fix: do not deadlock when the diff runs inside a view transition

[aralroca]

The bug

wait() settled the stream walk on requestAnimationFrame alone:

const wait = () => new Promise((resolve) => requestAnimationFrame(resolve));

rAF does not fire while a document is rendering-suppressed, and that is exactly what the View Transition API does to it for the whole duration of an update callback. Verified in Chrome:

callback entered
rAF did NOT fire within 1500ms (+1509ms)
microtask ok
setTimeout fired (+1519ms)

So a caller that wraps the whole diff in document.startViewTransition deadlocks here until the browser abandons the transition on its 4-second DOM-update timeout — the page freezes for four seconds and then swaps with no animation at all:

start
ready REJECTED: TimeoutError: Transition was aborted because of timeout in DOM update
updateDone +4009ms
finished +4009ms

This is not an exotic call pattern: it is the one this README already documents under Incremental vs full transition, and it is the only way to pair shared view-transition-name elements across a full page change. transition: true cannot do it — it starts one transition per DOM update and each new one skips the previous, so on a real page swap I measured 51 transitions started, 50 skipped, 1 animated: the page snaps through and only the last fragment fades.

The fix

Race a timer against the frame. Outside a transition the frame still arrives first, so the normal path is unchanged; inside one the timer is the only thing that fires. It also keeps the walk moving in a background tab, where frames are throttled away for the same reason.

-const wait = () => new Promise((resolve) => requestAnimationFrame(resolve));
+const wait = () =>
+  new Promise((resolve) => (requestAnimationFrame(resolve), setTimeout(resolve, 16)));

After, same page swap:

start
updateDone +23ms
ready +24ms
finished +282ms

One transition, the diff completing in 23 ms inside the callback, and a real 282 ms animation.

Test

should complete a diff that runs inside document.startViewTransition — runs a real diff inside a real startViewTransition and asserts the DOM updated, ready resolved (a timed-out transition still applies the change, so asserting only the DOM would pass on the bug), and the callback finished well inside the 4 s cap.

Red before the fix, green after:

	result
before	✗ expected "resolved", received "rejected: TimeoutError" — 4137 ms
after	✓ 643 ms

Full suite: 138 pass, 0 fail across Chrome, Firefox and Safari.

One thing to decide: the bundle budget

The fix costs 15 bytes gzipped (1479 → 1494) against a bundlewatch maxSize of 1.5 kB, so it passes with 6 bytes of headroom. That is tight enough that the next change is likely to trip it. Worth bumping maxSize to 1.6 kB in the same PR if you agree — say the word and I will add it.

🤖 Generated with Claude Code