test(SnapPaging): edge-move touches land on the scene's clock - #11068

Maintainer impact

On a starved runner, three SnapPaging interaction cases failed with raw assertions or a plain freeze timeout. Two are touch cases during an edge button's move, with a 150 ms and a 40 ms hold. The third checks that the scroll offset holds through a swipe. A raw failure reads as a product red, though such a run proves nothing about the product.

Intended invariant

On a runner with CI's CPU limits, the cases prove what they proved. A run whose input or frames the runner held back fails by name, as the harness's LiftPlacementError, FrameStarvationError and touch pacing contracts state.

Change

• The edge-move touch cases sent their tap after the thaw. A loaded runner's round trips could deliver it after the move ended. The row then rested on the move's target, and the case failed raw. The touch-down is now held to 100 ms into the move on the scene's clock (holdInputUntil). One that reaches the page later fails as LiftPlacementError.
• The press-in-settle freeze now logs the row on every frame it checks. A row that crossed its freeze window between two frames more than the starvation gap apart fails as FrameStarvationError, through requireFrames. Any other timeout fails as before.
• The swipe case judges the release the page reads against the advance speed, as the file's swipe helper does. A release read on the other side fails by name.

Evidence

• Failure before the change, on main's specs: 32 raw failures in 60 WebKit runs under load (Testing).
• Success after the change: no raw failure in 60 WebKit runs under the same load. On CI's profile, every case passes in both engines.
• Product/runtime behavior verified unchanged: no product file changes.
• CI impact: none; no workflow, action or CI script changes.

Scope

Components: SnapPaging

• No intended public API, product behavior, visual direction, or policy change.
• Product changes discovered during the work were removed or split.
• Public text and artifacts contain no internal Meta context.

Testing

• The load: WebKit ran in its Playwright container, limited to 2 CPUs on one pinned pair. Two busy loops shared that pair, with 4 workers.
• All runs used main's build at d130dff, with main's specs or this head's (e1ed32e). The runs predate one comment edit. Each case ran 20 times with each.
• The swipe case: main's specs 8 raw failures (a resting offset), 9 passes and 3 named. This head's: 18 passes and 2 TouchPageStallError.
• The 150 ms touch during an edge button's move: main's specs 13 raw failures and 7 passes. Each raw failure is a freeze timeout or a row on the move's target. This head's: 14 LiftPlacementError and 6 FrameStarvationError.
• The 40 ms touch: main's specs 11 raw failures and 9 passes. This head's: 17 LiftPlacementError and 3 FrameStarvationError.
• Under this load, the touch reached the page 36 to 602 ms after its time. So no edge-move touch run could pass, and each says so by name.
• On CI's profile at 1 worker: the swipe case and every press-in-settle case pass 3 of 3 in each engine.
• Strict ESLint reports 0, and Prettier is clean on the changed file. Core tsc reports 0 errors. No comment sentence on the changed lines runs over 25 words.