⚠️ No Changeset found

Latest commit: 9555c5b

Merging this PR will not cause a version bump for any packages. If these changes should not result in a new version, you're good to go. If these changes should result in a version bump, you need to add a changeset.

Click here to learn what changesets are, and how to add one.

Click here if you're a maintainer who wants to add a changeset to this PR

Binary Sizes

Commit: 98bf428

Merging this PR will not alter performance

✅ 218 untouched benchmarks

_{Comparing kdy1/fix-11650-jsx-rescan-alloc (9555c5b) with main (f62bfa9)}

PR Review: perf(swc_ecma_parser): reduce JSX identifier rescan allocations

Summary

Good performance optimization that eliminates intermediate String allocations in JSX identifier rescanning by:

1. Refactoring scan_identifier_parts → scan_identifier_parts_into to append into a caller-owned buffer
2. Adding an ASCII fast path that collects contiguous ASCII identifier bytes in one slice operation
3. Replacing format!/to_string() with direct input_slice_str lookups for the prefix

Bug Fix (nice catch)

The removal of self.bump(1); // bump 'u' before self.read_unicode_escape() at line 602 is a correctness fix, not just a refactor. read_unicode_escape() already bumps u internally (line 1464 of mod.rs), so the old scan_identifier_parts was double-bumping past u, which would have caused incorrect parsing of \u escapes in identifier parts. The new test case <if-\u{2F804}> exercises this path and correctly expects an error. Good to have this covered.

Code Quality

• The unsafe blocks for input_slice_str have appropriate safety comments and the invariants are sound — start and prefix_end/self.cur_pos() are all obtained from self.input.
• The ASCII fast path inner loop is clean — check byte, bump if ASCII ident part, then slice the whole chunk at once.

Minor Observations

1. Duplicated branches in scan_jsx_identifier (lines 281-299): The is_known_ident() || is_keyword() branch and the final else branch have identical bodies — both do input_slice_str(start, prefix_end). Consider extracting the common path:

   let v = if !v.is_empty() {
       let mut value = String::new();
       if let Some(TokenValue::Word(prefix)) = self.state.token_value.take() {
           value.reserve(prefix.len() + v.len());
           value.push_str(&prefix);
       } else {
           let prefix = unsafe { self.input_slice_str(start, prefix_end) };
           value.reserve(prefix.len() + v.len());
           value.push_str(prefix);
       }
       value.push_str(&v);
       self.atom(value)
   } ...

   This would eliminate the duplicated unsafe block. Though this is minor and the current form mirrors the existing pattern, so up to you.

2. String::new() followed by reserve() (line 280): String::with_capacity(prefix.len() + v.len()) would be slightly cleaner, but this would require computing the prefix first. Current approach is fine.

Performance

The optimization is well-targeted: hot-path JSX identifier scanning no longer allocates intermediate Strings or uses format!. The ASCII fast path avoids per-byte push calls in favor of a single push_str of a slice. This should be a measurable improvement for JSX-heavy codebases.

Test Coverage

• New fixture test for the unicode escape bug fix (issue-11650).
• The error output .swc-stderr file is included.
• Existing parser tests provide regression coverage.

Verdict

LGTM. Clean performance optimization with a nice correctness fix for the \u double-bump bug. The duplicated branches are a minor nit.