clamps both movements to the ceiling the island allows failed in run 34021869950 attempt 1 and passed on retry. It is not fixed by #1148 or #1151: ee7627dfc already contains both merge commits, and #1151 only rewrote comments in this file.
What failed
moving — frames later than began + 200 that are not yet where the slice settles — held exactly one entry:
{ at: 433.0999999999767, depth: 16.6802, height: 709.4375 }
One isolated frame, not a run of them. A movement that genuinely overran its ceiling would put every frame between began + 200 and its end into that array.
So either the sampler delivered a single frame across the whole movement, which requestAnimationFrame starvation on a runner with eight workers on four vCPUs would do, or began was set by a reading taken before the movement it is meant to mark — the array's first frame over depth > 0.1. A depth of 16.68 against a resting 18 is 93% of the way, which is a long way into a cubic-bezier(0.22, 1, 0.36, 1) ease and says nothing about which of the two it is.
Not reproduced
15 of 15 passed with the renderer throttled 4x and 8x, and 15 of 15 under twelve busy processes alongside. Heavier machine load only produced page.goto timeouts, which is a different failure.
Where a fix probably goes
Whoever takes this should make the helper report began and the frame gaps around it, so the next occurrence distinguishes the two hypotheses instead of leaving both open.
Split out of #1173.
clamps both movements to the ceiling the island allowsfailed in run 34021869950 attempt 1 and passed on retry. It is not fixed by #1148 or #1151:ee7627dfcalready contains both merge commits, and #1151 only rewrote comments in this file.What failed
moving— frames later thanbegan + 200that are not yet where the slice settles — held exactly one entry:One isolated frame, not a run of them. A movement that genuinely overran its ceiling would put every frame between
began + 200and its end into that array.So either the sampler delivered a single frame across the whole movement, which
requestAnimationFramestarvation on a runner with eight workers on four vCPUs would do, orbeganwas set by a reading taken before the movement it is meant to mark — the array's first frame overdepth > 0.1. Adepthof 16.68 against a resting 18 is 93% of the way, which is a long way into acubic-bezier(0.22, 1, 0.36, 1)ease and says nothing about which of the two it is.Not reproduced
15 of 15 passed with the renderer throttled 4x and 8x, and 15 of 15 under twelve busy processes alongside. Heavier machine load only produced
page.gototimeouts, which is a different failure.Where a fix probably goes
Whoever takes this should make the helper report
beganand the frame gaps around it, so the next occurrence distinguishes the two hypotheses instead of leaving both open.Split out of #1173.