Samuel Newman
27 lines
1.2 KiB
Markdown
27 lines
1.2 KiB
Markdown
## expo-modules-core Patch
|
|
|
|
This patch contains two unrelated fixes:
|
|
|
|
### Android: bitdrift interceptor
|
|
|
|
Fixes an issue where bitdrift's API stream gets blocked by the Expo interceptor used to power the devtools.
|
|
|
|
### iOS: Hermes startup race in `ExpoBridgeModule.setBridge:`
|
|
|
|
On the legacy bridge (old architecture, where `RCTRuntimeExecutor.h` is
|
|
absent), `setBridge:` installed the Expo runtime synchronously on whatever
|
|
thread called it - typically the main thread, since RN's lazy module-load
|
|
path ignores `+requiresMainQueueSetup`. The `_runtime.didSet` then ran
|
|
`prepareRuntime()` (JSI mutations) on the main thread while the JS thread was
|
|
concurrently inside `JSIExecutor::initializeRuntime()`. Two threads mutating
|
|
the same Hermes runtime corrupted Hades GC, producing intermittent
|
|
`EXC_BAD_ACCESS` launch crashes (e.g. `HadesGC::writeBarrierSlow`,
|
|
`prepareRuntime` / `bindNativePerformanceNow`).
|
|
|
|
The fix hops the runtime install onto `RCTJSThread` so all JSI mutation is
|
|
serialized on the JS thread. This backports the upstream fix discussed in
|
|
expo/expo#45374; the racy `ExpoBridgeModule` is removed entirely in SDK 55+
|
|
(expo/expo#44351), so this patch can be dropped on that upgrade.
|
|
|
|
Refs: expo/expo#43003, expo/expo#45374, expo/expo#44351
|