Summary

Add caching layers in the sourcemap resolution and error stack parsing paths
to eliminate redundant work when the same file is referenced by multiple
stack frames or errors within a single test run.

Not on the hot path today, but they are zero-cost correctness improvements:

• Function-scoped Map, created and discarded each call — no stale state, no watch-mode risk
• Eliminates redundant new TraceMap() (VLQ decode) and JSON.parse() when multiple stack frames reference the same file
• The effect is visible when a run produces many failures pointing to the same source files (e.g., a broken import causing 100+ test failures)
• Even when the savings are small, caching a pure lookup is the idiomatic pattern — it makes the intent clear and prevents future regressions if this path becomes hotter

Changes:

1. Cache TraceMap per file in both parseErrorStacktrace implementations
   (packages/core/src/utils/error.ts and packages/core/src/reporter/md.ts).
   Avoids redundant VLQ decoding when multiple stack frames originate from the
   same source file.

2. Memoize JSON.parse in getSourcemap (packages/core/src/core/runTests.ts).
   A per-run parsedSourceMapCache prevents repeated parsing of the same
   sourcemap JSON when the same file appears in multiple errors.

3. Use Set.has instead of Array.includes for coverage file lookup
   (packages/core/src/coverage/generate.ts). Converts O(n) linear scans to
   O(1) hash lookups when filtering covered/uncovered files.

4. Fix fullStack condition in md.ts reporter (packages/core/src/reporter/md.ts).
   Only re-parse the full stack when error.fullStack is truthy, avoiding
   unnecessary duplicate parsing when the full stack is not requested.

All caches are scoped to per-call or per-run() lifetime, so they are
automatically invalidated on each watch-mode rerun.

Related Links

• Node perf optimization skill: .agents/skills/node-perf-optimization/

Checklist

• Tests updated (or not required).
• Documentation updated (or not required).