extract_if`] method on [`HashSet`]. See its +/// documentation for more. +/// +/// [`extract_if`]: struct.HashSet.html#method.extract_if +/// [`HashSet`]: struct.HashSet.html +#[must_use = "Iterators are lazy unless consumed"] +pub struct ExtractIf<'a, K, F, A: Allocator = Global> +where + F: FnMut(&K) -> bool, +{ + f: F, + inner: ExtractIfInner<'a, K, (), A>, +} + +/// A lazy iterator producing elements in the intersection of `HashSet`s. +/// +/// This `struct` is created by the [`intersection`] method on [`HashSet`]. +/// See its documentation for more. +/// +/// [`HashSet`]: struct.HashSet.html +/// [`intersection`]: struct.HashSet.html#method.intersection +pub struct Intersection<'a, T, S, A: Allocator = Global> { + // iterator of the first set + iter: Iter<'a, T>, + // the second set + other: &'a HashSet<T, S, A>, +} + +/// A lazy iterator producing elements in the difference of `HashSet`s. +/// +/// This `struct` is created by the [`difference`] method on [`HashSet`]. +/// See its documentation for more. +/// +/// [`HashSet`]: struct.HashSet.html +/// [`difference`]: struct.HashSet.html#method.difference +pub struct Difference<'a, T, S, A: Allocator = Global> { + // iterator of the first set + iter: Iter<'a, T>, + // the second set + other: &'a HashSet<T, S, A>, +} + +/// A lazy iterator producing elements in the symmetric difference of `HashSet`s. +/// +/// This `struct` is created by the [`symmetric_difference`] method on +/// [`HashSet`]. See its documentation for more. +/// +/// [`HashSet`]: struct.HashSet.html +/// [`symmetric_difference`]: struct.HashSet.html#method.symmetric_difference +pub struct SymmetricDifference<'a, T, S, A: Allocator = Global> { + iter: Chain<Difference<'a, T, S, A>, Difference<'a, T, S, A>>, +} + +/// A lazy iterator producing elements in the union of `HashSet`s. +/// +/// This `struct` is created by the [`union`] method on [`HashSet`]. +/// See its documentation for more. +/// +/// [`HashSet`]: struct.HashSet.html +/// [`union`]: struct.HashSet.html#method.union +pub struct Union<'a, T, S, A: Allocator = Global> { + iter: Chain<Iter<'a, T>, Difference<'a, T, S, A>>, +} + +impl<'a, T, S, A: Allocator> IntoIterator for &'a HashSet<T, S, A> { + type Item = &'a T; + type IntoIter = Iter<'a, T>; + + #[cfg_attr(feature = "inline-more", inline)] + fn into_iter(self) -> Iter<'a, T> { + self.iter() + } +} + +impl<T, S, A: Allocator> IntoIterator for HashSet<T, S, A> { + type Item = T; + type IntoIter = IntoIter<T, A>; + + /// Creates a consuming iterator, that is, one that moves each value out + /// of the set in arbitrary order. The set cannot be used after calling + /// this. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashSet; + /// let mut set = HashSet::new(); + /// set.insert("a".to_string()); + /// set.insert("b".to_string()); + /// + /// // Not possible to collect to a Vec<String> with a regular `.iter()`. + /// let v: Vec<String> = set.into_iter().collect(); + /// + /// // Will print in an arbitrary order. + /// for x in &v { + /// println!("{}", x); + /// } + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + fn into_iter(self) -> IntoIter<T, A> { + IntoIter { + iter: self.map.into_iter(), + } + } +} + +impl<K> Clone for Iter<'_, K> { + #[cfg_attr(feature = "inline-more", inline)] + fn clone(&self) -> Self { + Iter { + iter: self.iter.clone(), + } + } +} +impl<'a, K> Iterator for Iter<'a, K> { + type Item = &'a K; + + #[cfg_attr(feature = "inline-more", inline)] + fn next(&mut self) -> Option<&'a K> { + self.iter.next() + } + #[cfg_attr(feature = "inline-more", inline)] + fn size_hint(&self) -> (usize, Option<usize>) { + self.iter.size_hint() + } +} +impl<'a, K> ExactSizeIterator for Iter<'a, K> { + #[cfg_attr(feature = "inline-more", inline)] + fn len(&self) -> usize { + self.iter.len() + } +} +impl<K> FusedIterator for Iter<'_, K> {} + +impl<K: fmt::Debug> fmt::Debug for Iter<'_, K> { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + f.debug_list().entries(self.clone()).finish() + } +} + +impl<K, A: Allocator> Iterator for IntoIter<K, A> { + type Item = K; + + #[cfg_attr(feature = "inline-more", inline)] + fn next(&mut self) -> Option<K> { + // Avoid `Option::map` because it bloats LLVM IR. + match self.iter.next() { + Some((k, _)) => Some(k), + None => None, + } + } + #[cfg_attr(feature = "inline-more", inline)] + fn size_hint(&self) -> (usize, Option<usize>) { + self.iter.size_hint() + } +} +impl<K, A: Allocator> ExactSizeIterator for IntoIter<K, A> { + #[cfg_attr(feature = "inline-more", inline)] + fn len(&self) -> usize { + self.iter.len() + } +} +impl<K, A: Allocator> FusedIterator for IntoIter<K, A> {} + +impl<K: fmt::Debug, A: Allocator> fmt::Debug for IntoIter<K, A> { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + let entries_iter = self.iter.iter().map(|(k, _)| k); + f.debug_list().entries(entries_iter).finish() + } +} + +impl<K, A: Allocator> Iterator for Drain<'_, K, A> { + type Item = K; + + #[cfg_attr(feature = "inline-more", inline)] + fn next(&mut self) -> Option<K> { + // Avoid `Option::map` because it bloats LLVM IR. + match self.iter.next() { + Some((k, _)) => Some(k), + None => None, + } + } + #[cfg_attr(feature = "inline-more", inline)] + fn size_hint(&self) -> (usize, Option<usize>) { + self.iter.size_hint() + } +} +impl<K, A: Allocator> ExactSizeIterator for Drain<'_, K, A> { + #[cfg_attr(feature = "inline-more", inline)] + fn len(&self) -> usize { + self.iter.len() + } +} +impl<K, A: Allocator> FusedIterator for Drain<'_, K, A> {} + +impl<K: fmt::Debug, A: Allocator> fmt::Debug for Drain<'_, K, A> { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + let entries_iter = self.iter.iter().map(|(k, _)| k); + f.debug_list().entries(entries_iter).finish() + } +} + +impl<K, F, A: Allocator> Iterator for ExtractIf<'_, K, F, A> +where + F: FnMut(&K) -> bool, +{ + type Item = K; + + #[cfg_attr(feature = "inline-more", inline)] + fn next(&mut self) -> Option<Self::Item> { + let f = &mut self.f; + let (k, _) = self.inner.next(&mut |k, _| f(k))?; + Some(k) + } + + #[inline] + fn size_hint(&self) -> (usize, Option<usize>) { + (0, self.inner.iter.size_hint().1) + } +} + +impl<K, F, A: Allocator> FusedIterator for ExtractIf<'_, K, F, A> where F: FnMut(&K) -> bool {} + +impl<T, S, A: Allocator> Clone for Intersection<'_, T, S, A> { + #[cfg_attr(feature = "inline-more", inline)] + fn clone(&self) -> Self { + Intersection { + iter: self.iter.clone(), + ..*self + } + } +} + +impl<'a, T, S, A> Iterator for Intersection<'a, T, S, A> +where + T: Eq + Hash, + S: BuildHasher, + A: Allocator, +{ + type Item = &'a T; + + #[cfg_attr(feature = "inline-more", inline)] + fn next(&mut self) -> Option<&'a T> { + loop { + let elt = self.iter.next()?; + if self.other.contains(elt) { + return Some(elt); + } + } + } + + #[cfg_attr(feature = "inline-more", inline)] + fn size_hint(&self) -> (usize, Option<usize>) { + let (_, upper) = self.iter.size_hint(); + (0, upper) + } +} + +impl<T, S, A> fmt::Debug for Intersection<'_, T, S, A> +where + T: fmt::Debug + Eq + Hash, + S: BuildHasher, + A: Allocator, +{ + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + f.debug_list().entries(self.clone()).finish() + } +} + +impl<T, S, A> FusedIterator for Intersection<'_, T, S, A> +where + T: Eq + Hash, + S: BuildHasher, + A: Allocator, +{ +} + +impl<T, S, A: Allocator> Clone for Difference<'_, T, S, A> { + #[cfg_attr(feature = "inline-more", inline)] + fn clone(&self) -> Self { + Difference { + iter: self.iter.clone(), + ..*self + } + } +} + +impl<'a, T, S, A> Iterator for Difference<'a, T, S, A> +where + T: Eq + Hash, + S: BuildHasher, + A: Allocator, +{ + type Item = &'a T; + + #[cfg_attr(feature = "inline-more", inline)] + fn next(&mut self) -> Option<&'a T> { + loop { + let elt = self.iter.next()?; + if !self.other.contains(elt) { + return Some(elt); + } + } + } + + #[cfg_attr(feature = "inline-more", inline)] + fn size_hint(&self) -> (usize, Option<usize>) { + let (_, upper) = self.iter.size_hint(); + (0, upper) + } +} + +impl<T, S, A> FusedIterator for Difference<'_, T, S, A> +where + T: Eq + Hash, + S: BuildHasher, + A: Allocator, +{ +} + +impl<T, S, A> fmt::Debug for Difference<'_, T, S, A> +where + T: fmt::Debug + Eq + Hash, + S: BuildHasher, + A: Allocator, +{ + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + f.debug_list().entries(self.clone()).finish() + } +} + +impl<T, S, A: Allocator> Clone for SymmetricDifference<'_, T, S, A> { + #[cfg_attr(feature = "inline-more", inline)] + fn clone(&self) -> Self { + SymmetricDifference { + iter: self.iter.clone(), + } + } +} + +impl<'a, T, S, A> Iterator for SymmetricDifference<'a, T, S, A> +where + T: Eq + Hash, + S: BuildHasher, + A: Allocator, +{ + type Item = &'a T; + + #[cfg_attr(feature = "inline-more", inline)] + fn next(&mut self) -> Option<&'a T> { + self.iter.next() + } + #[cfg_attr(feature = "inline-more", inline)] + fn size_hint(&self) -> (usize, Option<usize>) { + self.iter.size_hint() + } +} + +impl<T, S, A> FusedIterator for SymmetricDifference<'_, T, S, A> +where + T: Eq + Hash, + S: BuildHasher, + A: Allocator, +{ +} + +impl<T, S, A> fmt::Debug for SymmetricDifference<'_, T, S, A> +where + T: fmt::Debug + Eq + Hash, + S: BuildHasher, + A: Allocator, +{ + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + f.debug_list().entries(self.clone()).finish() + } +} + +impl<T, S, A: Allocator> Clone for Union<'_, T, S, A> { + #[cfg_attr(feature = "inline-more", inline)] + fn clone(&self) -> Self { + Union { + iter: self.iter.clone(), + } + } +} + +impl<T, S, A> FusedIterator for Union<'_, T, S, A> +where + T: Eq + Hash, + S: BuildHasher, + A: Allocator, +{ +} + +impl<T, S, A> fmt::Debug for Union<'_, T, S, A> +where + T: fmt::Debug + Eq + Hash, + S: BuildHasher, + A: Allocator, +{ + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + f.debug_list().entries(self.clone()).finish() + } +} + +impl<'a, T, S, A> Iterator for Union<'a, T, S, A> +where + T: Eq + Hash, + S: BuildHasher, + A: Allocator, +{ + type Item = &'a T; + + #[cfg_attr(feature = "inline-more", inline)] + fn next(&mut self) -> Option<&'a T> { + self.iter.next() + } + #[cfg_attr(feature = "inline-more", inline)] + fn size_hint(&self) -> (usize, Option<usize>) { + self.iter.size_hint() + } +} + +/// A view into a single entry in a set, which may either be vacant or occupied. +/// +/// This `enum` is constructed from the [`entry`] method on [`HashSet`]. +/// +/// [`HashSet`]: struct.HashSet.html +/// [`entry`]: struct.HashSet.html#method.entry +/// +/// # Examples +/// +/// ``` +/// use hashbrown::hash_set::{Entry, HashSet, OccupiedEntry}; +/// +/// let mut set = HashSet::new(); +/// set.extend(["a", "b", "c"]); +/// assert_eq!(set.len(), 3); +/// +/// // Existing value (insert) +/// let entry: Entry<_, _> = set.entry("a"); +/// let _raw_o: OccupiedEntry<_, _> = entry.insert(); +/// assert_eq!(set.len(), 3); +/// // Nonexistent value (insert) +/// set.entry("d").insert(); +/// +/// // Existing value (or_insert) +/// set.entry("b").or_insert(); +/// // Nonexistent value (or_insert) +/// set.entry("e").or_insert(); +/// +/// println!("Our HashSet: {:?}", set); +/// +/// let mut vec: Vec<_> = set.iter().copied().collect(); +/// // The `Iter` iterator produces items in arbitrary order, so the +/// // items must be sorted to test them against a sorted array. +/// vec.sort_unstable(); +/// assert_eq!(vec, ["a", "b", "c", "d", "e"]); +/// ``` +pub enum Entry<'a, T, S, A = Global> +where + A: Allocator, +{ + /// An occupied entry. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::hash_set::{Entry, HashSet}; + /// let mut set: HashSet<_> = ["a", "b"].into(); + /// + /// match set.entry("a") { + /// Entry::Vacant(_) => unreachable!(), + /// Entry::Occupied(_) => { } + /// } + /// ``` + Occupied(OccupiedEntry<'a, T, S, A>), + + /// A vacant entry. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::hash_set::{Entry, HashSet}; + /// let mut set: HashSet<&str> = HashSet::new(); + /// + /// match set.entry("a") { + /// Entry::Occupied(_) => unreachable!(), + /// Entry::Vacant(_) => { } + /// } + /// ``` + Vacant(VacantEntry<'a, T, S, A>), +} + +impl<T: fmt::Debug, S, A: Allocator> fmt::Debug for Entry<'_, T, S, A> { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + match *self { + Entry::Vacant(ref v) => f.debug_tuple("Entry").field(v).finish(), + Entry::Occupied(ref o) => f.debug_tuple("Entry").field(o).finish(), + } + } +} + +/// A view into an occupied entry in a `HashSet`. +/// It is part of the [`Entry`] enum. +/// +/// [`Entry`]: enum.Entry.html +/// +/// # Examples +/// +/// ``` +/// use hashbrown::hash_set::{Entry, HashSet, OccupiedEntry}; +/// +/// let mut set = HashSet::new(); +/// set.extend(["a", "b", "c"]); +/// +/// let _entry_o: OccupiedEntry<_, _> = set.entry("a").insert(); +/// assert_eq!(set.len(), 3); +/// +/// // Existing key +/// match set.entry("a") { +/// Entry::Vacant(_) => unreachable!(), +/// Entry::Occupied(view) => { +/// assert_eq!(view.get(), &"a"); +/// } +/// } +/// +/// assert_eq!(set.len(), 3); +/// +/// // Existing key (take) +/// match set.entry("c") { +/// Entry::Vacant(_) => unreachable!(), +/// Entry::Occupied(view) => { +/// assert_eq!(view.remove(), "c"); +/// } +/// } +/// assert_eq!(set.get(&"c"), None); +/// assert_eq!(set.len(), 2); +/// ``` +pub struct OccupiedEntry<'a, T, S, A: Allocator = Global> { + inner: map::OccupiedEntry<'a, T, (), S, A>, +} + +impl<T: fmt::Debug, S, A: Allocator> fmt::Debug for OccupiedEntry<'_, T, S, A> { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + f.debug_struct("OccupiedEntry") + .field("value", self.get()) + .finish() + } +} + +/// A view into a vacant entry in a `HashSet`. +/// It is part of the [`Entry`] enum. +/// +/// [`Entry`]: enum.Entry.html +/// +/// # Examples +/// +/// ``` +/// use hashbrown::hash_set::{Entry, HashSet, VacantEntry}; +/// +/// let mut set = HashSet::<&str>::new(); +/// +/// let entry_v: VacantEntry<_, _> = match set.entry("a") { +/// Entry::Vacant(view) => view, +/// Entry::Occupied(_) => unreachable!(), +/// }; +/// entry_v.insert(); +/// assert!(set.contains("a") && set.len() == 1); +/// +/// // Nonexistent key (insert) +/// match set.entry("b") { +/// Entry::Vacant(view) => view.insert(), +/// Entry::Occupied(_) => unreachable!(), +/// } +/// assert!(set.contains("b") && set.len() == 2); +/// ``` +pub struct VacantEntry<'a, T, S, A: Allocator = Global> { + inner: map::VacantEntry<'a, T, (), S, A>, +} + +impl<T: fmt::Debug, S, A: Allocator> fmt::Debug for VacantEntry<'_, T, S, A> { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + f.debug_tuple("VacantEntry").field(self.get()).finish() + } +} + +impl<'a, T, S, A: Allocator> Entry<'a, T, S, A> { + /// Sets the value of the entry, and returns an OccupiedEntry. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashSet; + /// + /// let mut set: HashSet<&str> = HashSet::new(); + /// let entry = set.entry("horseyland").insert(); + /// + /// assert_eq!(entry.get(), &"horseyland"); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn insert(self) -> OccupiedEntry<'a, T, S, A> + where + T: Hash, + S: BuildHasher, + { + match self { + Entry::Occupied(entry) => entry, + Entry::Vacant(entry) => entry.insert_entry(), + } + } + + /// Ensures a value is in the entry by inserting if it was vacant. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashSet; + /// + /// let mut set: HashSet<&str> = HashSet::new(); + /// + /// // nonexistent key + /// set.entry("poneyland").or_insert(); + /// assert!(set.contains("poneyland")); + /// + /// // existing key + /// set.entry("poneyland").or_insert(); + /// assert!(set.contains("poneyland")); + /// assert_eq!(set.len(), 1); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn or_insert(self) + where + T: Hash, + S: BuildHasher, + { + if let Entry::Vacant(entry) = self { + entry.insert(); + } + } + + /// Returns a reference to this entry's value. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashSet; + /// + /// let mut set: HashSet<&str> = HashSet::new(); + /// set.entry("poneyland").or_insert(); + /// // existing key + /// assert_eq!(set.entry("poneyland").get(), &"poneyland"); + /// // nonexistent key + /// assert_eq!(set.entry("horseland").get(), &"horseland"); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn get(&self) -> &T { + match *self { + Entry::Occupied(ref entry) => entry.get(), + Entry::Vacant(ref entry) => entry.get(), + } + } +} + +impl<T, S, A: Allocator> OccupiedEntry<'_, T, S, A> { + /// Gets a reference to the value in the entry. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::hash_set::{Entry, HashSet}; + /// + /// let mut set: HashSet<&str> = HashSet::new(); + /// set.entry("poneyland").or_insert(); + /// + /// match set.entry("poneyland") { + /// Entry::Vacant(_) => panic!(), + /// Entry::Occupied(entry) => assert_eq!(entry.get(), &"poneyland"), + /// } + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn get(&self) -> &T { + self.inner.key() + } + + /// Takes the value out of the entry, and returns it. + /// Keeps the allocated memory for reuse. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashSet; + /// use hashbrown::hash_set::Entry; + /// + /// let mut set: HashSet<&str> = HashSet::new(); + /// // The set is empty + /// assert!(set.is_empty() && set.capacity() == 0); + /// + /// set.entry("poneyland").or_insert(); + /// let capacity_before_remove = set.capacity(); + /// + /// if let Entry::Occupied(o) = set.entry("poneyland") { + /// assert_eq!(o.remove(), "poneyland"); + /// } + /// + /// assert_eq!(set.contains("poneyland"), false); + /// // Now set hold none elements but capacity is equal to the old one + /// assert!(set.len() == 0 && set.capacity() == capacity_before_remove); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn remove(self) -> T { + self.inner.remove_entry().0 + } + + /// Replaces the entry, returning the old value. The new value in the hash map will be + /// the value used to create this entry. + /// + /// # Panics + /// + /// Will panic if this OccupiedEntry was created through [`Entry::insert`]. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::hash_set::{Entry, HashSet}; + /// use std::rc::Rc; + /// + /// let mut set: HashSet<Rc<String>> = HashSet::new(); + /// let key_one = Rc::new("Stringthing".to_string()); + /// let key_two = Rc::new("Stringthing".to_string()); + /// + /// set.insert(key_one.clone()); + /// assert!(Rc::strong_count(&key_one) == 2 && Rc::strong_count(&key_two) == 1); + /// + /// match set.entry(key_two.clone()) { + /// Entry::Occupied(entry) => { + /// let old_key: Rc<String> = entry.replace(); + /// assert!(Rc::ptr_eq(&key_one, &old_key)); + /// } + /// Entry::Vacant(_) => panic!(), + /// } + /// + /// assert!(Rc::strong_count(&key_one) == 1 && Rc::strong_count(&key_two) == 2); + /// assert!(set.contains(&"Stringthing".to_owned())); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn replace(self) -> T { + self.inner.replace_key() + } +} + +impl<'a, T, S, A: Allocator> VacantEntry<'a, T, S, A> { + /// Gets a reference to the value that would be used when inserting + /// through the `VacantEntry`. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashSet; + /// + /// let mut set: HashSet<&str> = HashSet::new(); + /// assert_eq!(set.entry("poneyland").get(), &"poneyland"); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn get(&self) -> &T { + self.inner.key() + } + + /// Take ownership of the value. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::hash_set::{Entry, HashSet}; + /// + /// let mut set: HashSet<&str> = HashSet::new(); + /// + /// match set.entry("poneyland") { + /// Entry::Occupied(_) => panic!(), + /// Entry::Vacant(v) => assert_eq!(v.into_value(), "poneyland"), + /// } + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn into_value(self) -> T { + self.inner.into_key() + } + + /// Sets the value of the entry with the VacantEntry's value. + /// + /// # Examples + /// + /// ``` + /// use hashbrown::HashSet; + /// use hashbrown::hash_set::Entry; + /// + /// let mut set: HashSet<&str> = HashSet::new(); + /// + /// if let Entry::Vacant(o) = set.entry("poneyland") { + /// o.insert(); + /// } + /// assert!(set.contains("poneyland")); + /// ``` + #[cfg_attr(feature = "inline-more", inline)] + pub fn insert(self) + where + T: Hash, + S: BuildHasher, + { + self.inner.insert(()); + } + + #[cfg_attr(feature = "inline-more", inline)] + fn insert_entry(self) -> OccupiedEntry<'a, T, S, A> + where + T: Hash, + S: BuildHasher, + { + OccupiedEntry { + inner: self.inner.insert_entry(()), + } + } +} + +#[allow(dead_code)] +fn assert_covariance() { + fn set<'new>(v: HashSet<&'static str>) -> HashSet<&'new str> { + v + } + fn iter<'a, 'new>(v: Iter<'a, &'static str>) -> Iter<'a, &'new str> { + v + } + fn into_iter<'new, A: Allocator>(v: IntoIter<&'static str, A>) -> IntoIter<&'new str, A> { + v + } + fn difference<'a, 'new, A: Allocator>( + v: Difference<'a, &'static str, DefaultHashBuilder, A>, + ) -> Difference<'a, &'new str, DefaultHashBuilder, A> { + v + } + fn symmetric_difference<'a, 'new, A: Allocator>( + v: SymmetricDifference<'a, &'static str, DefaultHashBuilder, A>, + ) -> SymmetricDifference<'a, &'new str, DefaultHashBuilder, A> { + v + } + fn intersection<'a, 'new, A: Allocator>( + v: Intersection<'a, &'static str, DefaultHashBuilder, A>, + ) -> Intersection<'a, &'new str, DefaultHashBuilder, A> { + v + } + fn union<'a, 'new, A: Allocator>( + v: Union<'a, &'static str, DefaultHashBuilder, A>, + ) -> Union<'a, &'new str, DefaultHashBuilder, A> { + v + } + fn drain<'new, A: Allocator>(d: Drain<'static, &'static str, A>) -> Drain<'new, &'new str, A> { + d + } +} + +#[cfg(test)] +mod test_set { + use super::super::map::DefaultHashBuilder; + use super::HashSet; + use std::vec::Vec; + + #[test] + fn test_zero_capacities() { + type HS = HashSet<i32>; + + let s = HS::new(); + assert_eq!(s.capacity(), 0); + + let s = HS::default(); + assert_eq!(s.capacity(), 0); + + let s = HS::with_hasher(DefaultHashBuilder::default()); + assert_eq!(s.capacity(), 0); + + let s = HS::with_capacity(0); + assert_eq!(s.capacity(), 0); + + let s = HS::with_capacity_and_hasher(0, DefaultHashBuilder::default()); + assert_eq!(s.capacity(), 0); + + let mut s = HS::new(); + s.insert(1); + s.insert(2); + s.remove(&1); + s.remove(&2); + s.shrink_to_fit(); + assert_eq!(s.capacity(), 0); + + let mut s = HS::new(); + s.reserve(0); + assert_eq!(s.capacity(), 0); + } + + #[test] + fn test_disjoint() { + let mut xs = HashSet::new(); + let mut ys = HashSet::new(); + assert!(xs.is_disjoint(&ys)); + assert!(ys.is_disjoint(&xs)); + assert!(xs.insert(5)); + assert!(ys.insert(11)); + assert!(xs.is_disjoint(&ys)); + assert!(ys.is_disjoint(&xs)); + assert!(xs.insert(7)); + assert!(xs.insert(19)); + assert!(xs.insert(4)); + assert!(ys.insert(2)); + assert!(ys.insert(-11)); + assert!(xs.is_disjoint(&ys)); + assert!(ys.is_disjoint(&xs)); + assert!(ys.insert(7)); + assert!(!xs.is_disjoint(&ys)); + assert!(!ys.is_disjoint(&xs)); + } + + #[test] + fn test_subset_and_superset() { + let mut a = HashSet::new(); + assert!(a.insert(0)); + assert!(a.insert(5)); + assert!(a.insert(11)); + assert!(a.insert(7)); + + let mut b = HashSet::new(); + assert!(b.insert(0)); + assert!(b.insert(7)); + assert!(b.insert(19)); + assert!(b.insert(250)); + assert!(b.insert(11)); + assert!(b.insert(200)); + + assert!(!a.is_subset(&b)); + assert!(!a.is_superset(&b)); + assert!(!b.is_subset(&a)); + assert!(!b.is_superset(&a)); + + assert!(b.insert(5)); + + assert!(a.is_subset(&b)); + assert!(!a.is_superset(&b)); + assert!(!b.is_subset(&a)); + assert!(b.is_superset(&a)); + } + + #[test] + fn test_iterate() { + let mut a = HashSet::new(); + for i in 0..32 { + assert!(a.insert(i)); + } + let mut observed: u32 = 0; + for k in &a { + observed |= 1 << *k; + } + assert_eq!(observed, 0xFFFF_FFFF); + } + + #[test] + fn test_intersection() { + let mut a = HashSet::new(); + let mut b = HashSet::new(); + + assert!(a.insert(11)); + assert!(a.insert(1)); + assert!(a.insert(3)); + assert!(a.insert(77)); + assert!(a.insert(103)); + assert!(a.insert(5)); + assert!(a.insert(-5)); + + assert!(b.insert(2)); + assert!(b.insert(11)); + assert!(b.insert(77)); + assert!(b.insert(-9)); + assert!(b.insert(-42)); + assert!(b.insert(5)); + assert!(b.insert(3)); + + let mut i = 0; + let expected = [3, 5, 11, 77]; + for x in a.intersection(&b) { + assert!(expected.contains(x)); + i += 1; + } + assert_eq!(i, expected.len()); + } + + #[test] + fn test_difference() { + let mut a = HashSet::new(); + let mut b = HashSet::new(); + + assert!(a.insert(1)); + assert!(a.insert(3)); + assert!(a.insert(5)); + assert!(a.insert(9)); + assert!(a.insert(11)); + + assert!(b.insert(3)); + assert!(b.insert(9)); + + let mut i = 0; + let expected = [1, 5, 11]; + for x in a.difference(&b) { + assert!(expected.contains(x)); + i += 1; + } + assert_eq!(i, expected.len()); + } + + #[test] + fn test_symmetric_difference() { + let mut a = HashSet::new(); + let mut b = HashSet::new(); + + assert!(a.insert(1)); + assert!(a.insert(3)); + assert!(a.insert(5)); + assert!(a.insert(9)); + assert!(a.insert(11)); + + assert!(b.insert(-2)); + assert!(b.insert(3)); + assert!(b.insert(9)); + assert!(b.insert(14)); + assert!(b.insert(22)); + + let mut i = 0; + let expected = [-2, 1, 5, 11, 14, 22]; + for x in a.symmetric_difference(&b) { + assert!(expected.contains(x)); + i += 1; + } + assert_eq!(i, expected.len()); + } + + #[test] + fn test_union() { + let mut a = HashSet::new(); + let mut b = HashSet::new(); + + assert!(a.insert(1)); + assert!(a.insert(3)); + assert!(a.insert(5)); + assert!(a.insert(9)); + assert!(a.insert(11)); + assert!(a.insert(16)); + assert!(a.insert(19)); + assert!(a.insert(24)); + + assert!(b.insert(-2)); + assert!(b.insert(1)); + assert!(b.insert(5)); + assert!(b.insert(9)); + assert!(b.insert(13)); + assert!(b.insert(19)); + + let mut i = 0; + let expected = [-2, 1, 3, 5, 9, 11, 13, 16, 19, 24]; + for x in a.union(&b) { + assert!(expected.contains(x)); + i += 1; + } + assert_eq!(i, expected.len()); + } + + #[test] + fn test_from_map() { + let mut a = crate::HashMap::new(); + a.insert(1, ()); + a.insert(2, ()); + a.insert(3, ()); + a.insert(4, ()); + + let a: HashSet<_> = a.into(); + + assert_eq!(a.len(), 4); + assert!(a.contains(&1)); + assert!(a.contains(&2)); + assert!(a.contains(&3)); + assert!(a.contains(&4)); + } + + #[test] + fn test_from_iter() { + let xs = [1, 2, 2, 3, 4, 5, 6, 7, 8, 9]; + + let set: HashSet<_> = xs.iter().copied().collect(); + + for x in &xs { + assert!(set.contains(x)); + } + + assert_eq!(set.iter().len(), xs.len() - 1); + } + + #[test] + fn test_move_iter() { + let hs = { + let mut hs = HashSet::new(); + + hs.insert('a'); + hs.insert('b'); + + hs + }; + + let v = hs.into_iter().collect::<Vec<char>>(); + assert!(v == ['a', 'b'] || v == ['b', 'a']); + } + + #[test] + fn test_eq() { + // These constants once happened to expose a bug in insert(). + // I'm keeping them around to prevent a regression. + let mut s1 = HashSet::new(); + + s1.insert(1); + s1.insert(2); + s1.insert(3); + + let mut s2 = HashSet::new(); + + s2.insert(1); + s2.insert(2); + + assert!(s1 != s2); + + s2.insert(3); + + assert_eq!(s1, s2); + } + + #[test] + fn test_show() { + let mut set = HashSet::new(); + let empty = HashSet::<i32>::new(); + + set.insert(1); + set.insert(2); + + let set_str = format!("{set:?}"); + + assert!(set_str == "{1, 2}" || set_str == "{2, 1}"); + assert_eq!(format!("{empty:?}"), "{}"); + } + + #[test] + fn test_trivial_drain() { + let mut s = HashSet::<i32>::new(); + for _ in s.drain() {} + assert!(s.is_empty()); + drop(s); + + let mut s = HashSet::<i32>::new(); + drop(s.drain()); + assert!(s.is_empty()); + } + + #[test] + fn test_drain() { + let mut s: HashSet<_> = (1..100).collect(); + + // try this a bunch of times to make sure we don't screw up internal state. + for _ in 0..20 { + assert_eq!(s.len(), 99); + + { + let mut last_i = 0; + let mut d = s.drain(); + for (i, x) in d.by_ref().take(50).enumerate() { + last_i = i; + assert!(x != 0); + } + assert_eq!(last_i, 49); + } + + for _ in &s { + panic!("s should be empty!"); + } + + // reset to try again. + s.extend(1..100); + } + } + + #[test] + fn test_replace() { + use core::hash; + + #[derive(Debug)] + struct Foo(&'static str, i32); + + impl PartialEq for Foo { + fn eq(&self, other: &Self) -> bool { + self.0 == other.0 + } + } + + impl Eq for Foo {} + + impl hash::Hash for Foo { + fn hash<H: hash::Hasher>(&self, h: &mut H) { + self.0.hash(h); + } + } + + let mut s = HashSet::new(); + assert_eq!(s.replace(Foo("a", 1)), None); + assert_eq!(s.len(), 1); + assert_eq!(s.replace(Foo("a", 2)), Some(Foo("a", 1))); + assert_eq!(s.len(), 1); + + let mut it = s.iter(); + assert_eq!(it.next(), Some(&Foo("a", 2))); + assert_eq!(it.next(), None); + } + + #[test] + #[allow(clippy::needless_borrow)] + fn test_extend_ref() { + let mut a = HashSet::new(); + a.insert(1); + + a.extend([2, 3, 4]); + + assert_eq!(a.len(), 4); + assert!(a.contains(&1)); + assert!(a.contains(&2)); + assert!(a.contains(&3)); + assert!(a.contains(&4)); + + let mut b = HashSet::new(); + b.insert(5); + b.insert(6); + + a.extend(&b); + + assert_eq!(a.len(), 6); + assert!(a.contains(&1)); + assert!(a.contains(&2)); + assert!(a.contains(&3)); + assert!(a.contains(&4)); + assert!(a.contains(&5)); + assert!(a.contains(&6)); + } + + #[test] + fn test_retain() { + let xs = [1, 2, 3, 4, 5, 6]; + let mut set: HashSet<i32> = xs.iter().copied().collect(); + set.retain(|&k| k % 2 == 0); + assert_eq!(set.len(), 3); + assert!(set.contains(&2)); + assert!(set.contains(&4)); + assert!(set.contains(&6)); + } + + #[test] + fn test_extract_if() { + { + let mut set: HashSet<i32> = (0..8).collect(); + let drained = set.extract_if(|&k| k % 2 == 0); + let mut out = drained.collect::<Vec<_>>(); + out.sort_unstable(); + assert_eq!(vec![0, 2, 4, 6], out); + assert_eq!(set.len(), 4); + } + { + let mut set: HashSet<i32> = (0..8).collect(); + set.extract_if(|&k| k % 2 == 0).for_each(drop); + assert_eq!(set.len(), 4, "Removes non-matching items on drop"); + } + } + + #[test] + fn test_const_with_hasher() { + use core::hash::BuildHasher; + use std::collections::hash_map::DefaultHasher; + + #[derive(Clone)] + struct MyHasher; + impl BuildHasher for MyHasher { + type Hasher = DefaultHasher; + + fn build_hasher(&self) -> DefaultHasher { + DefaultHasher::new() + } + } + + const EMPTY_SET: HashSet<u32, MyHasher> = HashSet::with_hasher(MyHasher); + + let mut set = EMPTY_SET; + set.insert(19); + assert!(set.contains(&19)); + } + + #[test] + fn rehash_in_place() { + let mut set = HashSet::new(); + + for i in 0..224 { + set.insert(i); + } + + assert_eq!( + set.capacity(), + 224, + "The set must be at or close to capacity to trigger a re hashing" + ); + + for i in 100..1400 { + set.remove(&(i - 100)); + set.insert(i); + } + } + + #[test] + fn collect() { + // At the time of writing, this hits the ZST case in from_base_index + // (and without the `map`, it does not). + let mut _set: HashSet<_> = (0..3).map(|_| ()).collect(); + } +} |