summaryrefslogtreecommitdiffstats
path: root/third_party/rust/syn/src/tt.rs
diff options
context:
space:
mode:
Diffstat (limited to 'third_party/rust/syn/src/tt.rs')
-rw-r--r--third_party/rust/syn/src/tt.rs107
1 files changed, 107 insertions, 0 deletions
diff --git a/third_party/rust/syn/src/tt.rs b/third_party/rust/syn/src/tt.rs
new file mode 100644
index 0000000000..d87c0ed4dc
--- /dev/null
+++ b/third_party/rust/syn/src/tt.rs
@@ -0,0 +1,107 @@
+use proc_macro2::{Delimiter, TokenStream, TokenTree};
+use std::hash::{Hash, Hasher};
+
+pub struct TokenTreeHelper<'a>(pub &'a TokenTree);
+
+impl<'a> PartialEq for TokenTreeHelper<'a> {
+ fn eq(&self, other: &Self) -> bool {
+ use proc_macro2::Spacing;
+
+ match (self.0, other.0) {
+ (TokenTree::Group(g1), TokenTree::Group(g2)) => {
+ match (g1.delimiter(), g2.delimiter()) {
+ (Delimiter::Parenthesis, Delimiter::Parenthesis)
+ | (Delimiter::Brace, Delimiter::Brace)
+ | (Delimiter::Bracket, Delimiter::Bracket)
+ | (Delimiter::None, Delimiter::None) => {}
+ _ => return false,
+ }
+
+ let s1 = g1.stream().into_iter();
+ let mut s2 = g2.stream().into_iter();
+
+ for item1 in s1 {
+ let item2 = match s2.next() {
+ Some(item) => item,
+ None => return false,
+ };
+ if TokenTreeHelper(&item1) != TokenTreeHelper(&item2) {
+ return false;
+ }
+ }
+ s2.next().is_none()
+ }
+ (TokenTree::Punct(o1), TokenTree::Punct(o2)) => {
+ o1.as_char() == o2.as_char()
+ && match (o1.spacing(), o2.spacing()) {
+ (Spacing::Alone, Spacing::Alone) | (Spacing::Joint, Spacing::Joint) => true,
+ _ => false,
+ }
+ }
+ (TokenTree::Literal(l1), TokenTree::Literal(l2)) => l1.to_string() == l2.to_string(),
+ (TokenTree::Ident(s1), TokenTree::Ident(s2)) => s1 == s2,
+ _ => false,
+ }
+ }
+}
+
+impl<'a> Hash for TokenTreeHelper<'a> {
+ fn hash<H: Hasher>(&self, h: &mut H) {
+ use proc_macro2::Spacing;
+
+ match self.0 {
+ TokenTree::Group(g) => {
+ 0u8.hash(h);
+ match g.delimiter() {
+ Delimiter::Parenthesis => 0u8.hash(h),
+ Delimiter::Brace => 1u8.hash(h),
+ Delimiter::Bracket => 2u8.hash(h),
+ Delimiter::None => 3u8.hash(h),
+ }
+
+ for item in g.stream() {
+ TokenTreeHelper(&item).hash(h);
+ }
+ 0xffu8.hash(h); // terminator w/ a variant we don't normally hash
+ }
+ TokenTree::Punct(op) => {
+ 1u8.hash(h);
+ op.as_char().hash(h);
+ match op.spacing() {
+ Spacing::Alone => 0u8.hash(h),
+ Spacing::Joint => 1u8.hash(h),
+ }
+ }
+ TokenTree::Literal(lit) => (2u8, lit.to_string()).hash(h),
+ TokenTree::Ident(word) => (3u8, word).hash(h),
+ }
+ }
+}
+
+pub struct TokenStreamHelper<'a>(pub &'a TokenStream);
+
+impl<'a> PartialEq for TokenStreamHelper<'a> {
+ fn eq(&self, other: &Self) -> bool {
+ let left = self.0.clone().into_iter().collect::<Vec<_>>();
+ let right = other.0.clone().into_iter().collect::<Vec<_>>();
+ if left.len() != right.len() {
+ return false;
+ }
+ for (a, b) in left.into_iter().zip(right) {
+ if TokenTreeHelper(&a) != TokenTreeHelper(&b) {
+ return false;
+ }
+ }
+ true
+ }
+}
+
+impl<'a> Hash for TokenStreamHelper<'a> {
+ fn hash<H: Hasher>(&self, state: &mut H) {
+ let tts = self.0.clone().into_iter().collect::<Vec<_>>();
+ tts.len().hash(state);
+ for tt in tts {
+ TokenTreeHelper(&tt).hash(state);
+ }
+ }
+}