Repository navigation
Conversation
Throwaway branch. Contains only the workflow files from feature/api-surface-hardening on top of origin/main, so the Linux screenshot validation runs against main's picker code. Used to determine whether the 9 screenshot failures on that PR come from its changes or from the reference images being machine-specific. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
Important Review skippedDraft detected. Please check the settings in the CodeRabbit UI or the ⚙️ Run configurationConfiguration used: defaults Review profile: CHILL Plan: Pro Plus Run ID: You can disable this status message by setting the Use the checkbox below for a quick retry:
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
The gate failed all 9 screenshot comparisons on the Linux runner. A baseline run of the same workflow against unmodified main (PR #156) failed identically, so the cause is the reference images, not any change in this branch. Measured it rather than guessing: | | differing pixels | per-pixel delta | | --- | --- | --- | | Same references, Linux runner vs recording host | 0.164%-0.888% | exactly 1/255 | | One digit changed (LocalTime(13, 5) -> 13, 6), same host | 0.189% | up to 225/255 | Host noise is 16-90x the configured 0.01% `imageDifferenceThreshold`, and raising the threshold is not a fix: the setting counts differing pixels with no per-pixel tolerance, so any value large enough to pass on another host (>=0.9%) also masks a one-digit regression at 0.189%. What separates the two is how much each pixel differs - 1 versus 225 - which the setting cannot express. So the task stays where it already worked: locally on the recording machine and in the manual workflow. Making it a hosted gate means picking a canonical rendering environment and re-recording every reference there, which inverts the problem for local runs; that trade-off is the maintainer's to make and is recorded in docs/testing/compose-screenshot-tests.md, replacing the previous claim that real regressions are "far larger" than host noise. Also fixed the manual workflow's screenshot artifact, which uploaded the HTML report without the images it links to, and tidied the 0.8.0 changelog section into one block per heading. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
확인 완료. main 원본에서도 스크린샷 9개가 동일하게 실패해, #155의 변경과 무관한 기존 문제임이 확인됐습니다. 측정 결과:
호스트 노이즈가 설정된 #155에서 스크린샷 검증을 호스티드 PR 게이트에서 제외하고 |
임시 PR입니다. 머지하지 않고 확인 후 닫습니다.
#155에서 Linux runner의
validateDebugScreenshotTest가 9개 전부 실패했습니다. 로컬(macOS arm64)에서는clean후 재검증까지 통과하므로, 원인이 #155의 변경인지 기준 이미지가 머신 종속인지 구분해야 합니다.이 브랜치는
origin/main+ 워크플로 파일 2개뿐입니다. picker 코드는 main 그대로입니다.🤖 Generated with Claude Code