Repository navigation
fix(ci): re-measure the ModelEffortDropdown chunk's bundle-size ceiling - #23867
Conversation
Pristine main @ 0dab198 builds the shared chat-route chunk Rolldown names after the composer's model/effort picker at 517,094 B (505.0 KB), 5,094 B over the 500 KB default the Bundle Size Gate applies to a chunk with no entry of its own, so the gate fails on the merge ref of every open website PR (#23618 first) with no PR code in the tree. Attribution is measured per main commit since fada42f, where the chunk was 502,179 B: eight website commits of ordinary chat-route growth added 5,664 B, and #23271 (1a7190c) added one module and 9,251 B. No dependency joined the chunk. Give it a CHUNK_BUDGETS entry at the table's ~5% convention over the measurement (531 KB), with the attribution in its note, the way #19089 and #15514 re-measured App. Fixes #23866
First Principles Review (Opus 5.5) — ⏭️ skippedRevision |
UX Review (Opus 5.5) — ✅ PASSUX-level review of UX-Verdict: PASS Nothing to check. The only change is a bundle-size budget in a CI build script, so no user-visible control is added or changed and the missing screenshots leave no evidence gap. [UX-REVIEWED] 2e07c4b Description read[DESCRIPTION-READ] 82d888ccadf9cbe0203c83c1347fc5eb46b68ac6fdc82ff602e2e728bc84f98b That sha256 names everything this verdict read: the PR description, plus the 3 evidence file(s) the workflow collected from it and handed this review. It covers that evidence because the media strip replaces every attachment URL with the same placeholder, so an edit swapping one attachment for another would leave the description's captured text identical while what this review judged changed. Recompute it with |
Design Review (Opus 5.5) — ✅ PASSDesign-level review of Design-Verdict: PASS Nothing blocks: this is a measured, attributed re-baseline of one budget entry, done the way the table already prescribes. Notes
[DESIGN-REVIEWED] 2e07c4b Description read[DESCRIPTION-READ] e99081723a11eba2cd71dc1524637414b13bc0876a50ff18ee1274df7d119471 That sha256 names everything this verdict read: the PR description, plus the 1 evidence file(s) the workflow collected from it and handed this review. It covers that evidence because the media strip replaces every attachment URL with the same placeholder, so an edit swapping one attachment for another would leave the description's captured text identical while what this review judged changed. Recompute it with |
GPT 6.1 Review — ✅ no blocking findingsGPT 6.1 completed its review of This comment is updated in place on each push. Review detailsNo findings. False positive or not applicable? A repository writer can comment: |
Opus 5.5 Review — ✅ no blocking findingsReviewed Review detailsNo findings. Notes:
[OPUS-REVIEWED] 2e07c4b Verdict parsed from the review's SHA-scoped output markers for commit False positive or not applicable? A repository writer can comment: |
buluoray
left a comment
There was a problem hiding this comment.
Approve @ 2e07c4b. Fixes red main (#23866): main @ 0dab198 builds ModelEffortDropdown at 517,094 B against the 500 KB default. The new 531 KB entry follows the table's ~5% headroom convention and also covers #23697's 506.7 KB measurement. Bundle Size Gate and every review lane are green on this head.
Problem / Motivation
Goal: The Bundle Size Gate passes again on pristine
main, so website PRs go red only for growth they cause.Pristine
main@ 0dab198 fails the gate with no PR code in the tree:The gate runs on the PR merge ref, so every open PR that touches
website/now fails it whatever it changes; #23618 was the first (its own tree builds the chunk at 504,901 B, under budget).Why it matters
A gate that is red on main's own tip blocks every website PR and stops catching the regression it exists for (a new library landing in a chunk), because authors learn to read it as noise.
Not a goal
import()seam is a design change for the chat route, not a gate fix.What changed (motivation → approach → change)
Measured first. A detached worktree on
origin/main@ 0dab198 withnpm ciandnpx vite build --mode analyzeemits the chunk at 517,094 B (163 modules);node scripts/check-bundle-size.mjsfails on it exactly as CI does. Then, per main commit since fada42f (the PR base where the gate still passed):No dependency joined the chunk: the only
package.jsonchange in the range is #23634's React Compiler toolchain, with an empty opt-in list.The change is one
CHUNK_BUDGETSentry inwebsite/scripts/check-bundle-size.mjs,ModelEffortDropdown: 531 * KB, at the table's ~5% convention over the measurement (517,094 B × 1.05 = 542,949 B, rounded up to the whole-KB unit; 26,650 B / 5.15% headroom), with the attribution above in its note. Same shape as #19089 and #15514 for the App chunk.Backwards compatibility
Compatible: a CI gate's ceiling moves; no shipped code changes.
Removes nothing: one table entry is added.
Tests
Proof: not provable -- docs or config only (a CI gate budget table). Verified by hand: with pristine main's
bundle-report.jsonin place,node scripts/check-bundle-size.mjsgoes fromFAIL ... ModelEffortDropdownto927 chunks within budget (default 500.0 KB, 25 allowlisted).Manual verification
git worktree add --detach <dir> origin/main && cd <dir>/website && npm ci && npx vite build --mode analyze && node scripts/check-bundle-size.mjs→ FAIL on main's tip.python3 scripts/check_file_line_cap.py→ 0 failures.Related Issues
Fixes #23866
Refs #23618
Sibling sweep
Swept:
node scripts/check-bundle-size.mjsover pristine main's analyze build (every emitted chunk against its budget).Siblings: none -- it is the only chunk over budget on main @ 0dab198; the other 926 pass.
Pattern harvest
Not generalizable: the same recurrence the App entry's notes document (a chunk reaching a fixed ceiling through ordinary first-party growth); the fix each time is a re-measurement with attribution, which this table already prescribes.
Checklist
feat|fix|docs|style|refactor|perf|test|chore|ci|build|revert: ...)Contribution License Agreement
Contributions are accepted under the project's Apache-2.0 license; see the Licensing section of CONTRIBUTING.md.