Pplt 6073 : Make regression tolerance higher - #2431
Conversation
|
Important Review skippedAuto reviews are disabled on base/target branches other than the default branch. Please check the settings in the CodeRabbit UI or the ⚙️ Run configurationConfiguration used: Workspace UI Review profile: CHILL Plan: Advanced Run ID: You can disable this status message by setting the Use the checkbox below for a quick retry:
No actionable comments were generated in the recent review. 🎉 ℹ️ Recent review info⚙️ Run configurationConfiguration used: Workspace UI Review profile: CHILL Plan: Advanced Run ID: 📒 Files selected for processing (22)
Included review availability: Your plan provides up to 1 included review per hour; 0 remain after this review. 📝 WalkthroughWalkthroughThe PR promotes the repository packages from beta ChangesStable release promotion
PDF validation updates
Priority: ➖ Normal Estimated code review effort: 2 (Simple) | ~15 minutes Change: Other Suggested reviewers: Merge Risk: ⚪ Minimal · up to The release promotion and PDF validation updates do not show a concrete current-head risk requiring changes before merge. 🚥 Pre-merge checks | ✅ 5✅ Passed checks (5 passed)
✨ Finishing Touches📝 Generate docstrings
🧪 Generate unit tests (beta)
Comment |
Two CI failures on the 1.32.10 release PR, neither a product bug.
Regression (Track P): `jack sparrow resume.pdf` diffed by 648 pixels against
a 600 budget, in box [369, 1447, 692, 1448] at a channel delta of 4. That box
is not the circular photo clip the 600 was measured against — it is the
hairline rule under the LANGUAGES heading, whose two pixel rows are the only
non-white rows across those 324 columns. A second, previously unmeasured
source of run-to-run jitter on a fixed Chromium build (the manifest browser
and the run browser matched, so the goldens are not stale). Budget raised to
2000, ~2x the 918-pixel worst case of both sources landing at once and still
0.1% of the page, with the second source documented.
@percy/core: all 1322 specs passed; the job failed nyc's 100% threshold on
two uncovered spots.
- pdf-rasterize.js:128, the asInputError rethrow when a rendered page comes
back under Percy's 10px floor. fitScale only clamps the upper bound, so a
4x4pt MediaBox rasterizes to 8x8 at scale 2 and trips it. The new test
also pins that this surfaces as a 400, not a 500.
- pdf-snapshot.js:67-68, the decoded-size check. The existing oversize test
trips the encoded cap first, leaving the decoded one unreachable. The cap
rounds up to a whole base64 quantum, so 52428801 bytes encodes to exactly
MAX_PDF_BASE64_CHARS and clears it — the two-byte window the second check
exists to close.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
d1339ef to
96d0cc6
Compare
Summary by CodeRabbit
Release
Bug Fixes