From dc0db358abe19481e475e10c32149b53370f1a1c Mon Sep 17 00:00:00 2001 From: Daniel Baumann Date: Thu, 30 May 2024 05:57:31 +0200 Subject: Merging upstream version 1.72.1+dfsg1. Signed-off-by: Daniel Baumann --- vendor/hashbrown/.cargo-checksum.json | 2 +- vendor/hashbrown/CHANGELOG.md | 42 +- vendor/hashbrown/Cargo.toml | 30 +- vendor/hashbrown/README.md | 75 +- vendor/hashbrown/benches/bench.rs | 2 +- vendor/hashbrown/src/external_trait_impls/mod.rs | 2 + .../src/external_trait_impls/rayon/map.rs | 7 +- .../src/external_trait_impls/rkyv/hash_map.rs | 125 ++ .../src/external_trait_impls/rkyv/hash_set.rs | 123 ++ .../hashbrown/src/external_trait_impls/rkyv/mod.rs | 2 + vendor/hashbrown/src/lib.rs | 22 +- vendor/hashbrown/src/macros.rs | 2 +- vendor/hashbrown/src/map.rs | 284 +++-- vendor/hashbrown/src/raw/alloc.rs | 57 +- vendor/hashbrown/src/raw/bitmask.rs | 99 +- vendor/hashbrown/src/raw/generic.rs | 59 +- vendor/hashbrown/src/raw/mod.rs | 1246 +++++++++++++++++--- vendor/hashbrown/src/raw/neon.rs | 124 ++ vendor/hashbrown/src/raw/sse2.rs | 31 +- vendor/hashbrown/src/rustc_entry.rs | 6 +- vendor/hashbrown/src/set.rs | 90 +- vendor/hashbrown/tests/rayon.rs | 4 +- vendor/hashbrown/tests/set.rs | 2 +- 23 files changed, 1875 insertions(+), 561 deletions(-) create mode 100644 vendor/hashbrown/src/external_trait_impls/rkyv/hash_map.rs create mode 100644 vendor/hashbrown/src/external_trait_impls/rkyv/hash_set.rs create mode 100644 vendor/hashbrown/src/external_trait_impls/rkyv/mod.rs create mode 100644 vendor/hashbrown/src/raw/neon.rs (limited to 'vendor/hashbrown') diff --git a/vendor/hashbrown/.cargo-checksum.json b/vendor/hashbrown/.cargo-checksum.json index 095d0c719..aed788fc5 100644 --- a/vendor/hashbrown/.cargo-checksum.json +++ b/vendor/hashbrown/.cargo-checksum.json @@ -1 +1 @@ -{"files":{"CHANGELOG.md":"9f6a99787d52bc87da929df9348c5e6db403c54995be5a006eea4b641efb7652","Cargo.toml":"da2b6aaf324e787d10a6a3626347dae7c1b101614a04eba0e13a5f455201bdc7","LICENSE-APACHE":"a60eea817514531668d7e00765731449fe14d059d3249e0bc93b36de45f759f2","LICENSE-MIT":"ff8f68cb076caf8cefe7a6430d4ac086ce6af2ca8ce2c4e5a2004d4552ef52a2","README.md":"292bb49a47b4f75f0a94fd2316d43349cd3238823e5b9b96b9ee4d963a88b998","benches/bench.rs":"aadc39d815eadf094ed9357d946319df2d93194203bbccb7c33cea6951d654df","benches/insert_unique_unchecked.rs":"cb84275f22d5f95a5ac995ac6b2df74ffcf342765b401d27c95f2955c7b7cb9f","clippy.toml":"7535949f908c6d9aea4f9a9f3a7625552c93fc29e963d059d40f4def9d77ea7b","src/external_trait_impls/mod.rs":"d69528827794524cfd9acbeacc1ac4f6131e3c7574311e6d919f818f65fbff07","src/external_trait_impls/rayon/helpers.rs":"ba105bf0853ebc45157f22116ad0f55d3bdab75e721d8e7a677c7b912d0c0c6d","src/external_trait_impls/rayon/map.rs":"2809e2a0071db8101c38789deb955f3830c5c3455eb1794ff64a0cf2ceb53fc7","src/external_trait_impls/rayon/mod.rs":"156de9c1ad0123334ea3b7e5a17444faf1b8bf971aa88a1f23e2f2d1c3021141","src/external_trait_impls/rayon/raw.rs":"e62c5f3ca5fffea47357e64b6f8c34cec94af62d9bd28a2b87934da46c22b66e","src/external_trait_impls/rayon/set.rs":"c4c44d44e56c2f59e9e1355662e29d8744ac96645ca4414127a359fb46cb0fbf","src/external_trait_impls/serde.rs":"0bc1a1f218d1ae7a5262557a5e3737b9334caf7d50c136dbdc75ff75680c223b","src/lib.rs":"486e1ea3445e8438f0fb6256471d80621f6c59195f0c198d02c7d951838dd35b","src/macros.rs":"36fe532656879c80f7753d13354b889f5b45caef451a1bb3a27dbc32d74c9878","src/map.rs":"83748649c382b608de5f57e7d64a30455ff4270a24ada22159f9a6d8e084f202","src/raw/alloc.rs":"184a0345bc2c7544b65c28724063be26b1f2b28dbaaa028a0b01192ccac25557","src/raw/bitmask.rs":"820d90b19b7e3433a1048ace008c9526331cd53a576cb0cfc1ff9960b6fe52f8","src/raw/generic.rs":"51720f27d4b76ab411a9658affd0c6faf423402c9def0879481657dd7b1a7928","src/raw/mod.rs":"156c574d7e64f845902ac66032be80356acb1bca758839a92f660a233da91108","src/raw/sse2.rs":"838cfdb1daa1e70951ed25f985283b8b7ab4b46fa130f92eda152047ce6086f6","src/rustc_entry.rs":"cdd70972cba5b79ca1cad79869cb5e184d6dc4798ef90822e966ef89679ba011","src/scopeguard.rs":"1a246e08a63c06cd8ad934bd7da229421bf804f991ae93cd7e242da27ca6c601","src/set.rs":"3e8e6ceb1728dd9a2fa0db1cf351413a608dc73df905a1546e268d0e8e527493","tests/equivalent_trait.rs":"84faa3fe9d67c375d03fec81f0f1412c47862477d42e84e7d235258236338d5b","tests/hasher.rs":"9a8fdf67e4415618e16729969c386eefe71408cded5d46cf7b67d969276a3452","tests/raw.rs":"43ed2f98877533a0905611d9a30f26b183dd3e103e3856eeab80e7b8ac7894d3","tests/rayon.rs":"83d5289771542203f539a41cccb889fbe7ce70f5adf5b903ac9f051e3ba13cfa","tests/serde.rs":"6bac8054db722dd049901b37a6e006535bac30f425eb5cd91af19b5bc1dfe78e","tests/set.rs":"01cf39efb04646ef4c63a809ebb96dfa63cfec472bf8bdb6c121f6526d40c40e"},"package":"43a3c133739dddd0d2990f9a4bdf8eb4b21ef50e4851ca85ab661199821d510e"} \ No newline at end of file +{"files":{"CHANGELOG.md":"f4769b4c6f44e09c379f55694f89a189620a78586a030a6c6c8d183265c31c52","Cargo.toml":"7a3568541b22e0e7dd0a8c2227699c84474673870f5e46dbb7db61ac2eac391f","LICENSE-APACHE":"a60eea817514531668d7e00765731449fe14d059d3249e0bc93b36de45f759f2","LICENSE-MIT":"ff8f68cb076caf8cefe7a6430d4ac086ce6af2ca8ce2c4e5a2004d4552ef52a2","README.md":"00e45d59f6f8537aa0f8e3dec17e24e9838b52f35aa9c1815c71ab1e8f63888e","benches/bench.rs":"ef7bc025922f077d307c565640c005d056e3d6c1713448a95aae92d3c22c1005","benches/insert_unique_unchecked.rs":"cb84275f22d5f95a5ac995ac6b2df74ffcf342765b401d27c95f2955c7b7cb9f","clippy.toml":"7535949f908c6d9aea4f9a9f3a7625552c93fc29e963d059d40f4def9d77ea7b","src/external_trait_impls/mod.rs":"0625e6a5e3b8ecc8901a12aeeea54393fd84617fb3a14d98a34d2d2bddb8d257","src/external_trait_impls/rayon/helpers.rs":"ba105bf0853ebc45157f22116ad0f55d3bdab75e721d8e7a677c7b912d0c0c6d","src/external_trait_impls/rayon/map.rs":"c0f50c8c6f2f70c70994a3243d92de3bbda5e78519c906c4f81f207ed63e5cc3","src/external_trait_impls/rayon/mod.rs":"156de9c1ad0123334ea3b7e5a17444faf1b8bf971aa88a1f23e2f2d1c3021141","src/external_trait_impls/rayon/raw.rs":"e62c5f3ca5fffea47357e64b6f8c34cec94af62d9bd28a2b87934da46c22b66e","src/external_trait_impls/rayon/set.rs":"c4c44d44e56c2f59e9e1355662e29d8744ac96645ca4414127a359fb46cb0fbf","src/external_trait_impls/rkyv/hash_map.rs":"7abe24318143b776016052b05840656afc858b1ba5252f3d418d61972477f53d","src/external_trait_impls/rkyv/hash_set.rs":"38d969125d17d606492ec4ec9fc06b7e7118eb903240dacf40de21b9b06fa5c8","src/external_trait_impls/rkyv/mod.rs":"54399ce5574fd1d84b7b0cb4238fa3e898575e89a6724299be009d2172bda02e","src/external_trait_impls/serde.rs":"0bc1a1f218d1ae7a5262557a5e3737b9334caf7d50c136dbdc75ff75680c223b","src/lib.rs":"662765875308544b71a46e20f18782e7a3246fddb11496206e296a24b78b56a5","src/macros.rs":"98a26b908fc0fbe6a58d008a317e550013d615eb3cc17a5054a573c62c1d74cb","src/map.rs":"4f4bdc2a2eb3c4395d655f5ce12d217b82f6edeef3579efcb6b18d08345f1d52","src/raw/alloc.rs":"902f8588d0fdee3e5c3dc02410f41d4b38ac88843727387f929f3186b3a2d322","src/raw/bitmask.rs":"3b3dce8d6a48856ada19085abf43908f124ab3419fcf434b9ca64d7bff243f67","src/raw/generic.rs":"efc5e603be3e9a17935aef1836a38ce01c78a0093b2af0671548eb5459b37921","src/raw/mod.rs":"cecbe517b36042094818887d41226bdafd78c6f67657540a8c1ea020cd5d302f","src/raw/neon.rs":"9907d8ebc36fc3df562dde478ea9b72213fda65288a304718d8647f0029dc9ad","src/raw/sse2.rs":"39038e3344e49f4638e211bcdbf56565ac53e90dce56172cc3b526fea911c2af","src/rustc_entry.rs":"19d3346843bc62c7c0165e8824d26355ab2666086f3088b1150a8b3f59376a76","src/scopeguard.rs":"1a246e08a63c06cd8ad934bd7da229421bf804f991ae93cd7e242da27ca6c601","src/set.rs":"349a1523656a8a3f364b5313d98a969444717461a7ab7133c8e5f215ac2c329d","tests/equivalent_trait.rs":"84faa3fe9d67c375d03fec81f0f1412c47862477d42e84e7d235258236338d5b","tests/hasher.rs":"9a8fdf67e4415618e16729969c386eefe71408cded5d46cf7b67d969276a3452","tests/raw.rs":"43ed2f98877533a0905611d9a30f26b183dd3e103e3856eeab80e7b8ac7894d3","tests/rayon.rs":"39cb24ab45fce8087bb54948715c8b6973ebfba1a325292b5b3cd9aab50b5fd2","tests/serde.rs":"6bac8054db722dd049901b37a6e006535bac30f425eb5cd91af19b5bc1dfe78e","tests/set.rs":"9f8011c29d1059aadb54b6dd4623521d5178b4278b4a56021ef2cee4bbb19fd9"},"package":"2c6201b9ff9fd90a5a3bac2e56a830d0caa509576f0e503818ee82c181b3437a"} \ No newline at end of file diff --git a/vendor/hashbrown/CHANGELOG.md b/vendor/hashbrown/CHANGELOG.md index 960f9387a..8837c51ca 100644 --- a/vendor/hashbrown/CHANGELOG.md +++ b/vendor/hashbrown/CHANGELOG.md @@ -7,6 +7,43 @@ and this project adheres to [Semantic Versioning](https://semver.org/). ## [Unreleased] +## [v0.14.0] - 2023-06-01 + +### Added + +- Support for `allocator-api2` crate + for interfacing with custom allocators on stable. (#417) +- Optimized implementation for ARM using NEON instructions. (#430) +- Support for rkyv serialization. (#432) +- `Equivalent` trait to look up values without `Borrow`. (#345) +- `Hash{Map,Set}::raw_table_mut` is added whic returns a mutable reference. (#404) +- Fast path for `clear` on empty tables. (#428) + +### Changed + +- Optimized insertion to only perform a single lookup. (#277) +- `DrainFilter` has been renamed to `ExtractIf` and no longer drops remaining + elements when the iterator is dropped. #(374) +- Bumped MSRV to 1.64.0. (#431) +- `{Map,Set}::raw_table` now returns an immutable reference. (#404) +- `VacantEntry` and `OccupiedEntry` now use the default hasher if none is + specified in generics. (#389) +- `RawTable::data_start` now returns a `NonNull` to match `RawTable::data_end`. (#387) +- `RawIter::{reflect_insert, reflect_remove}` are now unsafe. (#429) +- `RawTable::find_potential` is renamed to `find_or_find_insert_slot` and returns an `InsertSlot`. (#429) +- `RawTable::remove` now also returns an `InsertSlot`. (#429) +- `InsertSlot` can be used to insert an element with `RawTable::insert_in_slot`. (#429) +- `RawIterHash` no longer has a lifetime tied to that of the `RawTable`. (#427) +- The trait bounds of `HashSet::raw_table` have been relaxed to not require `Eq + Hash`. (#423) +- `EntryRef::and_replace_entry_with` and `OccupiedEntryRef::replace_entry_with` + were changed to give a `&K` instead of a `&Q` to the closure. + +### Removed + +- Support for `bumpalo` as an allocator with custom wrapper. + Use `allocator-api2` feature in `bumpalo` to use it as an allocator + for `hashbrown` collections. (#417) + ## [v0.13.2] - 2023-01-12 ### Fixed @@ -248,7 +285,7 @@ This release was _yanked_ due to inconsistent hashes being generated with the `n ## [v0.6.2] - 2019-10-23 ### Added -- Added an `inline-more` feature (enabled by default) which allows choosing a tradeoff between +- Added an `inline-more` feature (enabled by default) which allows choosing a tradeoff between runtime performance and compilation time. (#119) ## [v0.6.1] - 2019-10-04 @@ -396,7 +433,8 @@ This release was _yanked_ due to a breaking change for users of `no-default-feat - Initial release -[Unreleased]: https://github.com/rust-lang/hashbrown/compare/v0.13.2...HEAD +[Unreleased]: https://github.com/rust-lang/hashbrown/compare/v0.14.0...HEAD +[v0.14.0]: https://github.com/rust-lang/hashbrown/compare/v0.13.2...v0.14.0 [v0.13.2]: https://github.com/rust-lang/hashbrown/compare/v0.13.1...v0.13.2 [v0.13.1]: https://github.com/rust-lang/hashbrown/compare/v0.12.3...v0.13.1 [v0.12.3]: https://github.com/rust-lang/hashbrown/compare/v0.12.2...v0.12.3 diff --git a/vendor/hashbrown/Cargo.toml b/vendor/hashbrown/Cargo.toml index c64471533..27e3f9b4f 100644 --- a/vendor/hashbrown/Cargo.toml +++ b/vendor/hashbrown/Cargo.toml @@ -11,9 +11,9 @@ [package] edition = "2021" -rust-version = "1.61.0" +rust-version = "1.64.0" name = "hashbrown" -version = "0.13.2" +version = "0.14.0" authors = ["Amanieu d'Antras "] exclude = [ ".github", @@ -52,9 +52,11 @@ version = "1.0.0" optional = true package = "rustc-std-workspace-alloc" -[dependencies.bumpalo] -version = "3.5.0" +[dependencies.allocator-api2] +version = "0.2.9" +features = ["alloc"] optional = true +default-features = false [dependencies.compiler_builtins] version = "0.1.2" @@ -69,11 +71,21 @@ package = "rustc-std-workspace-core" version = "1.0" optional = true +[dependencies.rkyv] +version = "0.7.42" +features = ["alloc"] +optional = true +default-features = false + [dependencies.serde] version = "1.0.25" optional = true default-features = false +[dev-dependencies.bumpalo] +version = "3.13.0" +features = ["allocator-api2"] + [dev-dependencies.doc-comment] version = "0.3.1" @@ -90,6 +102,10 @@ features = ["small_rng"] [dev-dependencies.rayon] version = "1.0" +[dev-dependencies.rkyv] +version = "0.7.42" +features = ["validation"] + [dev-dependencies.serde_test] version = "1.0" @@ -97,9 +113,13 @@ version = "1.0" default = [ "ahash", "inline-more", + "allocator-api2", ] inline-more = [] -nightly = [] +nightly = [ + "allocator-api2?/nightly", + "bumpalo/allocator_api", +] raw = [] rustc-dep-of-std = [ "nightly", diff --git a/vendor/hashbrown/README.md b/vendor/hashbrown/README.md index 9b2ca7eae..d0c4261e9 100644 --- a/vendor/hashbrown/README.md +++ b/vendor/hashbrown/README.md @@ -4,7 +4,7 @@ hashbrown [![Build Status](https://github.com/rust-lang/hashbrown/actions/workflows/rust.yml/badge.svg)](https://github.com/rust-lang/hashbrown/actions) [![Crates.io](https://img.shields.io/crates/v/hashbrown.svg)](https://crates.io/crates/hashbrown) [![Documentation](https://docs.rs/hashbrown/badge.svg)](https://docs.rs/hashbrown) -[![Rust](https://img.shields.io/badge/rust-1.61.0%2B-blue.svg?maxAge=3600)](https://github.com/rust-lang/hashbrown) +[![Rust](https://img.shields.io/badge/rust-1.64.0%2B-blue.svg?maxAge=3600)](https://github.com/rust-lang/hashbrown) 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` @@ -40,44 +40,44 @@ Compared to the previous implementation of `std::collections::HashMap` (Rust 1.3 With the hashbrown default AHash hasher: -| name | oldstdhash ns/iter | hashbrown ns/iter | diff ns/iter | diff % | speedup | -|:------------------------|:-------------------:|------------------:|:------------:|---------:|---------| -| insert_ahash_highbits | 18,865 | 8,020 | -10,845 | -57.49% | x 2.35 | -| insert_ahash_random | 19,711 | 8,019 | -11,692 | -59.32% | x 2.46 | -| insert_ahash_serial | 19,365 | 6,463 | -12,902 | -66.63% | x 3.00 | -| insert_erase_ahash_highbits | 51,136 | 17,916 | -33,220 | -64.96% | x 2.85 | -| insert_erase_ahash_random | 51,157 | 17,688 | -33,469 | -65.42% | x 2.89 | -| insert_erase_ahash_serial | 45,479 | 14,895 | -30,584 | -67.25% | x 3.05 | -| iter_ahash_highbits | 1,399 | 1,092 | -307 | -21.94% | x 1.28 | -| iter_ahash_random | 1,586 | 1,059 | -527 | -33.23% | x 1.50 | -| iter_ahash_serial | 3,168 | 1,079 | -2,089 | -65.94% | x 2.94 | -| lookup_ahash_highbits | 32,351 | 4,792 | -27,559 | -85.19% | x 6.75 | -| lookup_ahash_random | 17,419 | 4,817 | -12,602 | -72.35% | x 3.62 | -| lookup_ahash_serial | 15,254 | 3,606 | -11,648 | -76.36% | x 4.23 | -| lookup_fail_ahash_highbits | 21,187 | 4,369 | -16,818 | -79.38% | x 4.85 | -| lookup_fail_ahash_random | 21,550 | 4,395 | -17,155 | -79.61% | x 4.90 | -| lookup_fail_ahash_serial | 19,450 | 3,176 | -16,274 | -83.67% | x 6.12 | +| name | oldstdhash ns/iter | hashbrown ns/iter | diff ns/iter | diff % | speedup | +| :-------------------------- | :----------------: | ----------------: | :----------: | ------: | ------- | +| insert_ahash_highbits | 18,865 | 8,020 | -10,845 | -57.49% | x 2.35 | +| insert_ahash_random | 19,711 | 8,019 | -11,692 | -59.32% | x 2.46 | +| insert_ahash_serial | 19,365 | 6,463 | -12,902 | -66.63% | x 3.00 | +| insert_erase_ahash_highbits | 51,136 | 17,916 | -33,220 | -64.96% | x 2.85 | +| insert_erase_ahash_random | 51,157 | 17,688 | -33,469 | -65.42% | x 2.89 | +| insert_erase_ahash_serial | 45,479 | 14,895 | -30,584 | -67.25% | x 3.05 | +| iter_ahash_highbits | 1,399 | 1,092 | -307 | -21.94% | x 1.28 | +| iter_ahash_random | 1,586 | 1,059 | -527 | -33.23% | x 1.50 | +| iter_ahash_serial | 3,168 | 1,079 | -2,089 | -65.94% | x 2.94 | +| lookup_ahash_highbits | 32,351 | 4,792 | -27,559 | -85.19% | x 6.75 | +| lookup_ahash_random | 17,419 | 4,817 | -12,602 | -72.35% | x 3.62 | +| lookup_ahash_serial | 15,254 | 3,606 | -11,648 | -76.36% | x 4.23 | +| lookup_fail_ahash_highbits | 21,187 | 4,369 | -16,818 | -79.38% | x 4.85 | +| lookup_fail_ahash_random | 21,550 | 4,395 | -17,155 | -79.61% | x 4.90 | +| lookup_fail_ahash_serial | 19,450 | 3,176 | -16,274 | -83.67% | x 6.12 | With the libstd default SipHash hasher: -|name | oldstdhash ns/iter | hashbrown ns/iter | diff ns/iter | diff % | speedup | -|:------------------------|:-------------------:|------------------:|:------------:|---------:|---------| -|insert_std_highbits |19,216 |16,885 | -2,331 | -12.13% | x 1.14 | -|insert_std_random |19,179 |17,034 | -2,145 | -11.18% | x 1.13 | -|insert_std_serial |19,462 |17,493 | -1,969 | -10.12% | x 1.11 | -|insert_erase_std_highbits |50,825 |35,847 | -14,978 | -29.47% | x 1.42 | -|insert_erase_std_random |51,448 |35,392 | -16,056 | -31.21% | x 1.45 | -|insert_erase_std_serial |87,711 |38,091 | -49,620 | -56.57% | x 2.30 | -|iter_std_highbits |1,378 |1,159 | -219 | -15.89% | x 1.19 | -|iter_std_random |1,395 |1,132 | -263 | -18.85% | x 1.23 | -|iter_std_serial |1,704 |1,105 | -599 | -35.15% | x 1.54 | -|lookup_std_highbits |17,195 |13,642 | -3,553 | -20.66% | x 1.26 | -|lookup_std_random |17,181 |13,773 | -3,408 | -19.84% | x 1.25 | -|lookup_std_serial |15,483 |13,651 | -1,832 | -11.83% | x 1.13 | -|lookup_fail_std_highbits |20,926 |13,474 | -7,452 | -35.61% | x 1.55 | -|lookup_fail_std_random |21,766 |13,505 | -8,261 | -37.95% | x 1.61 | -|lookup_fail_std_serial |19,336 |13,519 | -5,817 | -30.08% | x 1.43 | +| name | oldstdhash ns/iter | hashbrown ns/iter | diff ns/iter | diff % | speedup | +| :------------------------ | :----------------: | ----------------: | :----------: | ------: | ------- | +| insert_std_highbits | 19,216 | 16,885 | -2,331 | -12.13% | x 1.14 | +| insert_std_random | 19,179 | 17,034 | -2,145 | -11.18% | x 1.13 | +| insert_std_serial | 19,462 | 17,493 | -1,969 | -10.12% | x 1.11 | +| insert_erase_std_highbits | 50,825 | 35,847 | -14,978 | -29.47% | x 1.42 | +| insert_erase_std_random | 51,448 | 35,392 | -16,056 | -31.21% | x 1.45 | +| insert_erase_std_serial | 87,711 | 38,091 | -49,620 | -56.57% | x 2.30 | +| iter_std_highbits | 1,378 | 1,159 | -219 | -15.89% | x 1.19 | +| iter_std_random | 1,395 | 1,132 | -263 | -18.85% | x 1.23 | +| iter_std_serial | 1,704 | 1,105 | -599 | -35.15% | x 1.54 | +| lookup_std_highbits | 17,195 | 13,642 | -3,553 | -20.66% | x 1.26 | +| lookup_std_random | 17,181 | 13,773 | -3,408 | -19.84% | x 1.25 | +| lookup_std_serial | 15,483 | 13,651 | -1,832 | -11.83% | x 1.13 | +| lookup_fail_std_highbits | 20,926 | 13,474 | -7,452 | -35.61% | x 1.55 | +| lookup_fail_std_random | 21,766 | 13,505 | -8,261 | -37.95% | x 1.61 | +| lookup_fail_std_serial | 19,336 | 13,519 | -5,817 | -30.08% | x 1.43 | ## Usage @@ -85,7 +85,7 @@ Add this to your `Cargo.toml`: ```toml [dependencies] -hashbrown = "0.13" +hashbrown = "0.14" ``` Then: @@ -101,12 +101,13 @@ This crate has the following Cargo features: - `nightly`: Enables nightly-only features including: `#[may_dangle]`. - `serde`: Enables serde serialization support. +- `rkyv`: Enables rkyv serialization support. - `rayon`: Enables rayon parallel iterator support. - `raw`: Enables access to the experimental and unsafe `RawTable` API. - `inline-more`: Adds inline hints to most functions, improving run-time performance at the cost of compilation time. (enabled by default) -- `bumpalo`: Provides a `BumpWrapper` type which allows `bumpalo` to be used for memory allocation. - `ahash`: Compiles with ahash as default hasher. (enabled by default) +- `allocator-api2`: Enables support for allocators that support `allocator-api2`. (enabled by default) ## License diff --git a/vendor/hashbrown/benches/bench.rs b/vendor/hashbrown/benches/bench.rs index c393b9a70..346bd7ef8 100644 --- a/vendor/hashbrown/benches/bench.rs +++ b/vendor/hashbrown/benches/bench.rs @@ -311,7 +311,7 @@ fn rehash_in_place(b: &mut Bencher) { // Each loop triggers one rehash for _ in 0..10 { - for i in 0..224 { + for i in 0..223 { set.insert(i); } diff --git a/vendor/hashbrown/src/external_trait_impls/mod.rs b/vendor/hashbrown/src/external_trait_impls/mod.rs index ef497836c..01d386b04 100644 --- a/vendor/hashbrown/src/external_trait_impls/mod.rs +++ b/vendor/hashbrown/src/external_trait_impls/mod.rs @@ -1,4 +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/map.rs b/vendor/hashbrown/src/external_trait_impls/rayon/map.rs index 14d91c220..1124bfd32 100644 --- a/vendor/hashbrown/src/external_trait_impls/rayon/map.rs +++ b/vendor/hashbrown/src/external_trait_impls/rayon/map.rs @@ -561,10 +561,7 @@ mod test_par_map { assert_eq!(value.load(Ordering::Relaxed), 100); // retain only half - let _v: Vec<_> = hm - .into_par_iter() - .filter(|&(ref key, _)| key.k < 50) - .collect(); + 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); @@ -611,7 +608,7 @@ mod test_par_map { assert_eq!(value.load(Ordering::Relaxed), 100); // retain only half - let _v: Vec<_> = hm.drain().filter(|&(ref key, _)| key.k < 50).collect(); + let _v: Vec<_> = hm.drain().filter(|(key, _)| key.k < 50).collect(); assert!(hm.is_empty()); assert_eq!(key.load(Ordering::Relaxed), 50); 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 000000000..fae7f7676 --- /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 Archive for HashMap +where + K::Archived: Hash + Eq, +{ + type Archived = ArchivedHashMap; + 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 Serialize for HashMap +where + K: Serialize + Hash + Eq, + K::Archived: Hash + Eq, + V: Serialize, + S: Serializer + ScratchSpace + ?Sized, +{ + #[inline] + fn serialize(&self, serializer: &mut S) -> Result { + unsafe { ArchivedHashMap::serialize_from_iter(self.iter(), serializer) } + } +} + +impl + Deserialize, D> for ArchivedHashMap +where + K::Archived: Deserialize + Hash + Eq, + V::Archived: Deserialize, +{ + #[inline] + fn deserialize(&self, deserializer: &mut D) -> Result, 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, V, AK: Hash + Eq, AV: PartialEq, S: BuildHasher> + PartialEq> for ArchivedHashMap +{ + #[inline] + fn eq(&self, other: &HashMap) -> bool { + if self.len() != other.len() { + false + } else { + self.iter() + .all(|(key, value)| other.get(key).map_or(false, |v| value.eq(v))) + } + } +} + +impl, V, AK: Hash + Eq, AV: PartialEq> + PartialEq> for HashMap +{ + #[inline] + fn eq(&self, other: &ArchivedHashMap) -> 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::>(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 = 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::>(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 000000000..c8a69cf4f --- /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 Archive for HashSet +where + K::Archived: Hash + Eq, +{ + type Archived = ArchivedHashSet; + type Resolver = HashSetResolver; + + #[inline] + unsafe fn resolve(&self, pos: usize, resolver: Self::Resolver, out: *mut Self::Archived) { + ArchivedHashSet::::resolve_from_len(self.len(), pos, resolver, out); + } +} + +impl Serialize for HashSet +where + K::Archived: Hash + Eq, + K: Serialize + Hash + Eq, + S: ScratchSpace + Serializer + ?Sized, +{ + #[inline] + fn serialize(&self, serializer: &mut S) -> Result { + unsafe { ArchivedHashSet::serialize_from_iter(self.iter(), serializer) } + } +} + +impl Deserialize, D> for ArchivedHashSet +where + K: Archive + Hash + Eq, + K::Archived: Deserialize + Hash + Eq, + D: Fallible + ?Sized, + S: Default + BuildHasher, +{ + #[inline] + fn deserialize(&self, deserializer: &mut D) -> Result, D::Error> { + let mut result = HashSet::with_hasher(S::default()); + for k in self.iter() { + result.insert(k.deserialize(deserializer)?); + } + Ok(result) + } +} + +impl, AK: Hash + Eq, S: BuildHasher> PartialEq> + for ArchivedHashSet +{ + #[inline] + fn eq(&self, other: &HashSet) -> bool { + if self.len() != other.len() { + false + } else { + self.iter().all(|key| other.get(key).is_some()) + } + } +} + +impl, AK: Hash + Eq, S: BuildHasher> PartialEq> + for HashSet +{ + #[inline] + fn eq(&self, other: &ArchivedHashSet) -> 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::>(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 = 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::>(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 000000000..2bde6a065 --- /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/lib.rs b/vendor/hashbrown/src/lib.rs index e43165dd6..013a9ddd9 100644 --- a/vendor/hashbrown/src/lib.rs +++ b/vendor/hashbrown/src/lib.rs @@ -20,9 +20,8 @@ extend_one, allocator_api, slice_ptr_get, - nonnull_slice_from_raw_parts, maybe_uninit_array_assume_init, - build_hasher_simple_hash_one + strict_provenance ) )] #![allow( @@ -37,6 +36,7 @@ )] #![warn(missing_docs)] #![warn(rust_2018_idioms)] +#![cfg_attr(feature = "nightly", warn(fuzzy_provenance_casts))] #[cfg(test)] #[macro_use] @@ -163,21 +163,3 @@ pub enum TryReserveError { layout: alloc::alloc::Layout, }, } - -/// Wrapper around `Bump` which allows it to be used as an allocator for -/// `HashMap`, `HashSet` and `RawTable`. -/// -/// `Bump` can be used directly without this wrapper on nightly if you enable -/// the `allocator-api` feature of the `bumpalo` crate. -#[cfg(feature = "bumpalo")] -#[derive(Clone, Copy, Debug)] -pub struct BumpWrapper<'a>(pub &'a bumpalo::Bump); - -#[cfg(feature = "bumpalo")] -#[test] -fn test_bumpalo() { - use bumpalo::Bump; - let bump = Bump::new(); - let mut map = HashMap::new_in(BumpWrapper(&bump)); - map.insert(0, 1); -} diff --git a/vendor/hashbrown/src/macros.rs b/vendor/hashbrown/src/macros.rs index f8ef917b1..eaba6bed1 100644 --- a/vendor/hashbrown/src/macros.rs +++ b/vendor/hashbrown/src/macros.rs @@ -37,7 +37,7 @@ macro_rules! cfg_if { // 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 approprate #[cfg]. The + // 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)* } diff --git a/vendor/hashbrown/src/map.rs b/vendor/hashbrown/src/map.rs index e238bf66b..548ca0f9e 100644 --- a/vendor/hashbrown/src/map.rs +++ b/vendor/hashbrown/src/map.rs @@ -260,29 +260,6 @@ where hash_builder.hash_one(val) } -#[cfg(not(feature = "nightly"))] -#[cfg_attr(feature = "inline-more", inline)] -pub(crate) fn make_insert_hash(hash_builder: &S, val: &K) -> u64 -where - K: Hash, - 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_insert_hash(hash_builder: &S, val: &K) -> u64 -where - K: Hash, - S: BuildHasher, -{ - hash_builder.hash_one(val) -} - #[cfg(feature = "ahash")] impl HashMap { /// Creates an empty `HashMap`. @@ -368,13 +345,11 @@ impl HashMap { /// # Examples /// /// ``` - /// # #[cfg(feature = "bumpalo")] - /// # fn test() { - /// use hashbrown::{HashMap, BumpWrapper}; + /// use hashbrown::HashMap; /// use bumpalo::Bump; /// /// let bump = Bump::new(); - /// let mut map = HashMap::new_in(BumpWrapper(&bump)); + /// let mut map = HashMap::new_in(&bump); /// /// // The created HashMap holds none elements /// assert_eq!(map.len(), 0); @@ -388,11 +363,6 @@ impl HashMap { /// assert_eq!(map.len(), 1); /// // And it also allocates some capacity /// assert!(map.capacity() > 1); - /// # } - /// # fn main() { - /// # #[cfg(feature = "bumpalo")] - /// # test() - /// # } /// ``` #[cfg_attr(feature = "inline-more", inline)] pub fn new_in(alloc: A) -> Self { @@ -419,13 +389,11 @@ impl HashMap { /// # Examples /// /// ``` - /// # #[cfg(feature = "bumpalo")] - /// # fn test() { - /// use hashbrown::{HashMap, BumpWrapper}; + /// use hashbrown::HashMap; /// use bumpalo::Bump; /// /// let bump = Bump::new(); - /// let mut map = HashMap::with_capacity_in(5, BumpWrapper(&bump)); + /// let mut map = HashMap::with_capacity_in(5, &bump); /// /// // The created HashMap holds none elements /// assert_eq!(map.len(), 0); @@ -444,11 +412,6 @@ impl HashMap { /// assert_eq!(map.len(), 5); /// // But its capacity isn't changed /// assert_eq!(map.capacity(), empty_map_capacity) - /// # } - /// # fn main() { - /// # #[cfg(feature = "bumpalo")] - /// # test() - /// # } /// ``` #[cfg_attr(feature = "inline-more", inline)] pub fn with_capacity_in(capacity: usize, alloc: A) -> Self { @@ -972,15 +935,12 @@ impl HashMap { /// In other words, move all pairs `(k, v)` such that `f(&k, &mut v)` returns `true` out /// into another iterator. /// - /// Note that `drain_filter` lets you mutate every value in the filter closure, regardless of + /// Note that `extract_if` lets you mutate every value in the filter closure, regardless of /// whether you choose to keep or remove it. /// - /// When the returned DrainedFilter is dropped, any remaining elements that satisfy - /// the predicate are dropped from the table. - /// - /// It is unspecified how many more elements will be subjected to the closure - /// if a panic occurs in the closure, or a panic occurs while dropping an element, - /// or if the `DrainFilter` value is leaked. + /// 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. /// @@ -991,7 +951,7 @@ impl HashMap { /// /// let mut map: HashMap = (0..8).map(|x| (x, x)).collect(); /// - /// let drained: HashMap = map.drain_filter(|k, _v| k % 2 == 0).collect(); + /// let drained: HashMap = map.extract_if(|k, _v| k % 2 == 0).collect(); /// /// let mut evens = drained.keys().cloned().collect::>(); /// let mut odds = map.keys().cloned().collect::>(); @@ -1004,21 +964,20 @@ impl HashMap { /// let mut map: HashMap = (0..8).map(|x| (x, x)).collect(); /// /// { // Iterator is dropped without being consumed. - /// let d = map.drain_filter(|k, _v| k % 2 != 0); + /// let d = map.extract_if(|k, _v| k % 2 != 0); /// } /// - /// // But the map lens have been reduced by half - /// // even if we do not use DrainFilter iterator. - /// assert_eq!(map.len(), 4); + /// // ExtractIf was not exhausted, therefore no elements were drained. + /// assert_eq!(map.len(), 8); /// ``` #[cfg_attr(feature = "inline-more", inline)] - pub fn drain_filter(&mut self, f: F) -> DrainFilter<'_, K, V, F, A> + pub fn extract_if(&mut self, f: F) -> ExtractIf<'_, K, V, F, A> where F: FnMut(&K, &mut V) -> bool, { - DrainFilter { + ExtractIf { f, - inner: DrainFilterInner { + inner: ExtractIfInner { iter: unsafe { self.table.iter() }, table: &mut self.table, }, @@ -1266,7 +1225,7 @@ where /// ``` #[cfg_attr(feature = "inline-more", inline)] pub fn entry(&mut self, key: K) -> Entry<'_, K, V, S, A> { - let hash = make_insert_hash::(&self.hash_builder, &key); + let hash = make_hash::(&self.hash_builder, &key); if let Some(elem) = self.table.find(hash, equivalent_key(&key)) { Entry::Occupied(OccupiedEntry { hash, @@ -1348,7 +1307,7 @@ where { // Avoid `Option::map` because it bloats LLVM IR. match self.get_inner(k) { - Some(&(_, ref v)) => Some(v), + Some((_, v)) => Some(v), None => None, } } @@ -1379,7 +1338,7 @@ where { // Avoid `Option::map` because it bloats LLVM IR. match self.get_inner(k) { - Some(&(ref key, ref value)) => Some((key, value)), + Some((key, value)) => Some((key, value)), None => None, } } @@ -1786,13 +1745,19 @@ where /// ``` #[cfg_attr(feature = "inline-more", inline)] pub fn insert(&mut self, k: K, v: V) -> Option { - let hash = make_insert_hash::(&self.hash_builder, &k); - if let Some((_, item)) = self.table.get_mut(hash, equivalent_key(&k)) { - Some(mem::replace(item, v)) - } else { - self.table - .insert(hash, (k, v), make_hasher::<_, V, S>(&self.hash_builder)); - None + let hash = make_hash::(&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 + } } } @@ -1847,7 +1812,7 @@ where /// ``` #[cfg_attr(feature = "inline-more", inline)] pub fn insert_unique_unchecked(&mut self, k: K, v: V) -> (&K, &mut V) { - let hash = make_insert_hash::(&self.hash_builder, &k); + let hash = make_hash::(&self.hash_builder, &k); let bucket = self .table .insert(hash, (k, v), make_hasher::<_, V, S>(&self.hash_builder)); @@ -2123,6 +2088,18 @@ impl HashMap { 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. /// @@ -2159,9 +2136,9 @@ impl HashMap { /// where /// F: Fn(&(K, V)) -> bool, /// { - /// let raw_table = map.raw_table(); + /// let raw_table = map.raw_table_mut(); /// match raw_table.find(hash, is_match) { - /// Some(bucket) => Some(unsafe { raw_table.remove(bucket) }), + /// Some(bucket) => Some(unsafe { raw_table.remove(bucket).0 }), /// None => None, /// } /// } @@ -2180,7 +2157,7 @@ impl HashMap { /// ``` #[cfg(feature = "raw")] #[cfg_attr(feature = "inline-more", inline)] - pub fn raw_table(&mut self) -> &mut RawTable<(K, V), A> { + pub fn raw_table_mut(&mut self) -> &mut RawTable<(K, V), A> { &mut self.table } } @@ -2711,10 +2688,10 @@ impl Drain<'_, K, V, A> { /// 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 [`drain_filter`] method on [`HashMap`]. See its +/// This `struct` is created by the [`extract_if`] method on [`HashMap`]. See its /// documentation for more. /// -/// [`drain_filter`]: struct.HashMap.html#method.drain_filter +/// [`extract_if`]: struct.HashMap.html#method.extract_if /// [`HashMap`]: struct.HashMap.html /// /// # Examples @@ -2724,54 +2701,31 @@ impl Drain<'_, K, V, A> { /// /// let mut map: HashMap = [(1, "a"), (2, "b"), (3, "c")].into(); /// -/// let mut drain_filter = map.drain_filter(|k, _v| k % 2 != 0); -/// let mut vec = vec![drain_filter.next(), drain_filter.next()]; +/// let mut extract_if = map.extract_if(|k, _v| k % 2 != 0); +/// let mut vec = vec![extract_if.next(), extract_if.next()]; /// -/// // The `DrainFilter` iterator produces items in arbitrary order, so the +/// // 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!(drain_filter.next(), None); -/// assert_eq!(drain_filter.next(), None); -/// drop(drain_filter); +/// assert_eq!(extract_if.next(), None); +/// assert_eq!(extract_if.next(), None); +/// drop(extract_if); /// /// assert_eq!(map.len(), 1); /// ``` -pub struct DrainFilter<'a, K, V, F, A: Allocator + Clone = Global> +#[must_use = "Iterators are lazy unless consumed"] +pub struct ExtractIf<'a, K, V, F, A: Allocator + Clone = Global> where F: FnMut(&K, &mut V) -> bool, { f: F, - inner: DrainFilterInner<'a, K, V, A>, -} - -impl<'a, K, V, F, A> Drop for DrainFilter<'a, K, V, F, A> -where - F: FnMut(&K, &mut V) -> bool, - A: Allocator + Clone, -{ - #[cfg_attr(feature = "inline-more", inline)] - fn drop(&mut self) { - while let Some(item) = self.next() { - let guard = ConsumeAllOnDrop(self); - drop(item); - mem::forget(guard); - } - } + inner: ExtractIfInner<'a, K, V, A>, } -pub(super) struct ConsumeAllOnDrop<'a, T: Iterator>(pub &'a mut T); - -impl Drop for ConsumeAllOnDrop<'_, T> { - #[cfg_attr(feature = "inline-more", inline)] - fn drop(&mut self) { - self.0.for_each(drop); - } -} - -impl Iterator for DrainFilter<'_, K, V, F, A> +impl Iterator for ExtractIf<'_, K, V, F, A> where F: FnMut(&K, &mut V) -> bool, A: Allocator + Clone, @@ -2789,15 +2743,15 @@ where } } -impl FusedIterator for DrainFilter<'_, K, V, F> where F: FnMut(&K, &mut V) -> bool {} +impl FusedIterator for ExtractIf<'_, K, V, F> where F: FnMut(&K, &mut V) -> bool {} -/// Portions of `DrainFilter` shared with `set::DrainFilter` -pub(super) struct DrainFilterInner<'a, K, V, A: Allocator + Clone> { +/// Portions of `ExtractIf` shared with `set::ExtractIf` +pub(super) struct ExtractIfInner<'a, K, V, A: Allocator + Clone> { pub iter: RawIter<(K, V)>, pub table: &'a mut RawTable<(K, V), A>, } -impl DrainFilterInner<'_, K, V, A> { +impl ExtractIfInner<'_, K, V, A> { #[cfg_attr(feature = "inline-more", inline)] pub(super) fn next(&mut self, f: &mut F) -> Option<(K, V)> where @@ -2807,7 +2761,7 @@ impl DrainFilterInner<'_, K, V, A> { 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)); + return Some(self.table.remove(item).0); } } } @@ -3360,7 +3314,7 @@ impl<'a, K, V, S, A: Allocator + Clone> RawEntryBuilder<'a, K, V, S, A> { F: FnMut(&K) -> bool, { match self.map.table.get(hash, |(k, _)| is_match(k)) { - Some(&(ref key, ref value)) => Some((key, value)), + Some((key, value)) => Some((key, value)), None => None, } } @@ -3756,7 +3710,7 @@ impl<'a, K, V, S, A: Allocator + Clone> RawOccupiedEntryMut<'a, K, V, S, A> { #[cfg_attr(feature = "inline-more", inline)] pub fn get_key_value(&self) -> (&K, &V) { unsafe { - let &(ref key, ref value) = self.elem.as_ref(); + let (key, value) = self.elem.as_ref(); (key, value) } } @@ -3928,7 +3882,7 @@ impl<'a, K, V, S, A: Allocator + Clone> RawOccupiedEntryMut<'a, K, V, S, A> { /// ``` #[cfg_attr(feature = "inline-more", inline)] pub fn remove_entry(self) -> (K, V) { - unsafe { self.table.remove(self.elem) } + unsafe { self.table.remove(self.elem).0 } } /// Provides shared access to the key and owned access to the value of @@ -4012,7 +3966,7 @@ impl<'a, K, V, S, A: Allocator + Clone> RawVacantEntryMut<'a, K, V, S, A> { K: Hash, S: BuildHasher, { - let hash = make_insert_hash::(self.hash_builder, &key); + let hash = make_hash::(self.hash_builder, &key); self.insert_hashed_nocheck(hash, key, value) } @@ -4120,7 +4074,7 @@ impl<'a, K, V, S, A: Allocator + Clone> RawVacantEntryMut<'a, K, V, S, A> { K: Hash, S: BuildHasher, { - let hash = make_insert_hash::(self.hash_builder, &key); + let hash = make_hash::(self.hash_builder, &key); let elem = self.table.insert( hash, (key, value), @@ -4297,7 +4251,7 @@ impl Debug for Entry<'_, K, V, S, A /// assert_eq!(map.get(&"c"), None); /// assert_eq!(map.len(), 2); /// ``` -pub struct OccupiedEntry<'a, K, V, S, A: Allocator + Clone = Global> { +pub struct OccupiedEntry<'a, K, V, S = DefaultHashBuilder, A: Allocator + Clone = Global> { hash: u64, key: Option, elem: Bucket<(K, V)>, @@ -4360,7 +4314,7 @@ impl Debug for OccupiedEntry<'_, K, /// } /// assert!(map[&"b"] == 20 && map.len() == 2); /// ``` -pub struct VacantEntry<'a, K, V, S, A: Allocator + Clone = Global> { +pub struct VacantEntry<'a, K, V, S = DefaultHashBuilder, A: Allocator + Clone = Global> { hash: u64, key: K, table: &'a mut HashMap, @@ -4568,7 +4522,7 @@ impl, Q: ?Sized + Debug, V: Debug, S, A: Allocator + Clone> Debug { fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { f.debug_struct("OccupiedEntryRef") - .field("key", &self.key()) + .field("key", &self.key().borrow()) .field("value", &self.get()) .finish() } @@ -5301,7 +5255,7 @@ impl<'a, K, V, S, A: Allocator + Clone> OccupiedEntry<'a, K, V, S, A> { /// ``` #[cfg_attr(feature = "inline-more", inline)] pub fn remove_entry(self) -> (K, V) { - unsafe { self.table.table.remove(self.elem) } + unsafe { self.table.table.remove(self.elem).0 } } /// Gets a reference to the value in the entry. @@ -5838,7 +5792,7 @@ impl<'a, 'b, K, Q: ?Sized, V, S, A: Allocator + Clone> EntryRef<'a, 'b, K, Q, V, K: Borrow, { match *self { - EntryRef::Occupied(ref entry) => entry.key(), + EntryRef::Occupied(ref entry) => entry.key().borrow(), EntryRef::Vacant(ref entry) => entry.key(), } } @@ -5934,8 +5888,7 @@ impl<'a, 'b, K, Q: ?Sized, V, S, A: Allocator + Clone> EntryRef<'a, 'b, K, Q, V, #[cfg_attr(feature = "inline-more", inline)] pub fn and_replace_entry_with(self, f: F) -> Self where - F: FnOnce(&Q, V) -> Option, - K: Borrow, + F: FnOnce(&K, V) -> Option, { match self { EntryRef::Occupied(entry) => entry.replace_entry_with(f), @@ -5994,11 +5947,8 @@ impl<'a, 'b, K, Q: ?Sized, V, S, A: Allocator + Clone> OccupiedEntryRef<'a, 'b, /// } /// ``` #[cfg_attr(feature = "inline-more", inline)] - pub fn key(&self) -> &Q - where - K: Borrow, - { - unsafe { &self.elem.as_ref().0 }.borrow() + pub fn key(&self) -> &K { + unsafe { &self.elem.as_ref().0 } } /// Take the ownership of the key and value from the map. @@ -6027,7 +5977,7 @@ impl<'a, 'b, K, Q: ?Sized, V, S, A: Allocator + Clone> OccupiedEntryRef<'a, 'b, /// ``` #[cfg_attr(feature = "inline-more", inline)] pub fn remove_entry(self) -> (K, V) { - unsafe { self.table.table.remove(self.elem) } + unsafe { self.table.table.remove(self.elem).0 } } /// Gets a reference to the value in the entry. @@ -6303,8 +6253,7 @@ impl<'a, 'b, K, Q: ?Sized, V, S, A: Allocator + Clone> OccupiedEntryRef<'a, 'b, #[cfg_attr(feature = "inline-more", inline)] pub fn replace_entry_with(self, f: F) -> EntryRef<'a, 'b, K, Q, V, S, A> where - F: FnOnce(&Q, V) -> Option, - K: Borrow, + F: FnOnce(&K, V) -> Option, { unsafe { let mut spare_key = None; @@ -6312,7 +6261,7 @@ impl<'a, 'b, K, Q: ?Sized, V, S, A: Allocator + Clone> OccupiedEntryRef<'a, 'b, self.table .table .replace_bucket_with(self.elem.clone(), |(key, value)| { - if let Some(new_value) = f(key.borrow(), value) { + if let Some(new_value) = f(&key, value) { Some((key, new_value)) } else { spare_key = Some(KeyOrRef::Owned(key)); @@ -6926,7 +6875,6 @@ mod test_map { } }); - #[allow(clippy::let_underscore_drop)] // kind-of a false positive for _ in half.by_ref() {} DROP_VECTOR.with(|v| { @@ -7254,10 +7202,10 @@ mod test_map { map.insert(1, 2); map.insert(3, 4); - let map_str = format!("{:?}", map); + let map_str = format!("{map:?}"); assert!(map_str == "{1: 2, 3: 4}" || map_str == "{3: 4, 1: 2}"); - assert_eq!(format!("{:?}", empty), "{}"); + assert_eq!(format!("{empty:?}"), "{}"); } #[test] @@ -7573,7 +7521,7 @@ mod test_map { // Test for #19292 fn check(m: &HashMap) { for k in m.keys() { - assert!(m.contains_key(k), "{} is in keys() but not in the map?", k); + assert!(m.contains_key(k), "{k} is in keys() but not in the map?"); } } @@ -7609,7 +7557,7 @@ mod test_map { // Test for #19292 fn check(m: &HashMap) { for k in m.keys() { - assert!(m.contains_key(k), "{} is in keys() but not in the map?", k); + assert!(m.contains_key(k), "{k} is in keys() but not in the map?"); } } @@ -7658,6 +7606,7 @@ mod test_map { } #[test] + #[allow(clippy::needless_borrow)] fn test_extend_ref_kv_tuple() { use std::ops::AddAssign; let mut a = HashMap::new(); @@ -8080,7 +8029,7 @@ mod test_map { // Test for #19292 fn check(m: &HashMap) { for k in m.keys() { - assert!(m.contains_key(k), "{} is in keys() but not in the map?", k); + assert!(m.contains_key(k), "{k} is in keys() but not in the map?"); } } @@ -8110,7 +8059,7 @@ mod test_map { // Test for #19292 fn check(m: &HashMap) { for k in m.keys() { - assert!(m.contains_key(k), "{} is in keys() but not in the map?", k); + assert!(m.contains_key(k), "{k} is in keys() but not in the map?"); } } @@ -8148,10 +8097,10 @@ mod test_map { } #[test] - fn test_drain_filter() { + fn test_extract_if() { { let mut map: HashMap = (0..8).map(|x| (x, x * 10)).collect(); - let drained = map.drain_filter(|&k, _| k % 2 == 0); + let drained = map.extract_if(|&k, _| k % 2 == 0); let mut out = drained.collect::>(); out.sort_unstable(); assert_eq!(vec![(0, 0), (2, 20), (4, 40), (6, 60)], out); @@ -8159,7 +8108,7 @@ mod test_map { } { let mut map: HashMap = (0..8).map(|x| (x, x * 10)).collect(); - drop(map.drain_filter(|&k, _| k % 2 == 0)); + map.extract_if(|&k, _| k % 2 == 0).for_each(drop); assert_eq!(map.len(), 4); } } @@ -8208,7 +8157,7 @@ mod test_map { let mut map: HashMap<_, _> = xs.iter().copied().collect(); let compute_hash = |map: &HashMap, k: i32| -> u64 { - super::make_insert_hash::(map.hasher(), &k) + super::make_hash::(map.hasher(), &k) }; // Existing key (insert) @@ -8370,17 +8319,17 @@ mod test_map { loop { // occasionally remove some elements if i < n && rng.gen_bool(0.1) { - let hash_value = super::make_insert_hash(&hash_builder, &i); + 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); + let t = map.table.remove(e).0; removed.push(t); left -= 1; } else { - assert!(removed.contains(&(i, 2 * i)), "{} not in {:?}", i, removed); + assert!(removed.contains(&(i, 2 * i)), "{i} not in {removed:?}"); let e = map.table.insert( hash_value, (i, 2 * i), @@ -8509,4 +8458,49 @@ mod test_map { 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, + } + 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(); + } } diff --git a/vendor/hashbrown/src/raw/alloc.rs b/vendor/hashbrown/src/raw/alloc.rs index ba09ea9de..15299e7b0 100644 --- a/vendor/hashbrown/src/raw/alloc.rs +++ b/vendor/hashbrown/src/raw/alloc.rs @@ -1,5 +1,9 @@ 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; @@ -7,28 +11,44 @@ mod inner { use core::ptr::NonNull; #[allow(clippy::map_err_ignore)] - pub fn do_alloc(alloc: &A, layout: Layout) -> Result, ()> { + pub(crate) fn do_alloc(alloc: &A, layout: Layout) -> Result, ()> { match alloc.allocate(layout) { Ok(ptr) => Ok(ptr.as_non_null_ptr()), Err(_) => Err(()), } } +} - #[cfg(feature = "bumpalo")] - unsafe impl Allocator for crate::BumpWrapper<'_> { - #[inline] - fn allocate(&self, layout: Layout) -> Result, core::alloc::AllocError> { - match self.0.try_alloc_layout(layout) { - Ok(ptr) => Ok(NonNull::slice_from_raw_parts(ptr, layout.size())), - Err(_) => Err(core::alloc::AllocError), - } +// 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(alloc: &A, layout: Layout) -> Result, ()> { + match alloc.allocate(layout) { + Ok(ptr) => Ok(ptr.cast()), + Err(_) => Err(()), } - #[inline] - unsafe fn deallocate(&self, _ptr: NonNull, _layout: Layout) {} } } -#[cfg(not(feature = "nightly"))] +// 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; @@ -41,6 +61,7 @@ mod inner { #[derive(Copy, Clone)] pub struct Global; + unsafe impl Allocator for Global { #[inline] fn allocate(&self, layout: Layout) -> Result, ()> { @@ -51,6 +72,7 @@ mod inner { dealloc(ptr.as_ptr(), layout); } } + impl Default for Global { #[inline] fn default() -> Self { @@ -58,16 +80,7 @@ mod inner { } } - pub fn do_alloc(alloc: &A, layout: Layout) -> Result, ()> { + pub(crate) fn do_alloc(alloc: &A, layout: Layout) -> Result, ()> { alloc.allocate(layout) } - - #[cfg(feature = "bumpalo")] - unsafe impl Allocator for crate::BumpWrapper<'_> { - #[allow(clippy::map_err_ignore)] - fn allocate(&self, layout: Layout) -> Result, ()> { - self.0.try_alloc_layout(layout).map_err(|_| ()) - } - unsafe fn deallocate(&self, _ptr: NonNull, _layout: Layout) {} - } } diff --git a/vendor/hashbrown/src/raw/bitmask.rs b/vendor/hashbrown/src/raw/bitmask.rs index 7d4f9fc38..6576b3c5c 100644 --- a/vendor/hashbrown/src/raw/bitmask.rs +++ b/vendor/hashbrown/src/raw/bitmask.rs @@ -1,6 +1,6 @@ -use super::imp::{BitMaskWord, BITMASK_MASK, BITMASK_STRIDE}; -#[cfg(feature = "nightly")] -use core::intrinsics; +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. @@ -8,75 +8,55 @@ use core::intrinsics; /// 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, so -/// that there is only one bit-per-byte used (the high bit, 7). If this is the +/// 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 struct BitMask(pub BitMaskWord); +pub(crate) struct BitMask(pub(crate) BitMaskWord); #[allow(clippy::use_self)] impl BitMask { /// Returns a new `BitMask` with all bits inverted. #[inline] #[must_use] - pub fn invert(self) -> Self { + #[allow(dead_code)] + pub(crate) fn invert(self) -> Self { BitMask(self.0 ^ BITMASK_MASK) } - /// 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 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 ^= mask; - // The bit was set if the bit is now 0. - self.0 & mask == 0 - } - /// Returns a new `BitMask` with the lowest bit removed. #[inline] #[must_use] - pub fn remove_lowest_bit(self) -> Self { + fn remove_lowest_bit(self) -> Self { BitMask(self.0 & (self.0 - 1)) } + /// Returns whether the `BitMask` has at least one set bit. #[inline] - pub fn any_bit_set(self) -> bool { + 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 fn lowest_set_bit(self) -> Option { - if self.0 == 0 { - None + pub(crate) fn lowest_set_bit(self) -> Option { + if let Some(nonzero) = NonZeroBitMaskWord::new(self.0) { + Some(Self::nonzero_trailing_zeros(nonzero)) } else { - Some(unsafe { self.lowest_set_bit_nonzero() }) + None } } - /// Returns the first set bit in the `BitMask`, if there is one. The - /// bitmask must not be empty. - #[inline] - #[cfg(feature = "nightly")] - pub unsafe fn lowest_set_bit_nonzero(self) -> usize { - intrinsics::cttz_nonzero(self.0) as usize / BITMASK_STRIDE - } - #[inline] - #[cfg(not(feature = "nightly"))] - pub unsafe fn lowest_set_bit_nonzero(self) -> usize { - self.trailing_zeros() - } - /// Returns the number of trailing zeroes in the `BitMask`. #[inline] - pub fn trailing_zeros(self) -> usize { + 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 @@ -89,9 +69,21 @@ impl BitMask { } } + /// 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 fn leading_zeros(self) -> usize { + pub(crate) fn leading_zeros(self) -> usize { self.0.leading_zeros() as usize / BITMASK_STRIDE } } @@ -102,13 +94,32 @@ impl IntoIterator for BitMask { #[inline] fn into_iter(self) -> BitMaskIter { - BitMaskIter(self) + // 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. -pub struct BitMaskIter(BitMask); +#[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; diff --git a/vendor/hashbrown/src/raw/generic.rs b/vendor/hashbrown/src/raw/generic.rs index 52955a45b..c668b0642 100644 --- a/vendor/hashbrown/src/raw/generic.rs +++ b/vendor/hashbrown/src/raw/generic.rs @@ -5,26 +5,29 @@ 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(any( - target_pointer_width = "64", - target_arch = "aarch64", - target_arch = "x86_64", - target_arch = "wasm32", -))] -type GroupWord = u64; -#[cfg(all( - any(target_pointer_width = "32", target_pointer_width = "16"), - not(target_arch = "aarch64"), - not(target_arch = "x86_64"), - not(target_arch = "wasm32"), -))] -type GroupWord = u32; -pub type BitMaskWord = GroupWord; -pub const BITMASK_STRIDE: usize = 8; +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 const BITMASK_MASK: BitMaskWord = 0x8080_8080_8080_8080_u64 as GroupWord; +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] @@ -37,7 +40,7 @@ fn repeat(byte: u8) -> GroupWord { /// /// This implementation uses a word-sized integer. #[derive(Copy, Clone)] -pub struct Group(GroupWord); +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 @@ -46,14 +49,14 @@ pub struct Group(GroupWord); #[allow(clippy::use_self)] impl Group { /// Number of bytes in the group. - pub const WIDTH: usize = mem::size_of::(); + pub(crate) const WIDTH: usize = mem::size_of::(); /// 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 const fn static_empty() -> &'static [u8; Group::WIDTH] { + pub(crate) const fn static_empty() -> &'static [u8; Group::WIDTH] { #[repr(C)] struct AlignedBytes { _align: [Group; 0], @@ -69,7 +72,7 @@ impl Group { /// Loads a group of bytes starting at the given address. #[inline] #[allow(clippy::cast_ptr_alignment)] // unaligned load - pub unsafe fn load(ptr: *const u8) -> Self { + pub(crate) unsafe fn load(ptr: *const u8) -> Self { Group(ptr::read_unaligned(ptr.cast())) } @@ -77,7 +80,7 @@ impl Group { /// aligned to `mem::align_of::()`. #[inline] #[allow(clippy::cast_ptr_alignment)] - pub unsafe fn load_aligned(ptr: *const u8) -> Self { + 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::() - 1), 0); Group(ptr::read(ptr.cast())) @@ -87,7 +90,7 @@ impl Group { /// aligned to `mem::align_of::()`. #[inline] #[allow(clippy::cast_ptr_alignment)] - pub unsafe fn store_aligned(self, ptr: *mut u8) { + 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::() - 1), 0); ptr::write(ptr.cast(), self.0); @@ -104,7 +107,7 @@ impl Group { /// - 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 fn match_byte(self, byte: u8) -> BitMask { + 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); @@ -114,7 +117,7 @@ impl Group { /// Returns a `BitMask` indicating all bytes in the group which are /// `EMPTY`. #[inline] - pub fn match_empty(self) -> BitMask { + 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. @@ -124,14 +127,14 @@ impl Group { /// Returns a `BitMask` indicating all bytes in the group which are /// `EMPTY` or `DELETED`. #[inline] - pub fn match_empty_or_deleted(self) -> BitMask { + 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 fn match_full(self) -> BitMask { + pub(crate) fn match_full(self) -> BitMask { self.match_empty_or_deleted().invert() } @@ -140,7 +143,7 @@ impl Group { /// - `DELETED => EMPTY` /// - `FULL => DELETED` #[inline] - pub fn convert_special_to_empty_and_full_to_deleted(self) -> Self { + 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 // diff --git a/vendor/hashbrown/src/raw/mod.rs b/vendor/hashbrown/src/raw/mod.rs index 0e96306ef..1a6dced4b 100644 --- a/vendor/hashbrown/src/raw/mod.rs +++ b/vendor/hashbrown/src/raw/mod.rs @@ -25,8 +25,10 @@ cfg_if! { ))] { mod sse2; use sse2 as imp; + } else if #[cfg(all(target_arch = "aarch64", target_feature = "neon"))] { + mod neon; + use neon as imp; } else { - #[path = "generic.rs"] mod generic; use generic as imp; } @@ -37,36 +39,26 @@ pub(crate) use self::alloc::{do_alloc, Allocator, Global}; mod bitmask; -use self::bitmask::{BitMask, BitMaskIter}; +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}; -// On stable we can use #[cold] to get a equivalent effect: this attributes -// suggests that the function is unlikely to be called -#[cfg(not(feature = "nightly"))] -#[inline] -#[cold] -fn cold() {} - -#[cfg(not(feature = "nightly"))] -#[inline] -fn likely(b: bool) -> bool { - if !b { - cold(); - } - b -} +// 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] -fn unlikely(b: bool) -> bool { - if b { - cold(); - } - b +#[inline(always)] +fn invalid_mut(addr: usize) -> *mut T { + addr as *mut T } #[inline] @@ -272,6 +264,11 @@ impl TableLayout { } } +/// 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 @@ -299,11 +296,79 @@ impl Clone for Bucket { impl Bucket { const IS_ZERO_SIZED_TYPE: bool = mem::size_of::() == 0; + /// 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::()` 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::() != 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`]; + /// + /// * `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::() == 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`]: RawTableInner::data_end + /// [`RawTable::buckets`]: crate::raw::RawTable::buckets + /// [`RawTableInner::buckets`]: RawTableInner::buckets #[inline] unsafe fn from_base_index(base: NonNull, index: usize) -> Self { + // If mem::size_of::() != 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` 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 Self::IS_ZERO_SIZED_TYPE { - // won't overflow because index must be less than length - (index + 1) as *mut T + // 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) }; @@ -311,27 +376,183 @@ impl Bucket { 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::()`]. + /// + /// 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::() != 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`]; + /// + /// * `self` also must not contain dangling pointer; + /// + /// * both `self` and `base` must be created from the same [`RawTable`] + /// (or [`RawTableInner`]). + /// + /// If `mem::size_of::() == 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`]: RawTableInner::data_end + /// [`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) -> usize { + // If mem::size_of::() != 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::() + // | + // | `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` 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::() + // + // where: T0...Tlast - our stored data; C0...Clast - control bytes or metadata for data. if Self::IS_ZERO_SIZED_TYPE { + // 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; + /// + /// fn make_hash(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 Self::IS_ZERO_SIZED_TYPE { // Just return an arbitrary ZST pointer which is properly aligned - mem::align_of::() as *mut T + // invalid pointer is good enough for ZST + invalid_mut(mem::align_of::()) } 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::()` bytes. + /// + /// # Safety + /// + /// If `mem::size_of::() != 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::() == 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 Self::IS_ZERO_SIZED_TYPE { - (self.ptr.as_ptr() as usize + offset) as *mut T + // invalid pointer is good enough for ZST + invalid_mut(self.ptr.as_ptr() as usize + offset) } else { self.ptr.as_ptr().sub(offset) }; @@ -339,26 +560,212 @@ impl Bucket { 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; + /// + /// fn make_hash(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; + /// + /// fn make_hash(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` 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) { @@ -516,9 +923,9 @@ impl RawTable { /// Returns pointer to start of data table. #[inline] - #[cfg(feature = "nightly")] - pub unsafe fn data_start(&self) -> *mut T { - self.data_end().as_ptr().wrapping_sub(self.buckets()) + #[cfg(any(feature = "raw", feature = "nightly"))] + pub unsafe fn data_start(&self) -> NonNull { + NonNull::new_unchecked(self.data_end().as_ptr().wrapping_sub(self.buckets())) } /// Return the information about memory allocated by the table. @@ -581,11 +988,18 @@ impl RawTable { } /// 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 { + pub unsafe fn remove(&mut self, item: Bucket) -> (T, InsertSlot) { self.erase_no_drop(&item); - item.read() + ( + item.read(), + InsertSlot { + index: self.bucket_index(&item), + }, + ) } /// Finds and removes an element from the table, returning it. @@ -593,7 +1007,7 @@ impl RawTable { pub fn remove_entry(&mut self, hash: u64, eq: impl FnMut(&T) -> bool) -> Option { // Avoid `Option::map` because it bloats LLVM IR. match self.find(hash, eq) { - Some(bucket) => Some(unsafe { self.remove(bucket) }), + Some(bucket) => Some(unsafe { self.remove(bucket).0 }), None => None, } } @@ -607,6 +1021,10 @@ impl RawTable { /// 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 { @@ -663,7 +1081,7 @@ impl RawTable { /// without reallocation. #[cfg_attr(feature = "inline-more", inline)] pub fn reserve(&mut self, additional: usize, hasher: impl Fn(&T) -> u64) { - if additional > self.table.growth_left { + if unlikely(additional > self.table.growth_left) { // Avoid `Result::unwrap_or_else` because it bloats LLVM IR. if self .reserve_rehash(additional, hasher, Fallibility::Infallible) @@ -737,22 +1155,18 @@ impl RawTable { #[cfg_attr(feature = "inline-more", inline)] pub fn insert(&mut self, hash: u64, value: T, hasher: impl Fn(&T) -> u64) -> Bucket { unsafe { - let mut index = self.table.find_insert_slot(hash); + 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. - let old_ctrl = *self.table.ctrl(index); + let old_ctrl = *self.table.ctrl(slot.index); if unlikely(self.table.growth_left == 0 && special_is_empty(old_ctrl)) { self.reserve(1, hasher); - index = self.table.find_insert_slot(hash); + slot = self.table.find_insert_slot(hash); } - self.table.record_item_insert_at(index, old_ctrl, hash); - - let bucket = self.bucket(index); - bucket.write(value); - bucket + self.insert_in_slot(hash, slot, value) } } @@ -820,7 +1234,7 @@ impl RawTable { 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); + 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); @@ -832,6 +1246,49 @@ impl RawTable { } } + /// 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, InsertSlot> { + self.reserve(1, hasher); + + match self + .table + .find_or_find_insert_slot_inner(hash, &mut |index| unsafe { + eq(self.bucket(index).as_ref()) + }) { + Ok(index) => Ok(unsafe { 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 { + 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> { @@ -984,7 +1441,7 @@ impl RawTable { /// `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, A> { + pub unsafe fn iter_hash(&self, hash: u64) -> RawIterHash { RawIterHash::new(self, hash) } @@ -1072,6 +1529,11 @@ where } impl RawTableInner { + /// 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_in(alloc: A) -> Self { Self { @@ -1086,6 +1548,18 @@ impl RawTableInner { } 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( alloc: A, @@ -1106,6 +1580,7 @@ impl RawTableInner { 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, @@ -1116,6 +1591,10 @@ impl RawTableInner { }) } + /// 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( alloc: A, @@ -1126,11 +1605,16 @@ impl RawTableInner { if capacity == 0 { Ok(Self::new_in(alloc)) } 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) @@ -1138,66 +1622,200 @@ impl RawTableInner { } } + /// Fixes up an insertion slot due to false positives for groups smaller than the group width. + /// This must only be used on insertion slots found by `find_insert_slot_in_group`. + #[inline] + unsafe fn fix_insert_slot(&self, mut index: usize) -> InsertSlot { + // 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, but once masked 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). + if unlikely(self.is_bucket_full(index)) { + debug_assert!(self.bucket_mask < Group::WIDTH); + // SAFETY: + // + // * We are in range and `ptr = self.ctrl(0)` are valid for reads + // and properly aligned, because the table is already allocated + // (see `TableLayout::calculate_layout_for` and `ptr::read`); + // + // * 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, 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. + #[inline] + fn find_insert_slot_in_group(&self, group: &Group, probe_seq: &ProbeSeq) -> Option { + let bit = group.match_empty_or_deleted().lowest_set_bit(); + + if likely(bit.is_some()) { + 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. + /// + /// This uses dynamic dispatch to reduce the amount of code generated, but that is + /// eliminated by LLVM optimizations. + #[inline] + fn find_or_find_insert_slot_inner( + &self, + hash: u64, + eq: &mut dyn FnMut(usize) -> bool, + ) -> Result { + let mut insert_slot = None; + + let h2_hash = h2(hash); + let mut probe_seq = self.probe_seq(hash); + + loop { + 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 { + 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. /// /// There must be at least 1 empty bucket in the table. #[inline] unsafe fn prepare_insert_slot(&self, hash: u64) -> (usize, u8) { - let index = self.find_insert_slot(hash); + let index = self.find_insert_slot(hash).index; 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. + /// a new element, returning the `index` for the new [`Bucket`]. /// - /// There must be at least 1 empty bucket in the table. + /// 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. + /// + /// # Note + /// + /// Calling this function is always safe, but attempting 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 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] - fn find_insert_slot(&self, hash: u64) -> usize { + fn find_insert_slot(&self, hash: u64) -> InsertSlot { let mut probe_seq = self.probe_seq(hash); loop { + // SAFETY: + // * `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 comment for masking below). + // + // * 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). unsafe { let group = Group::load(self.ctrl(probe_seq.pos)); - if let Some(bit) = group.match_empty_or_deleted().lowest_set_bit() { - let result = (probe_seq.pos + bit) & self.bucket_mask; - - // In tables smaller than the group width, trailing control - // bytes outside the range of the table are filled with - // EMPTY entries. These will unfortunately trigger a - // match, but once masked 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). - if unlikely(self.is_bucket_full(result)) { - debug_assert!(self.bucket_mask < Group::WIDTH); - debug_assert_ne!(probe_seq.pos, 0); - return Group::load_aligned(self.ctrl(0)) - .match_empty_or_deleted() - .lowest_set_bit_nonzero(); - } + let index = self.find_insert_slot_in_group(&group, &probe_seq); - return result; + if likely(index.is_some()) { + return self.fix_insert_slot(index.unwrap_unchecked()); } } probe_seq.move_next(self.bucket_mask); } } - /// Searches for an element in the table. This uses dynamic dispatch to reduce the amount of - /// code generated, but it is eliminated by LLVM optimizations. + /// 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). #[inline(always)] fn find_inner(&self, hash: u64, eq: &mut dyn FnMut(usize) -> bool) -> Option { let h2_hash = h2(hash); let mut probe_seq = self.probe_seq(hash); loop { + // SAFETY: + // * `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)) { @@ -1213,12 +1831,49 @@ impl RawTableInner { } } + /// 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. + /// + /// 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. + // 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(); @@ -1227,10 +1882,19 @@ impl RawTableInner { // Fix up the trailing control bytes. See the comments in set_ctrl // for the handling of tables smaller than the group width. - if self.buckets() < 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); } @@ -1274,7 +1938,7 @@ impl RawTableInner { #[cfg(feature = "raw")] #[inline] unsafe fn prepare_insert_no_grow(&mut self, hash: u64) -> Result { - let index = self.find_insert_slot(hash); + 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(()) @@ -1301,13 +1965,68 @@ impl RawTableInner { /// 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(&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(&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 @@ -1315,6 +2034,28 @@ impl RawTableInner { /// 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(&self, index: usize, ctrl: u8) { // Replicate the first Group::WIDTH control bytes at the end of @@ -1335,16 +2076,43 @@ impl RawTableInner { // --------------------------------------------- // | [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) } @@ -1541,7 +2309,7 @@ impl RawTableInner { let hash = hasher(*guard, i); // Search for a suitable place to put it - let new_i = guard.find_insert_slot(hash); + 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 @@ -1626,27 +2394,95 @@ impl RawTableInner { 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(); - // If we are inside a continuous block of Group::WIDTH full or deleted - // cells then a probe window may have seen a full block when trying to - // insert. We therefore need to keep that block non-empty so that - // lookups will continue searching to the next probe window. + // 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: // - // 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. + // - 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; } @@ -1752,7 +2588,8 @@ impl RawTableClone for RawTable { .copy_to_nonoverlapping(self.table.ctrl(0), self.table.num_ctrl_bytes()); source .data_start() - .copy_to_nonoverlapping(self.data_start(), self.table.buckets()); + .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; @@ -1776,7 +2613,7 @@ impl RawTable { // to make sure we drop only the elements that have been // cloned so far. let mut guard = guard((0, &mut *self), |(index, self_)| { - if Self::DATA_NEEDS_DROP && !self_.is_empty() { + if Self::DATA_NEEDS_DROP { for i in 0..=*index { if self_.is_bucket_full(i) { self_.bucket(i).drop(); @@ -1896,7 +2733,7 @@ impl IntoIterator for RawTable { pub(crate) struct RawIterRange { // Mask of full buckets in the current group. Bits are cleared from this // mask as each element is processed. - current_group: BitMask, + current_group: BitMaskIter, // Pointer to the buckets for the current group. data: Bucket, @@ -1924,7 +2761,7 @@ impl RawIterRange { let next_ctrl = ctrl.add(Group::WIDTH); Self { - current_group, + current_group: current_group.into_iter(), data, next_ctrl, end, @@ -1981,8 +2818,7 @@ impl RawIterRange { #[cfg_attr(feature = "inline-more", inline)] unsafe fn next_impl(&mut self) -> Option> { loop { - if let Some(index) = self.current_group.lowest_set_bit() { - self.current_group = self.current_group.remove_lowest_bit(); + if let Some(index) = self.current_group.next() { return Some(self.data.next_n(index)); } @@ -1995,7 +2831,7 @@ impl RawIterRange { // 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(); + 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); } @@ -2074,7 +2910,7 @@ impl RawIter { /// 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 fn reflect_remove(&mut self, b: &Bucket) { + pub unsafe fn reflect_remove(&mut self, b: &Bucket) { self.reflect_toggle_full(b, false); } @@ -2088,36 +2924,76 @@ impl RawIter { /// /// This method should be called _after_ the given insert is made. #[cfg(feature = "raw")] - pub fn reflect_insert(&mut self, b: &Bucket) { + pub unsafe fn reflect_insert(&mut self, b: &Bucket) { 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")] - fn reflect_toggle_full(&mut self, b: &Bucket, is_insert: bool) { - unsafe { - 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; + unsafe fn reflect_toggle_full(&mut self, b: &Bucket, 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 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 is_insert { + self.items += 1; + } else { + self.items -= 1; + } - 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)); - } + 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; @@ -2125,60 +3001,18 @@ impl RawIter { 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.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; + 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 { - 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.lowest_set_bit(), Some(index)); - } else { - // We should not have changed what bucket comes next. - debug_assert_eq!(self.iter.current_group.lowest_set_bit(), Some(index)); - } + // 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. } + } else { + // We must have already iterated past the removed item. } } @@ -2217,9 +3051,8 @@ impl Iterator for RawIter { self.iter.next_impl::() }; - if nxt.is_some() { - self.items -= 1; - } + debug_assert!(nxt.is_some()); + self.items -= 1; nxt } @@ -2388,13 +3221,28 @@ impl FusedIterator for RawDrain<'_, T, A> {} /// `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. -pub struct RawIterHash<'a, T, A: Allocator + Clone = Global> { - inner: RawIterHashInner<'a, A>, +/// +/// 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 { + inner: RawIterHashInner, _marker: PhantomData, } -struct RawIterHashInner<'a, A: Allocator + Clone> { - table: &'a RawTableInner, +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, // The top 7 bits of the hash. h2_hash: u8, @@ -2408,65 +3256,77 @@ struct RawIterHashInner<'a, A: Allocator + Clone> { bitmask: BitMaskIter, } -impl<'a, T, A: Allocator + Clone> RawIterHash<'a, T, A> { +impl RawIterHash { #[cfg_attr(feature = "inline-more", inline)] #[cfg(feature = "raw")] - fn new(table: &'a RawTable, hash: u64) -> Self { + unsafe fn new(table: &RawTable, hash: u64) -> Self { RawIterHash { inner: RawIterHashInner::new(&table.table, hash), _marker: PhantomData, } } } -impl<'a, A: Allocator + Clone> RawIterHashInner<'a, A> { +impl RawIterHashInner { #[cfg_attr(feature = "inline-more", inline)] #[cfg(feature = "raw")] - fn new(table: &'a RawTableInner, hash: u64) -> Self { - unsafe { - 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 { - table, - h2_hash, - probe_seq, - group, - bitmask, - } + 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<'a, T, A: Allocator + Clone> Iterator for RawIterHash<'a, T, A> { +impl Iterator for RawIterHash { type Item = Bucket; fn next(&mut self) -> Option> { unsafe { match self.inner.next() { - Some(index) => Some(self.inner.table.bucket(index)), + 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<'a, A: Allocator + Clone> Iterator for RawIterHashInner<'a, A> { +impl Iterator for RawIterHashInner { type Item = usize; fn next(&mut self) -> Option { unsafe { loop { if let Some(bit) = self.bitmask.next() { - let index = (self.probe_seq.pos + bit) & self.table.bucket_mask; + 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.table.bucket_mask); - self.group = Group::load(self.table.ctrl(self.probe_seq.pos)); + 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(); } } diff --git a/vendor/hashbrown/src/raw/neon.rs b/vendor/hashbrown/src/raw/neon.rs new file mode 100644 index 000000000..44e82d57d --- /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::(); + + /// 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::()`. + #[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::() - 1), 0); + Group(neon::vld1_u8(ptr)) + } + + /// Stores the group of bytes to the given address, which must be + /// aligned to `mem::align_of::()`. + #[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::() - 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 index a0bf6da80..956ba5d26 100644 --- a/vendor/hashbrown/src/raw/sse2.rs +++ b/vendor/hashbrown/src/raw/sse2.rs @@ -1,28 +1,31 @@ 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 type BitMaskWord = u16; -pub const BITMASK_STRIDE: usize = 1; -pub const BITMASK_MASK: BitMaskWord = 0xffff; +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 struct Group(x86::__m128i); +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 const WIDTH: usize = mem::size_of::(); + pub(crate) const WIDTH: usize = mem::size_of::(); /// Returns a full group of empty bytes, suitable for use as the initial /// value for an empty hash table. @@ -30,7 +33,7 @@ impl Group { /// This is guaranteed to be aligned to the group size. #[inline] #[allow(clippy::items_after_statements)] - pub const fn static_empty() -> &'static [u8; Group::WIDTH] { + pub(crate) const fn static_empty() -> &'static [u8; Group::WIDTH] { #[repr(C)] struct AlignedBytes { _align: [Group; 0], @@ -46,7 +49,7 @@ impl Group { /// Loads a group of bytes starting at the given address. #[inline] #[allow(clippy::cast_ptr_alignment)] // unaligned load - pub unsafe fn load(ptr: *const u8) -> Self { + pub(crate) unsafe fn load(ptr: *const u8) -> Self { Group(x86::_mm_loadu_si128(ptr.cast())) } @@ -54,7 +57,7 @@ impl Group { /// aligned to `mem::align_of::()`. #[inline] #[allow(clippy::cast_ptr_alignment)] - pub unsafe fn load_aligned(ptr: *const u8) -> Self { + 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::() - 1), 0); Group(x86::_mm_load_si128(ptr.cast())) @@ -64,7 +67,7 @@ impl Group { /// aligned to `mem::align_of::()`. #[inline] #[allow(clippy::cast_ptr_alignment)] - pub unsafe fn store_aligned(self, ptr: *mut u8) { + 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::() - 1), 0); x86::_mm_store_si128(ptr.cast(), self.0); @@ -73,7 +76,7 @@ impl Group { /// Returns a `BitMask` indicating all bytes in the group which have /// the given value. #[inline] - pub fn match_byte(self, byte: u8) -> BitMask { + pub(crate) fn match_byte(self, byte: u8) -> BitMask { #[allow( clippy::cast_possible_wrap, // byte: u8 as i8 // byte: i32 as u16 @@ -91,14 +94,14 @@ impl Group { /// Returns a `BitMask` indicating all bytes in the group which are /// `EMPTY`. #[inline] - pub fn match_empty(self) -> BitMask { + 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 fn match_empty_or_deleted(self) -> BitMask { + 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 @@ -114,7 +117,7 @@ impl Group { /// Returns a `BitMask` indicating all bytes in the group which are full. #[inline] - pub fn match_full(&self) -> BitMask { + pub(crate) fn match_full(&self) -> BitMask { self.match_empty_or_deleted().invert() } @@ -123,7 +126,7 @@ impl Group { /// - `DELETED => EMPTY` /// - `FULL => DELETED` #[inline] - pub fn convert_special_to_empty_and_full_to_deleted(self) -> Self { + 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 // diff --git a/vendor/hashbrown/src/rustc_entry.rs b/vendor/hashbrown/src/rustc_entry.rs index 2e8459526..89447d27d 100644 --- a/vendor/hashbrown/src/rustc_entry.rs +++ b/vendor/hashbrown/src/rustc_entry.rs @@ -1,5 +1,5 @@ use self::RustcEntry::*; -use crate::map::{make_insert_hash, Drain, HashMap, IntoIter, Iter, IterMut}; +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}; @@ -32,7 +32,7 @@ where /// ``` #[cfg_attr(feature = "inline-more", inline)] pub fn rustc_entry(&mut self, key: K) -> RustcEntry<'_, K, V, A> { - let hash = make_insert_hash(&self.hash_builder, &key); + 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), @@ -330,7 +330,7 @@ impl<'a, K, V, A: Allocator + Clone> RustcOccupiedEntry<'a, K, V, A> { /// ``` #[cfg_attr(feature = "inline-more", inline)] pub fn remove_entry(self) -> (K, V) { - unsafe { self.table.remove(self.elem) } + unsafe { self.table.remove(self.elem).0 } } /// Gets a reference to the value in the entry. diff --git a/vendor/hashbrown/src/set.rs b/vendor/hashbrown/src/set.rs index a8f24de80..52f6fdaf2 100644 --- a/vendor/hashbrown/src/set.rs +++ b/vendor/hashbrown/src/set.rs @@ -5,10 +5,9 @@ use alloc::borrow::ToOwned; use core::fmt; use core::hash::{BuildHasher, Hash}; use core::iter::{Chain, FromIterator, FusedIterator}; -use core::mem; use core::ops::{BitAnd, BitOr, BitXor, Sub}; -use super::map::{self, ConsumeAllOnDrop, DefaultHashBuilder, DrainFilterInner, HashMap, Keys}; +use super::map::{self, DefaultHashBuilder, ExtractIfInner, HashMap, Keys}; use crate::raw::{Allocator, Global}; // Future Optimization (FIXME!) @@ -380,8 +379,9 @@ impl HashSet { /// In other words, move all elements `e` such that `f(&e)` returns `true` out /// into another iterator. /// - /// When the returned DrainedFilter is dropped, any remaining elements that satisfy - /// the predicate are dropped from the set. + /// 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 /// @@ -389,7 +389,7 @@ impl HashSet { /// use hashbrown::HashSet; /// /// let mut set: HashSet = (0..8).collect(); - /// let drained: HashSet = set.drain_filter(|v| v % 2 == 0).collect(); + /// let drained: HashSet = set.extract_if(|v| v % 2 == 0).collect(); /// /// let mut evens = drained.into_iter().collect::>(); /// let mut odds = set.into_iter().collect::>(); @@ -400,13 +400,13 @@ impl HashSet { /// assert_eq!(odds, vec![1, 3, 5, 7]); /// ``` #[cfg_attr(feature = "inline-more", inline)] - pub fn drain_filter(&mut self, f: F) -> DrainFilter<'_, T, F, A> + pub fn extract_if(&mut self, f: F) -> ExtractIf<'_, T, F, A> where F: FnMut(&T) -> bool, { - DrainFilter { + ExtractIf { f, - inner: DrainFilterInner { + inner: ExtractIfInner { iter: unsafe { self.map.table.iter() }, table: &mut self.map.table, }, @@ -1221,8 +1221,10 @@ where None => None, } } +} - /// Returns a mutable reference to the [`RawTable`] used underneath [`HashSet`]. +impl HashSet { + /// Returns a reference to the [`RawTable`] used underneath [`HashSet`]. /// This function is only available if the `raw` feature of the crate is enabled. /// /// # Note @@ -1238,9 +1240,29 @@ where /// [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 self) -> &mut RawTable<(T, ()), A> { + 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 PartialEq for HashSet @@ -1547,17 +1569,18 @@ pub struct Drain<'a, K, A: Allocator + Clone = Global> { /// A draining iterator over entries of a `HashSet` which don't satisfy the predicate `f`. /// -/// This `struct` is created by the [`drain_filter`] method on [`HashSet`]. See its +/// This `struct` is created by the [`extract_if`] method on [`HashSet`]. See its /// documentation for more. /// -/// [`drain_filter`]: struct.HashSet.html#method.drain_filter +/// [`extract_if`]: struct.HashSet.html#method.extract_if /// [`HashSet`]: struct.HashSet.html -pub struct DrainFilter<'a, K, F, A: Allocator + Clone = Global> +#[must_use = "Iterators are lazy unless consumed"] +pub struct ExtractIf<'a, K, F, A: Allocator + Clone = Global> where F: FnMut(&K) -> bool, { f: F, - inner: DrainFilterInner<'a, K, (), A>, + inner: ExtractIfInner<'a, K, (), A>, } /// A lazy iterator producing elements in the intersection of `HashSet`s. @@ -1748,21 +1771,7 @@ impl fmt::Debug for Drain<'_, K, A> { } } -impl<'a, K, F, A: Allocator + Clone> Drop for DrainFilter<'a, K, F, A> -where - F: FnMut(&K) -> bool, -{ - #[cfg_attr(feature = "inline-more", inline)] - fn drop(&mut self) { - while let Some(item) = self.next() { - let guard = ConsumeAllOnDrop(self); - drop(item); - mem::forget(guard); - } - } -} - -impl Iterator for DrainFilter<'_, K, F, A> +impl Iterator for ExtractIf<'_, K, F, A> where F: FnMut(&K) -> bool, { @@ -1781,10 +1790,7 @@ where } } -impl FusedIterator for DrainFilter<'_, K, F, A> where - F: FnMut(&K) -> bool -{ -} +impl FusedIterator for ExtractIf<'_, K, F, A> where F: FnMut(&K) -> bool {} impl Clone for Intersection<'_, T, S, A> { #[cfg_attr(feature = "inline-more", inline)] @@ -2712,10 +2718,10 @@ mod test_set { set.insert(1); set.insert(2); - let set_str = format!("{:?}", set); + let set_str = format!("{set:?}"); assert!(set_str == "{1, 2}" || set_str == "{2, 1}"); - assert_eq!(format!("{:?}", empty), "{}"); + assert_eq!(format!("{empty:?}"), "{}"); } #[test] @@ -2790,6 +2796,7 @@ mod test_set { } #[test] + #[allow(clippy::needless_borrow)] fn test_extend_ref() { let mut a = HashSet::new(); a.insert(1); @@ -2829,10 +2836,10 @@ mod test_set { } #[test] - fn test_drain_filter() { + fn test_extract_if() { { let mut set: HashSet = (0..8).collect(); - let drained = set.drain_filter(|&k| k % 2 == 0); + let drained = set.extract_if(|&k| k % 2 == 0); let mut out = drained.collect::>(); out.sort_unstable(); assert_eq!(vec![0, 2, 4, 6], out); @@ -2840,7 +2847,7 @@ mod test_set { } { let mut set: HashSet = (0..8).collect(); - drop(set.drain_filter(|&k| k % 2 == 0)); + set.extract_if(|&k| k % 2 == 0).for_each(drop); assert_eq!(set.len(), 4, "Removes non-matching items on drop"); } } @@ -2886,4 +2893,11 @@ mod test_set { 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(); + } } diff --git a/vendor/hashbrown/tests/rayon.rs b/vendor/hashbrown/tests/rayon.rs index 8c603c5c4..d55e5a980 100644 --- a/vendor/hashbrown/tests/rayon.rs +++ b/vendor/hashbrown/tests/rayon.rs @@ -356,7 +356,9 @@ fn set_seq_par_equivalence_into_iter_empty() { let vec_seq = SET_EMPTY.clone().into_iter().collect::>(); let vec_par = SET_EMPTY.clone().into_par_iter().collect::>(); - assert_eq3!(vec_seq, vec_par, []); + // Work around type inference failure introduced by rend dev-dependency. + let empty: [char; 0] = []; + assert_eq3!(vec_seq, vec_par, empty); } #[test] diff --git a/vendor/hashbrown/tests/set.rs b/vendor/hashbrown/tests/set.rs index 5ae1ec98e..86ec96476 100644 --- a/vendor/hashbrown/tests/set.rs +++ b/vendor/hashbrown/tests/set.rs @@ -27,7 +27,7 @@ fn test_hashset_insert_remove() { assert_eq!(m.insert(x.clone()), true); } for (i, x) in tx.iter().enumerate() { - println!("removing {} {:?}", i, x); + println!("removing {i} {x:?}"); assert_eq!(m.remove(x), true); } } -- cgit v1.2.3