ci: track tree-shaken scenario sizes per PR with size-limit (#2883)#2891
Conversation
Five import-cost scenarios (signals floor / +createStore / +isPending, latest; the render+signal simple-app floor; a representative CSR app) with ~5% headroom limits as the tree-shaking regression gate — a re-coupled feature module costs hundreds of bytes against that headroom. The simple-app ceiling is pinned at 10 KB (brotli) deliberately. The size-limit action comments base-vs-head deltas on every PR; the retained-module-graph test in solid-signals stays the precise which-module diagnostic. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
|
Bugbot is not enabled for your account, so this pull request was not reviewed. Enable Bugbot in the Cursor dashboard to get automatic reviews on future PRs. |
|
Coverage Report for CI Build 29453133491Warning No base build found for commit Coverage: 74.74%Details
Uncovered ChangesNo uncovered changes found. Coverage RegressionsRequires a base build to compare against. How to fix this → Coverage Stats
💛 - Coveralls |
…lta comment The size-limit action compares against the base branch, which fails on any PR whose base lacks the size-limit setup (including this bootstrap PR) with 'Unexpected end of JSON input'. The hard gate is now its own job — install, build, pnpm size on the head only — and the base-vs-head comment job is continue-on-error, so it informs when it can and never blocks. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Merging this PR will not alter performance
Comparing 
Footnotes
|
Adding solid-js/@solidjs/* as root devDeps re-keyed pnpm's peer graph (vitest and vite-plugin-solid instances gained new peer contexts), relocating the benchmark harness's physical dependencies — which CodSpeed reports as instruction-count regressions despite zero code changes. Aliases in .size-limit.js point bare specifiers at the built browser-prod artifacts instead; the lockfile now only adds size-limit's own isolated tree. Same measured numbers on every scenario. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Any change to the workspace dependency graph re-keys pnpm peer instances — removing the workspace devDeps still moved @codspeed/vitest-plugin and vite instance keys (vite's optional terser peer context) — and relocating the benchmark harness shows up as phantom CodSpeed regressions. size/ is now a plain npm package outside the pnpm workspace with its own lockfile; the root package.json and pnpm-lock.yaml are byte-identical to next. Fixtures resolve bare specifiers via esbuild aliases to the built browser-prod artifacts. Same measured numbers on every scenario. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Same isolated-from-the-workspace setup, less top-level clutter; README documents why it must stay outside the pnpm workspace graph. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2 participants
Re-lands the size tracking that was committed to the phase-3 branch after #2890 had already merged (the commit was stranded on the closed branch).
Five import-cost scenarios in
.size-limit.jsonwith ~5% headroom limits as the tree-shaking regression gate — a re-coupled feature module costs hundreds of bytes against that headroom, so a breach means shaking broke (or a deliberate feature: bump the ceiling in the same PR with a reason). The render+one-signal simple-app scenario is pinned at 10 KB (brotli), making that mark a CI invariant. The size-limit GitHub action comments base-vs-head deltas per scenario on every PR;pnpm sizeruns the same gate locally. The retained-module-graph test in solid-signals remains the which-module-re-coupled diagnostic underneath.Current numbers: signals floor 6.76 KB (limit 7.1) · +createStore 11.05 (11.6) · +isPending/latest 8.32 (8.75) · simple app 9.42 (10) · CSR app 11.41 (12), all brotli.
Companion guard for the runtime side: ryansolid/dom-expressions#545.
🤖 Generated with Claude Code