perf(formatter): Improve is_expression_parenthesized performance (#5825)

crates/ruff_python_formatter/src/expression/parentheses.rs

@@ -1,6 +1,6 @@
 use crate::context::NodeLevel;
 use crate::prelude::*;
-use crate::trivia::{first_non_trivia_token, first_non_trivia_token_rev, Token, TokenKind};
+use crate::trivia::{first_non_trivia_token, SimpleTokenizer, Token, TokenKind};
 use ruff_formatter::prelude::tag::Condition;
 use ruff_formatter::{format_args, write, Argument, Arguments};
 use ruff_python_ast::node::AnyNodeRef;
@@ -72,19 +72,27 @@ pub enum Parentheses {
 }
 
 pub(crate) fn is_expression_parenthesized(expr: AnyNodeRef, contents: &str) -> bool {
-    matches!(
+    // First test if there's a closing parentheses because it tends to be cheaper.
+    if matches!(
         first_non_trivia_token(expr.end(), contents),
         Some(Token {
             kind: TokenKind::RParen,
             ..
         })
-    ) && matches!(
-        first_non_trivia_token_rev(expr.start(), contents),
-        Some(Token {
-            kind: TokenKind::LParen,
-            ..
-        })
-    )
+    ) {
+        let mut tokenizer =
+            SimpleTokenizer::up_to_without_back_comment(expr.start(), contents).skip_trivia();
+
+        matches!(
+            tokenizer.next_back(),
+            Some(Token {
+                kind: TokenKind::LParen,
+                ..
+            })
+        )
+    } else {
+        false
+    }
 }
 
 /// Formats `content` enclosed by the `left` and `right` parentheses. The implementation also ensures

crates/ruff_python_formatter/src/trivia.rs

@@ -274,10 +274,20 @@ impl<'a> SimpleTokenizer<'a> {
         Self::new(source, range)
     }
 
+    /// Creates a tokenizer that lexes tokens from the start of `source` up to `offset`.
     pub(crate) fn up_to(offset: TextSize, source: &'a str) -> Self {
         Self::new(source, TextRange::up_to(offset))
     }
 
+    /// Creates a tokenizer that lexes tokens from the start of `source` up to `offset`, and informs
+    /// the lexer that the line at `offset` contains no comments. This can significantly speed up backwards lexing
+    /// because the lexer doesn't need to scan for comments.
+    pub(crate) fn up_to_without_back_comment(offset: TextSize, source: &'a str) -> Self {
+        let mut tokenizer = Self::up_to(offset, source);
+        tokenizer.back_line_has_no_comment = true;
+        tokenizer
+    }
+
     fn to_keyword_or_other(&self, range: TextRange) -> TokenKind {
         let source = &self.source[range];
         match source {