You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
[Due for payment 2026-09-22] [$250] [Sentry: APP-93Z] HybridApp iOS EXC_BREAKPOINT in CACElementNumberStabilizer mergeLists #96177
iOS 26 Accessibility element-number stabilizer crashes during list merge, triggered from a background TLS certificate serialization path — may be an OS-level regression interacting with network activity.
Reproduction
Unknown — see Sentry events linked from the Sentry issue.
The crash frame CACElementNumberStabilizer mergeLists belongs to Apple's private Core Accessibility framework, not to first-party JavaScript. On iOS 26, when VoiceOver/accessibility recomputes element ordering it calls mergeLists, and the sibling frames (_xzm_xzone_malloc_freelist_outlined, SecTrustSerialize, NWIOConnection TLS trust evaluation) show this runs concurrently with a background TLS certificate serialization. The EXC_BREAKPOINT is a malloc freelist integrity trap that fires when the accessibility element list is being mutated on one thread while the OS stabilizer merges it on another. So the user-visible consequence (a hard crash on iOS 26 HybridApp, versions 9.4.25 through 9.4.31) is triggered inside Apple code that the app cannot call or patch directly.
The app-side factor that raises exposure is how often we rebuild large accessibility element trees. Heavy SelectionList screens render many accessibility nodes per row. UserSelectionListItem re-derives display data and passes keyForList={item.keyForList} on every render:
When keyForList or item identity changes frequently, rows remount and the native accessibility element list is torn down and merged again. The more often that merge runs while background network/TLS activity is live, the more often iOS 26's stabilizer hits the unstable freelist and traps.
What changes do you think we should make in order to solve the problem?
Confirm the scope first: this is an iOS 26 OS regression, so the crash cannot be fully eliminated from JS. The realistic goal is to reduce how often the native accessibility element tree is rebuilt during list rendering, which reduces how often the Apple stabilizer runs and can trap.
Stabilize the accessibility element count coming from SelectionList rows. In UserSelectionListItem.tsx, the memoized userHandle and userDisplayName are already stable, but the row still exposes each icon and control as its own accessibility element. Group the row into a single accessibility element so the stabilizer merges fewer nodes:
Collapsing each row to one accessibility element shrinks the list the OS must merge, which lowers the chance of tripping mergeLists.
Because the crash originates in Apple's framework, also add a Sentry inbound-filter/beforeSend entry for the CACElementNumberStabilizer mergeLists signature so this OS-level noise stops dominating the dashboard while the mitigation ships and iOS point releases land. This uses the existing Sentry wiring referenced by the stub in sentry-react-native.ts:
What alternative solutions did you explore? (Optional)
Reminder: Please use plain English, be brief and avoid jargon. Feel free to use images, charts or pseudo-code if necessary. Do not post large multi-line diffs or write walls of text. Do not create PRs unless you have been hired for this job.
Please re-state the problem that we are trying to solve in this issue.
iOS 26 HybridApp is hard-crashing with an EXC_BREAKPOINT whose top frames flip between Apple's accessibility element merge (CACElementNumberStabilizer mergeLists) and a background TLS trust evaluation (SecTrustSerialize / NWIOConnection), with _xzm_xzone_malloc_freelist_outlined sitting in the middle. 186 users, app versions 9.4.25 to 9.4.31, no known repro.
What is the root cause of that problem?
The only frame that names the actual failure is _xzm_xzone_malloc_freelist_outlined. That is iOS's malloc catching a corrupted zone freelist and trapping with EXC_BREAKPOINT. So this is a heap-corruption crash, not an accessibility logic bug. The bad write happens earlier, and the trap fires on the next allocation that touches the poisoned zone. That is why the "cause" frame changes between events: accessibility element merging in some, TLS certificate serialization in others. Whoever corrupts the heap has to live in both of those worlds.
The one thing in the app that instruments both the view/accessibility tree and background TLS network activity is FullStory. Its session-replay engine walks the native view hierarchy including accessibility elements to build each replay frame, and it continuously uploads captured session bundles over HTTPS on background threads, which is precisely SecTrustSerialize / NWIOConnection. You can see how deep it reaches into RN internals in node_modules/@fullstory/react-native/ios/FullStory.mm, where FSReactSwizzleBootstrap swizzles RCTComponentData viewConfig, every RCTViewComponentView subclass's handleCommand:args:, and RNNComponentViewController renderReactViewIfNeeded.
We already have a use-after-free from this exact SDK in HybridApp: #91225, patched in this same file. That patch (patches/@fullstory/react-native/@fullstory+react-native+1.9.0.patch) added an invalidate that clears _eventEmitterCallback when the NewDot bridge tears down on the ND to OD switch. But it only clears the callback. It never stops FullStory's capture engine. After the bridge invalidates, the FS singleton is still registered as the SDK delegate (FS.delegate = self in init) and keeps walking now-freed RN view objects and firing background upload callbacks against a half torn-down runtime. That is the corruption source, and the malloc trap lands wherever the next allocation happens, accessibility or TLS. It reads as iOS-26-only and HybridApp-only because iOS 26 tightened the malloc zone integrity checks, so a pre-existing corruption that used to slide by now traps, and only HybridApp tears the RN bridge down while a FullStory session is live.
I can't point at the exact offending line because the corruptor is inside FullStory's closed FullStory xcframework, pinned at 1.70.1 in ios/Podfile:133. The app-side surface we do control is when the SDK runs relative to bridge teardown, and which SDK build we ship.
What changes do you think we should make in order to solve the problem?
Extend the existing FullStory patch so the SDK actually stops capturing when the bridge tears down, instead of only nulling the callback. shutdown is already exported by the module (RCT_EXPORT_METHOD(shutdown) calling [FS shutdown]) but nothing in JS ever calls it. Add it to invalidate in patches/@fullstory/react-native/@fullstory+react-native+1.9.0.patch:
- (void)invalidate {
// Stop capture before the bridge tears down so FullStory's screen-capture and// background upload threads stop touching freed RN views (APP-93Z, #91225).
[FSshutdown];
_eventEmitterCallback = nullptr;
}
This is safe because capture already gets re-armed on the way back. consentAndIdentify in src/libs/Fullstory/index.native.ts:44 calls FullStory.restart(), and that fires whenever UserMetadata updates, wired at src/libs/actions/Session/index.ts:114. So NewDot coming back turns FullStory on again exactly like it does today.
Then bump the pinned FullStory native SDK past 1.70.1 at ios/Podfile:133 and re-run pod install to refresh Podfile.lock. 1.70.1 predates iOS 26's stricter malloc zones, and FullStory ships iOS 26 memory fixes in newer builds. The teardown-stop closes the HybridApp race; the SDK bump covers the case where 1.70.1 corrupts the heap during normal capture on iOS 26 even without a teardown. I'd open a FullStory support ticket with the Sentry stack so they can confirm the fixed version before we pin it.
What alternative solutions did you explore? (Optional)
I looked at the "it's just an Apple accessibility regression, collapse the SelectionList rows into one accessibility element" route, but that can't explain the events that crash in the TLS path, and you don't fix a malloc freelist trap by shrinking the accessibility node count. I also checked whether another native SDK (the Sentry profiler, group-ib) could be the corruptor, but FullStory is the only one that touches both the accessibility tree and background TLS uploads, and it already has a documented HybridApp UAF right here. A Sentry inbound filter for this signature is worth adding to cut dashboard noise while the real fix ships, but it has to go on the native Sentry init (ios/SentryNativeSDKManager.m) or Sentry's server-side inbound filters, not the JS beforeSend in setupSentry.ts, since this is a native mach crash the JS layer never sees. That only hides it though, so it's a companion, not the fix.
@bernhardoj this is a bit of an edge case, I sent wtih 25% mainly cuz I don't think we should pay out 1.5x or higher on issues where C+ should be able to see them through, from start to finish (or ask to reassign and not be due compensation).
@ZhenjaHorbach plz complete the Contributor+ checklist and tag me in a post once you have. Thx
@mallenexpensify
I think the checklist isn't needed here
This is an iOS 26 OS-level crash, not a regression from a PR, and there are no repro steps for a regression test
And the fix is verified via Sentry
Activity
melvin-bot commentedon Jul 20, 2026
Triggered auto assignment to Contributor-plus team member for initial proposal review - @bernhardoj (
External)samranahm commentedon Jul 20, 2026
Proposal
What is the root cause of that problem?
The crash frame
CACElementNumberStabilizer mergeListsbelongs to Apple's private Core Accessibility framework, not to first-party JavaScript. On iOS 26, when VoiceOver/accessibility recomputes element ordering it callsmergeLists, and the sibling frames (_xzm_xzone_malloc_freelist_outlined,SecTrustSerialize,NWIOConnectionTLS trust evaluation) show this runs concurrently with a background TLS certificate serialization. TheEXC_BREAKPOINTis a malloc freelist integrity trap that fires when the accessibility element list is being mutated on one thread while the OS stabilizer merges it on another. So the user-visible consequence (a hard crash on iOS 26 HybridApp, versions 9.4.25 through 9.4.31) is triggered inside Apple code that the app cannot call or patch directly.The app-side factor that raises exposure is how often we rebuild large accessibility element trees. Heavy
SelectionListscreens render many accessibility nodes per row.UserSelectionListItemre-derives display data and passeskeyForList={item.keyForList}on every render:App/src/components/SelectionList/ListItem/UserSelectionListItem.tsx
Lines 25 to 104 in d24f5d7
When
keyForListor item identity changes frequently, rows remount and the native accessibility element list is torn down and merged again. The more often that merge runs while background network/TLS activity is live, the more often iOS 26's stabilizer hits the unstable freelist and traps.What changes do you think we should make in order to solve the problem?
Confirm the scope first: this is an iOS 26 OS regression, so the crash cannot be fully eliminated from JS. The realistic goal is to reduce how often the native accessibility element tree is rebuilt during list rendering, which reduces how often the Apple stabilizer runs and can trap.
Stabilize the accessibility element count coming from
SelectionListrows. InUserSelectionListItem.tsx, the memoizeduserHandleanduserDisplayNameare already stable, but the row still exposes each icon and control as its own accessibility element. Group the row into a single accessibility element so the stabilizer merges fewer nodes:App/src/components/SelectionList/ListItem/UserSelectionListItem.tsx
Lines 25 to 104 in d24f5d7
Collapsing each row to one accessibility element shrinks the list the OS must merge, which lowers the chance of tripping
mergeLists.beforeSendentry for theCACElementNumberStabilizer mergeListssignature so this OS-level noise stops dominating the dashboard while the mitigation ships and iOS point releases land. This uses the existing Sentry wiring referenced by the stub insentry-react-native.ts:App/server/stubs/sentry-react-native.ts
Lines 20 to 21 in d24f5d7
What alternative solutions did you explore? (Optional)
Reminder: Please use plain English, be brief and avoid jargon. Feel free to use images, charts or pseudo-code if necessary. Do not post large multi-line diffs or write walls of text. Do not create PRs unless you have been hired for this job.
melvin-bot commentedon Jul 20, 2026
Job added to Upwork: https://www.upwork.com/jobs/~022079255410690029390
[-][Sentry: APP-93Z] HybridApp iOS EXC_BREAKPOINT in CACElementNumberStabilizer mergeLists[/-][+][$250] [Sentry: APP-93Z] HybridApp iOS EXC_BREAKPOINT in CACElementNumberStabilizer mergeLists[/+]yuvrajangadsingh commentedon Jul 20, 2026
Proposal
Please re-state the problem that we are trying to solve in this issue.
iOS 26 HybridApp is hard-crashing with an
EXC_BREAKPOINTwhose top frames flip between Apple's accessibility element merge (CACElementNumberStabilizer mergeLists) and a background TLS trust evaluation (SecTrustSerialize/NWIOConnection), with_xzm_xzone_malloc_freelist_outlinedsitting in the middle. 186 users, app versions 9.4.25 to 9.4.31, no known repro.What is the root cause of that problem?
The only frame that names the actual failure is
_xzm_xzone_malloc_freelist_outlined. That is iOS's malloc catching a corrupted zone freelist and trapping withEXC_BREAKPOINT. So this is a heap-corruption crash, not an accessibility logic bug. The bad write happens earlier, and the trap fires on the next allocation that touches the poisoned zone. That is why the "cause" frame changes between events: accessibility element merging in some, TLS certificate serialization in others. Whoever corrupts the heap has to live in both of those worlds.The one thing in the app that instruments both the view/accessibility tree and background TLS network activity is FullStory. Its session-replay engine walks the native view hierarchy including accessibility elements to build each replay frame, and it continuously uploads captured session bundles over HTTPS on background threads, which is precisely
SecTrustSerialize/NWIOConnection. You can see how deep it reaches into RN internals innode_modules/@fullstory/react-native/ios/FullStory.mm, whereFSReactSwizzleBootstrapswizzlesRCTComponentData viewConfig, everyRCTViewComponentViewsubclass'shandleCommand:args:, andRNNComponentViewController renderReactViewIfNeeded.We already have a use-after-free from this exact SDK in HybridApp: #91225, patched in this same file. That patch (
patches/@fullstory/react-native/@fullstory+react-native+1.9.0.patch) added aninvalidatethat clears_eventEmitterCallbackwhen the NewDot bridge tears down on the ND to OD switch. But it only clears the callback. It never stops FullStory's capture engine. After the bridge invalidates, theFSsingleton is still registered as the SDK delegate (FS.delegate = selfininit) and keeps walking now-freed RN view objects and firing background upload callbacks against a half torn-down runtime. That is the corruption source, and the malloc trap lands wherever the next allocation happens, accessibility or TLS. It reads as iOS-26-only and HybridApp-only because iOS 26 tightened the malloc zone integrity checks, so a pre-existing corruption that used to slide by now traps, and only HybridApp tears the RN bridge down while a FullStory session is live.I can't point at the exact offending line because the corruptor is inside FullStory's closed
FullStoryxcframework, pinned at1.70.1inios/Podfile:133. The app-side surface we do control is when the SDK runs relative to bridge teardown, and which SDK build we ship.What changes do you think we should make in order to solve the problem?
Two things, mirroring how #91225 was handled.
Extend the existing FullStory patch so the SDK actually stops capturing when the bridge tears down, instead of only nulling the callback.
shutdownis already exported by the module (RCT_EXPORT_METHOD(shutdown)calling[FS shutdown]) but nothing in JS ever calls it. Add it toinvalidateinpatches/@fullstory/react-native/@fullstory+react-native+1.9.0.patch:This is safe because capture already gets re-armed on the way back.
consentAndIdentifyinsrc/libs/Fullstory/index.native.ts:44callsFullStory.restart(), and that fires wheneverUserMetadataupdates, wired atsrc/libs/actions/Session/index.ts:114. So NewDot coming back turns FullStory on again exactly like it does today.Then bump the pinned FullStory native SDK past
1.70.1atios/Podfile:133and re-run pod install to refreshPodfile.lock. 1.70.1 predates iOS 26's stricter malloc zones, and FullStory ships iOS 26 memory fixes in newer builds. The teardown-stop closes the HybridApp race; the SDK bump covers the case where 1.70.1 corrupts the heap during normal capture on iOS 26 even without a teardown. I'd open a FullStory support ticket with the Sentry stack so they can confirm the fixed version before we pin it.What alternative solutions did you explore? (Optional)
I looked at the "it's just an Apple accessibility regression, collapse the SelectionList rows into one accessibility element" route, but that can't explain the events that crash in the TLS path, and you don't fix a malloc freelist trap by shrinking the accessibility node count. I also checked whether another native SDK (the Sentry profiler, group-ib) could be the corruptor, but FullStory is the only one that touches both the accessibility tree and background TLS uploads, and it already has a documented HybridApp UAF right here. A Sentry inbound filter for this signature is worth adding to cut dashboard noise while the real fix ships, but it has to go on the native Sentry init (
ios/SentryNativeSDKManager.m) or Sentry's server-side inbound filters, not the JSbeforeSendinsetupSentry.ts, since this is a native mach crash the JS layer never sees. That only hides it though, so it's a companion, not the fix.bernhardoj commentedon Jul 21, 2026
@MelvinBot please investigate this issue. Can you check if this is related to an iOS 26 native Network module regression?
melvin-bot commentedon Jul 21, 2026
I can't run change requests for your access level. I can investigate, or file an issue in
Expensify/Expensifyinstead.Failed after 0s · 0 tools used · view debug log
63 remaining items
ZhenjaHorbach commentedon Sep 22, 2026
I reviewed one PR!
bernhardoj commentedon Sep 22, 2026
I only analyze the root cause but don't review the PR since I don't have access to physical iOS.
melvin-bot commentedon Sep 22, 2026
Payment Summary
BugZero Checklist (@mallenexpensify)
mallenexpensify commentedon Sep 24, 2026
@bernhardoj @ZhenjaHorbach can you provide the PR link/s and tag me? Thx
melvin-bot commentedon Oct 1, 2026
@mallenexpensify Uh oh! This issue is overdue by 2 days. Don't forget to update your issues!
mallenexpensify commentedon Oct 3, 2026
Payment Summary
Contributor+: @ZhenjaHorbach due $250 via NewDot
Contributor+: @bernhardoj due $75 via NewDot
@bernhardoj this is a bit of an edge case, I sent wtih 25% mainly cuz I don't think we should pay out 1.5x or higher on issues where C+ should be able to see them through, from start to finish (or ask to reassign and not be due compensation).
@ZhenjaHorbach plz complete the Contributor+ checklist and tag me in a post once you have. Thx
ZhenjaHorbach commentedon Oct 3, 2026
Oh
Sorry
I’ll add the checklist today or tomorrow!
ZhenjaHorbach commentedon Oct 5, 2026
@mallenexpensify
I think the checklist isn't needed here
This is an iOS 26 OS-level crash, not a regression from a PR, and there are no repro steps for a regression test
And the fix is verified via Sentry
garrettmknight commentedon Nov 5, 2026
$250 approved for @ZhenjaHorbach
flaviadefaria commentedon Nov 9, 2026
Approved: @bernhardoj due $75 via NewDot