Repository navigation
fix: Remotion ports keep box sizes, the manifest hook runs after format, a drag test stops flaking - #6392
Conversation
There was a problem hiding this comment.
Verdict: REQUEST CHANGES · 1 blocker · 0 important · 0 nits · head 7df9ab55
- blocker — the committed
skills-manifest.jsonhash forremotion-to-hyperframes(cf320d781a133445) doesn't match the skill's contents at this head, so CI's required Skills: manifest in sync job fails. I reproduced it on a clean checkout withgen-skills-manifest.ts --check, which wants37ed4536305b3c64, andmainhasn't touched this skill since the merge base, so the drift comes from this branch. Runbun run --cwd packages/cli gen:skills-manifestand commit the result.
Scope
I confirmed the premise: Remotion's packages/core/src/default-css.ts sets * { box-sizing: border-box }. Sub-compositions are inlined <template>s, not shadow DOM, so one root rule reaches scene-2's bordered .stat-card. boxSizing.test.mjs passes locally (4/4) and the test:skills glob picks it up. I didn't re-run the corpus SSIM, as the PR body says.
|
Fixed at 35375eb: the manifest hash for remotion-to-hyperframes is now 37ed4536305b3c64, the value the check wanted. Cause: the commit hook's format pass reflowed api-map.md after the manifest was generated, so the hash I committed was of the unformatted file. gen-skills-manifest.ts --check passes on this head. |
…ordered boxes keep their size
35375eb to
1b2cee7
Compare
There was a problem hiding this comment.
Verdict: REQUEST CHANGES · 1 blocker · 0 important · 0 nits · head 1b2cee72
The box-sizing and manifest commits are unchanged from the head I approved. The piped format → skills-manifest group passes lefthook validate on 2.1.4, and lefthook-order.test.mjs is wired into test:scripts. The one blocker is in the drag-bench change, inline at sequences.mjs:277.
Edit accuracy: accurate 271 (base branch 271), smooth 129 of thoseThe gate passes. Quarantined, measured but not gated (0) Unstable (2)
|
|
Fixed at 6018d37 as suggested: the waiting marker is set on the selection overlay (previewGestureStarts.ts boxRef), so it belongs to the selected element; another element is now read bare until it is selected, and the marker-aware read is used only when the element is already selected. Ran the ab, held and inflight sequence cases (14 cases) 3 times on a 2-CPU Linux box: no verdict changed against baseline.json. |
Three small fixes in one PR (each is under the 100-line floor alone). They are independent; the commits are separate.
1. A Remotion port keeps its bordered boxes at their size
Remotion's render page sets
* { box-sizing: border-box }, so awidththere already includes the border and padding. A browser's own default iscontent-box, so a port without the rule draws every bordered or padded element larger. A box withwidth:200px; height:100px; padding:20px; border:10px solidmeasures 200x100 with the rule and 260x160 without, in Chrome.SKILL.mdstep 2 tells the port to put* { box-sizing: border-box; }once in the rootindex.html<style>;references/api-map.mdhas the row and an example.<style>.skills/remotion-to-hyperframes/scripts/tests/boxSizing.test.mjsgoes red if a corpus root loses the rule (3 pass, 1 fail with the tier-1 rule removed).2. The commit hook hashes the skills manifest after the formatter has run
pre-commitranformatandskills-manifestin parallel, so the manifest could be hashed from a skill document the formatter was about to reflow; CI's "Skills: manifest in sync" then failed (this is what happened to the first version of this PR). They are now a piped group,formatfirst; every other job is still parallel.commands:becamejobs:because lefthook only has groups underjobs.scripts/lefthook-order.test.mjs(intest:scripts) fails on the old file ("skills-manifest must sit in a group") and passes on this one.3. The edit accuracy drag case stops failing at random
seqheldtwice-tlhold-px-r0-nested-z100failed now and then withpressJump312.29 andtrackingabout 120. When a press lands while the previous save is in flight, Studio queues the drag, shows a "press waiting" marker where the box is meant to be, and replays the drag when the save lands; the element itself sits at its base position until then. The bench's drag step took its baseline from the bare element, so it read 312 px (the move's own vector) off. A user cannot see it: with the marker counted, the painted frames show the move continuous at 0.07 px.measureShownshared with the drag measurement. The press still lands with the earlier save in flight.pressJump312.29. After: 30 of 30 pass the gate's verdict,pressJumpat most 0.11.settle.test.mjscoversmeasureShown(red if the marker is ignored).