fix(RuntimeScheduler): flush JS microtasks after work-loop to prevent…

[SudoPlz]

🐛 Fabric + Reanimated microtasks never flush (animations freeze)

TL;DR
React-Freeze’s infiniteThenable leaves Fabric’s RuntimeScheduler_Legacy stuck in an endless work-loop.
JS microtasks never flush → Reanimated worklets never run → all animations stall.
This issue proposes calling performMicrotaskCheckpoint(runtime) inside the loop so the JS queue is processed each iteration.

---

Symptoms

⚠️ What we saw	Notes
Animations randomly freeze	Reanimated worklets never execute.
global.queueMicrotask logs enqueue events	but global.__callMicrotasks never logs execution.
Adding a lone console.log “fixes” the bug	crossing the bridge forces a microtask flush (Heisenbug).

---

Investigation summary

1. React-Freeze (v1.0.4) suspends trees with an infiniteThenable.
2. React repeatedly calls commitRootImpl → schedules tasks via Scheduler.unstable_scheduleCallback.
3. Native side enters RuntimeScheduler_Legacy::startWorkLoop and spins forever:

while (!taskQueue_.empty()) {
  // process task …
}

4. Because the loop never finishes, JS is never asked to flush its microtasks (MessageQueue.__callReactNativeMicrotasks).
5. Console logs appear to “fix” the bug because they cross the RN bridge and force a JS flush.

---

Proposed fix (implemented in PR)

• File: RuntimeScheduler_Legacy.cpp
• Change: after each executeTask(...) call we now run

performMicrotaskCheckpoint(runtime);          // NEW

• performMicrotaskCheckpoint calls the global _flushReactNativeMicrotasks JS helper (already registered by JSTimers) which empties the queue via JSTimers.callReactNativeMicrotasks().

Why this is safe

• Only runs after a task is executed (identical spot where the queue would normally be flushed if the loop exited).
• No-op cost when queue is empty.
• Works in classic bridge and Fabric-bridgeless.
• Is actually the way the standard modern runtime scheduler works (when in bridgeless)

---

Ticket

https://app.asana.com/1/236888843494340/project/1199705967702853/task/1210188774106277?focus=true

---

Credits

Big shout-out to Szymon for spotting the infinite loop and mapping the bridge call-chain! 🎉

[Flewp]

Nice! 🎉

Is this upstream-able? Should it be?
Any expected impact on CPU usage?

[SudoPlz]

@Flewp yes it's upstreamable, I didn't notice any impact on CPU usage, essentially it's flushing the microtask queue after every task completion, which is the same thing the modern RuntimeScheduler does as well.

This PR is bringing that change to the Legacy runtime scheduler (the one we're using now).

After we turn bridgeless mode on, we will be using the modern RuntimeScheduler and this change will be useless, but for the time being we need to flush the micro task queue manually because of react-freeze.