Summary

In the concurrent-renamer path, child rename maps are built with std::collections::HashMap and then merged into an FxHashMap.

This introduces avoidable hasher overhead and rehashing in a path explicitly chosen for large rename workloads.

Evidence

crates/swc_ecma_transforms_base/src/rename/analyzer/scope.rs:

• Concurrent branch allocates std map per child: lines 247-259.
• Results are flattened into to: FxHashMap<Id, V>: lines 263-264.
• This branch is selected when parallel is true (high-cost rename cases).

Why this is a perf issue

Using std hash maps here means:

• slower hashing during child map build (SipHash),
• and extra hashing cost again when reinserting into FxHashMap.

For large scopes with many child bindings this adds measurable overhead to the parallel fast path.

Suggested direction

• Use FxHashMap<Id, V> for new_map in the concurrent branch.
• Optionally reserve capacity from child scope queue size when available.

Notes

Static-analysis report only (no benchmark attached yet).