mirror of
https://github.com/firebase/firebase-ios-sdk
synced 2026-11-09 22:36:01 +05:45
Konstantin Varlamov
f1294cee91
If a user calls terminate and immediately disposes their reference to Firestore, it's possible that while termination is in process, the last remaining reference (more precisely, shared pointer) to Firestore is in a listener. When that listener is destroyed as part of the termination, it leads to `api::Firestore` being destroyed. Importantly, termination happens on the worker queue. The destructor of `api::Firestore` calls `Dispose` which presumes it's not called from the worker queue and tries to enqueue work, leading to a failing assertion and a crash. The solution is simply to remove the sequential order checks when the queue enters/is in restricted mode. There is a legitimate case where the Firestore destructor can run on the worker queue, as the original issue shows. The complexity of #7412 also indicates that a simpler solution is preferable. Fixes #6909.