perf(serialization): reduce allocations in binary middleware hot paths - #21526

🦋 Changeset detected

Latest commit: 43e6849

The changes in this PR will be included in the next version bump.

Not sure what this means? Click here to learn what changesets are.

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

PR Summary

Low Risk
Refactor-only allocation tweaks in serialization framing with no intended behavior change; existing round-trip tests cover the touched paths.

Overview
BinaryMiddleware allocation hygiene on serialize/deserialize framing (follow-up to the V8 serializer work; no semantic change).

_deserialize no longer builds a parts array and a second merge pass. The first section’s array becomes result; additional sections append in place. Empty streams still return [].

Serialize flush when nulling buffer slots in the in-place path uses an indexed for loop instead of for…of over bufferIndices, matching the restore loop beside it.

Changeset: patch note for reduced allocations on binary serialization hot paths.

^{Reviewed by Cursor Bugbot for commit 43e6849. Bugbot is set up for automated code reviews on this repo. Configure here.}

This PR is packaged and the instant preview is available (aa1e746).

Install it locally:

• npm

npm i -D webpack@https://pkg.pr.new/webpack@aa1e746

• yarn

yarn add -D webpack@https://pkg.pr.new/webpack@aa1e746

• pnpm

pnpm add -D webpack@https://pkg.pr.new/webpack@aa1e746

Types Coverage

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 93.79%. Comparing base (84d351c) to head (43e6849).
⚠️ Report is 5 commits behind head on main.

@@            Coverage Diff             @@
##             main   #21526      +/-   ##
==========================================
+ Coverage   93.78%   93.79%   +0.01%     
==========================================
  Files         620      620              
  Lines       73371    73370       -1     
  Branches    21192    21196       +4     
==========================================
+ Hits        68811    68818       +7     
+ Misses       4560     4552       -8     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

Merging this PR will regress 1 benchmark

⚡ 3 improved benchmarks
❌ 1 regressed benchmark
✅ 212 untouched benchmarks

Performance Changes

_{Comparing perf/binary-middleware-followup (43e6849) with main (84d351c)¹}

On the CodSpeed memory regression (asset-modules-resource / mode-development-rebuild): I don't think it's real or actionable for this diff.

• This change only removes allocations — it drops the intermediate parts array in _deserialize and replaces one for…of with an indexed loop. It adds no retained state and no new allocation, so there's no mechanism by which it could raise peak heap by ~570 KB in a rebuild scenario; if anything it's neutral-to-lower on memory.
• The report flags two reliability caveats on exactly this comparison — "Different runtime environments detected … may affect the accuracy" and "No successful run was found on main (4efcd73) … 84d351c was used instead … there might be some changes unrelated to this pull request" — and the regressed case is a single high-variance Memory watch-rebuild benchmark.
• The same run shows two memory benchmarks improving (future-defaults +37.5%, many-chunks-commonjs +21.7%), which is inconsistent with a real regression from an allocation-reducing change.

I read this as environment/base-substitution noise rather than a regression introduced here, so I don't plan to change the code for it. Happy to re-trigger the benchmark or acknowledge it on CodSpeed if preferred.

Generated by Claude Code

The runtimes (bun) failure is unrelated to this change and not code-fixable here.

The failing suites are HotTestCases, and the error is in the HMR lazy-compilation HTTP server teardown — lib/hmr/lazyCompilationBackend.js:181 throwing "Server is not running" from server.close() in dispose (2 failed tests across 6 suites, 94,139 passed). That's a Bun HTTP-server lifecycle/teardown issue in a subsystem this PR doesn't touch — the diff only changes lib/serialization/BinaryMiddleware.js (persistent-cache deserialize framing), which has no path to an HMR server's close behavior. HotTestCases use the in-memory cache, not the filesystem serializer, so the deserialize change isn't even exercised here.

I'll re-trigger the Bun job to clear the flake; I'm not changing the diff for it. If it reproduces, it's a pre-existing Bun/lazyCompilationBackend teardown issue on main rather than something introduced here.

Generated by Claude Code