Repository navigation
useEffect + setState panics with RefCell already borrowed on macOS ARM64 #4
Description
Hey! Humongous fan of perry!
Been playing around with perry-react today and hit a crash that I think is also responsible for the weather-app demo crashing on launch.
Calling a state setter inside useEffect panics every time:
function App() {
const [status, setStatus] = useState("idle")
useEffect(() => {
setStatus("started")
}, [])
return <div><h1>Status: {status}</h1></div>
}
thread '<unnamed>' panicked at crates/perry-ui-macos/src/state.rs:382:13:
RefCell already borrowed
My guess is the useEffect fires and triggers a re-render while the initial render is still holding a borrow on the state RefCell.
I initially found this when cloning & testing the weather-app and it crashes on launch with the same abort — pretty sure it's the same root cause since it calls setState from inside a fetch callback on startup. I couldn't get weather-app working so started a new react repo and ultimately hit this error
macOS ARM64, built from source, perry-react from PerryTS/react.
Activity
Fix RefCell state callback panic (GH-4) and fetch linker error (GH-5)
proggeramlug commentedon Mar 14, 2026
Hey Thomas, thanks for the detailed report — your diagnosis was spot on.
Perry-react is admittedly the least tested part of our current UI surface, so really appreciate you putting it through its paces and taking the time to file this.
Root cause:
state_set()held an immutable borrow on theON_CHANGE_CALLBACKSRefCell while iterating and invoking user closures viajs_closure_call1. In perry-react,useEffectfires synchronously during the initial render — so whensetStatus("started")triggersstate_set→onChangecallback →_scheduleRerender→ component re-render → any newuseStatecallssig.onChange()→state_on_change()triesborrow_mut()on the same RefCell → panic.Fix:
state_set()now snapshots the callback list into a localVecbefore invoking any closures, releasing the RefCell borrow first. Same treatment applied toFOR_EACH_BINDINGS(which also calls user closures). Fixed across all 5 platform crates (macOS, iOS, GTK4, Windows, Android).Fixed in
b521064.tomkennedy22 commentedon Mar 14, 2026
Awesome! Amazing response time, much appreciated. I'll test it out more on Saturday
feat(ui): #185 Phase C step 6 — hex strings + gradient (v0.5.310)
perf(transform): static-trip-count for-loop full-unroll + smart chann…
fix(compile): honor "*"/"@scope/*" wildcard in perry.compilePackages …
fix: class async / async-generator method test262 parity
fix: class async / async-generator method test262 parity (#4768)
fix(compile): warm auto-optimize cache dropped CPU-only well-known li…
fix(codegen): globalize module-init lets nested in destructure Try sc…
perf(codegen+runtime): representation-selection Phase 4a — fast plain…
fix(gc): run old-gen reclaim precisely at safepoints, stop scanning o…
fix(object): a Buffer/DataView receiver must not reach the ordinary o…
perf(runtime): route the shadow-stack TLS through the hot-cache fast …
perf(runtime): route the shadow-stack TLS through the hot-cache fast …
refactor(stdlib): remove nanoid native binding
refactor(stdlib): remove nanoid native binding
revert(runtime): drop the canonical slot-index carry (2b2cb6e, 3f73dd4)
revert(runtime): drop the canonical slot-index carry (2b2cb6e, 3f73dd4)
revert(runtime): drop the canonical slot-index carry (2b2cb6e, 3f73dd4)
revert(runtime): drop the canonical slot-index carry (2b2cb6e, 3f73dd4)
revert(runtime): drop the canonical slot-index carry (2b2cb6e, 3f73dd4)
gc-effects: nounwind from the linked table (slice 9); keys arrays res…
cargo config: Android std has real thread-locals (one emutls key, not…