test(studio): fail edit accuracy cases whose preview flashes or paints an edit late - #4796
Draft
miguel-heygen wants to merge 12 commits into
Draft
miguel-heygen wants to merge 12 commits into
miguel-heygen wants to merge 12 commits into
Conversation
Work in progress: the ratchet and its tests; the CI job follows.
Shares the built CLI from the Build job, re-runs twice each case that passes on the base branch and failed, and posts one sticky comment from the gate job.
… banking them The shards re-ran only regressions, so one lucky pass of a base-failing case failed the gate as unbanked and was never listed as unstable. Every case whose verdict differs from the base branch, either way, is now re-run twice and judged 2 of 3. A render error fails the gate like an unsettled snapshot, a malformed baseline.json fails the job instead of reading as empty, and the header is 3 lines.
…s off-state frames
…s an edit late Edit-to-paint: the first frame after a pointer-up, key, undo or redo shows the after-state. A metric the base baseline never measured cannot regress the gate; it counts for banking.
…t a blank control PNG capture misses 11-23% of frames even on a blank page, so every case read uncovered.
… than a dropped frame A pointer-up whose next frame arrived 507 ms later passed on frame count alone.
…so flash capture keeps up Under swiftshader GL the screencast saw 40-70% of Studio's frames; software compositing sees ~95%.
…sh controls and input log The previous commit only switched the frame check to the live URL; the iframe walk did not land.
Edit accuracy: 264 passing here, 434 on the base branchThe gate fails: 170 case(s) that pass on the base branch fail here; the passing count fell from 434 to 264; baseline.json marks 434 failing case(s) as passing. Regressed (170)
Marked passing in baseline.json but failing (434)
|
This branch has not been deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What
The edit accuracy bench gains two metrics that feed the case pass:
How
Page.startScreencast(JPEG q95, every frame, each frame acked at once) runs only in the untraced windows after the drag; the drag keeps its trace.uncovered(never a pass) under 90%, or under the same run's blank-page control minus 5 points. The table lists each case's 1-frame miss chance (1 - coverage).--disable-gpu-compositing: under swiftshader GL the screencast saw 40-70% of Studio's frames during heavy windows (Chrome drew 87 frames, the screencast delivered 35); with software compositing it sees about 95%.Controls
Main at ca3a3f6, this bench, one case (
rotate-none-px-r0-root-z100) unless named, one job per run.move-none-px-r0-nested-z100, with marker"Caught" for a blink means a preview-pane bad frame within 4 frames of the pointer-up (the blink is injected by a page call after the release). No gesture on main is free of flashes today: every case screened shows the timeline emptying for several frames on undo and redo, and rotate shows 1-2 frames of mixed selection chrome after release. So a whole-case "0 flashes over 10 runs" control is not available yet; the no-move tap stands in for it. With and without the marker the counts are separate runs of a defect whose length varies, and without the marker repeated frames cannot be deduped, so they do not agree within one frame.
Slice (gsap=none, main ca3a3f6)
(filled in with the run)
Before
Without the flash metric the bench saw only the settled undo, which matches the state before the move.
After
The flash metric flags the frames in between: mid-undo, the timeline shows no clips at all.