refactor: reduce duplication and name magic constants across lib, speed up AggressiveMergingPlugin - #21525

PR Summary

Medium Risk
Touches core compilation, chunk optimization, runtime injection, and federation hoisting, but the PR positions everything as behavior-preserving with identical test output; the main risk is subtle equivalence bugs in merging selection or hoist paths.

Overview
Behavior-preserving refactors across lib/ plus two targeted wins: faster chunk merging and lower post-make memory from clearing unsafe-cache restore tables.

AggressiveMergingPlugin no longer builds and sorts every chunk pair each pass—it tracks the single best pair in one O(chunks²) scan (same merge choice as before via strict > / first-encountered max).

Compilation clears _restoredUnsafeCacheModuleEntries and _restoredUnsafeCacheEntries in seal’s finalCallback alongside other make-phase bookkeeping so restored modules aren’t pinned for the rest of the compilation.

FileSystemInfo documents Snapshot._flags with named SNAPSHOT_FLAG_* constants (persisted-cache contract; values unchanged). IdHelpers replaces magic lengths and char codes with named constants.

RuntimePlugin registers simple zero-arg tree runtime modules via SIMPLE_TREE_RUNTIME_MODULES instead of many duplicate taps; conditional handlers stay explicit. DefaultStatsPrinterPlugin registers all *_SIMPLE_PRINTERS maps in one loop.

HoistContainerReferencesPlugin shares a traceDep callback and extracts hoistReferencedModules for duplicated hoist logic. ModuleConcatenationPlugin centralizes failure caching in cacheFailure. ResourceHintPlugin adds shared helpers (iterateHintedUrlAssets, type guards, JS_CHUNK_FILE_REGEXP) to dedupe chunk×module×dependency walks.

Patch changeset only; no public API changes.

^{Reviewed by Cursor Bugbot for commit b35dc23. Bugbot is set up for automated code reviews on this repo. Configure here.}

🦋 Changeset detected

Latest commit: b35dc23

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

This PR is packaged and the instant preview is available (e28262e).

Install it locally:

• npm

npm i -D webpack@https://pkg.pr.new/webpack@e28262e

• yarn

yarn add -D webpack@https://pkg.pr.new/webpack@e28262e

• pnpm

pnpm add -D webpack@https://pkg.pr.new/webpack@e28262e

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 93.79%. Comparing base (3151fa6) to head (b35dc23).
⚠️ Report is 5 commits behind head on main.

@@           Coverage Diff            @@
##             main   #21525    +/-   ##
========================================
  Coverage   93.79%   93.79%            
========================================
  Files         620      620            
  Lines       73761    73276   -485     
  Branches    21289    21181   -108     
========================================
- Hits        69183    68732   -451     
+ Misses       4578     4544    -34     

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

Merging this PR will degrade performance by 20.09%

⚡ 1 improved benchmark
❌ 1 regressed benchmark
✅ 214 untouched benchmarks

Performance Changes

_{Comparing refactor/lib-dedupe-and-constants (b35dc23) with main (84d351c)}

Types Coverage

Regarding the CodSpeed memory report — I believe the flagged regression is measurement noise, not a real effect:

• The report flipped between the last two runs even though the code is unchanged. The previous run (1956798) reported an overall ×2.4 improvement (e.g. wasm-modules-async rebuild 814 KB → 195 KB); this run (b35dc23) reports a ~20% regression on a different benchmark (asset-modules-resource rebuild 635 KB → 1202 KB). b35dc23 only edits .changeset/*.md — its lib/ code is byte-for-byte identical to 1956798 — so these sub-MB, watch-rebuild memory benchmarks are dominated by GC/sampling variance (the report also notes "different runtime environments" and, earlier, no successful base run on main).
• The one large, stable benchmark is consistent in both runs: future-defaults production memory improves (9.6 → 7.2 MB, then 10.6 → 8.8 MB). That's the intended effect of clearing the unsafe-cache restore tables after seal.

None of the changes here plausibly increase asset-module rebuild memory (ResourceHintPlugin isn't exercised without resourceHints; the Compilation change only releases memory). So the net memory effect looks like an improvement and the asset-modules-resource delta looks like noise. Happy to acknowledge it on CodSpeed if you'd prefer.

Generated by Claude Code

The runtimes (bun) failure is pre-existing on the base branch and unrelated to this PR — this diff touches zero serialization code. The single failing test is BinaryMiddleware › invalid streams › should throw on a payload from a newer V8 format version, added by #21514 (feat(perf): use v8 for serialization) which merged into main today; CI runs the PR merged into main, so this lane inherited it.

Root cause: the test crafts a payload via v8.serialize([1, "x"]) then sets payload[1] = 0x7f, expecting the version guard to throw. That guard is if (payload[0] === 0xff && payload[1] > V8_FORMAT_VERSION). On Node, v8.serialize emits the V8 wire format ([0xff, 15, …]), so the guard fires. Under Bun, node:v8 uses JavaScriptCore's format (v8.serialize(null) → [13, 0, 0, 0, 4], i.e. payload[0] !== 0xff), so the guard is skipped and nothing throws — the toThrow assertion fails.

Everything else is green (full integration matrix across all OS/Node versions, Codecov patch + project). Will re-run once the base branch's Bun lane is fixed.

Generated by Claude Code