[Cherry]

We noticed this while profiling ESLint on our projects. eslint-plugin-lodash-template calls require("espree") on every parse, and espree is already loaded by ESLint from a different directory, so each call was resolved from disk again. In one lint run that was 297 repeat resolutions taking ~70 ms on Windows, all for packages that use "exports".

Module._findPath() caches results in Module._pathCache, but packages resolved through "exports" return before that write. #32351 added that early return; before it, "exports" targets fell through to the cached path - I'm not sure if this was intentional or not, but can't see any clear reason why.

This caches those results separately, only for bare specifiers and the default conditions, since the _pathCache key doesn't include conditions. Replacing Module._pathCache clears it, and a VFS unmount purges it like the other module caches.

require() of a loaded package from another directory	before	after
with "exports"	8,400 ops/s	~420,000 ops/s
without "exports"	unchanged	unchanged

Windows x64 release build, benchmark/module/module-require-exports.js.

One behavior difference: if node_modules changes on disk while the process runs, for example a closer copy of an "exports" package is installed, or its target file is deleted, a repeat require() with the same lookup paths returns the earlier resolution instead of checking the disk again. Packages without "exports" already behave this way through _pathCache, as "exports" packages did before #32351.

Each entry holds the key and resolved path strings, resulting in about 75 KB for the ~135 entries in that lint run.

[nodejs-github-bot]

Review requested:

• @nodejs/loaders
• @nodejs/performance

[github-actions]

Caution

AgentScan found account activity patterns that may be consistent with automation. This is a heuristic, not proof that this pull request was opened by an agent or violates policy. AI-assisted contributions are permitted, but automated tooling must not open pull requests without advance approval, and contributors must personally understand, test, verify, and take responsibility for every submitted change. See the AgentScan analysis, AI use policy, and automation policy for additional context.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 90.58%. Comparing base (91c4436) to head (3a837a2).

Additional details and impacted files

@@            Coverage Diff             @@
##             main   #66514      +/-   ##
==========================================
- Coverage   92.93%   90.58%   -2.36%     
==========================================
  Files         428      797     +369     
  Lines      195236   278322   +83086     
  Branches    30176    53520   +23344     
==========================================
+ Hits       181442   252114   +70672     
- Misses      13467    16577    +3110     
- Partials      327     9631    +9304     

Files with missing lines	Coverage Δ
lib/internal/modules/cjs/loader.js	98.01% <100.00%> (+19.21%)	⬆️

... and 500 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.