perf(image): make HTMLImageElement.src setter async

[andycall]

Summary

• Route img.src = url through SetBindingPropertyAsync instead of the sync SetBindingProperty path.
• The sync path forced a FlushUICommand per write plus a sync FFI round-trip; both are unnecessary for src because the setter is fire-and-forget, the network load is async on Dart anyway, and any subsequent JS read of img.src already calls FlushUICommand internally before its sync read so it still sees the just-written value.
• Read-after-write semantics, MutationObserver behavior (which never fired for HTMLImageElement to begin with), and the attributes_ mirror are all preserved.

Why

In profiles of the OTC mini-program, 59 sync setProperty(src) calls in a single route render burst forced 59 separate FlushUICommand drains. Each drain entered the styleRecalc cascade producing ~2,000 nested recalc spans per inserted node, contributing to a 16,500-span recalc total (1.69 s self) during the hot 67–69 s window with two 246 ms / 196 ms drawFrames.

Folding these writes into the next natural flush should drop ~250 ms of JS-thread block plus ~600-800 ms of cascading Dart-side styleRecalc work, smoothing the worst frames in the burst.

What's preserved

read scenario	today	after this PR
img.src getter immediately after write	flush + sync read returns new value	flush + sync read returns new value (GetBindingProperty calls FlushUICommand before the sync read at binding_object.cc:311)
img.src = url1; img.src = url2; (last-wins)	last value wins on Dart immediately	last value wins on Dart at next flush
img.getAttribute('src') after img.setAttribute('src', url)	returns url	returns url (unchanged path through ElementAttributes::setAttribute)
img.getAttribute('src') after img.src = url	returns null*	returns null* (pre-existing limitation, attributes_ mirror is WidgetElement-only)

* Pre-existing limitation; not introduced by this PR.

The only actual semantic change is a ≤ 1-frame delay before the network load starts on the Dart side. In practice that's smaller than today's per-write flush stalls (38 ms p95, 250+ ms tail).

Test plan

• Build bridge: npm run build:bridge:macos
• Existing image tests still pass: cd integration_tests && npm run integration (image-related specs in particular)
• New regression check: img.src = url; expect(img.src).toBe(url) returns the just-set value (validates the read-side flush trigger)
• Image onload fires correctly after img.src = url (load lifecycle)
• Re-capture profile from same flow that produced 59-call burst; confirm count drops and styleRecalc cascade shrinks

🤖 Generated with Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
use-case	Error		May 7, 2026 8:21am
vue_usecase	Error		May 7, 2026 8:21am