Crash on com.apollographql.websocket on _inputStreamCallbackFunc

[aaronbarsky]

Summary

Crashlytics is reporting a very rare but consistent crash on the com.apollograph.websocket queue.

Crashed: com.apollographql.websocket
EXC_BAD_ACCESS KERN_INVALID_ADDRESS 0x0000000f9d3993a0
0 libobjc.A.dylib objc_retain + 16
1 libobjc.A.dylib objc_retain_x0 + 16
2 CoreFoundation _inputStreamCallbackFunc + 60
3 CoreFoundation _signalEventSync + 216
4 CoreFoundation ___signalEventQueue_block_invoke + 28
5 libdispatch.dylib _dispatch_call_block_and_release + 32

The app is a mix of 60% foreground and 40% background.

Version

1.9.2

Steps to reproduce the behavior

Unfortunately I only have crash logs

Logs

No response

Anything else?

My hunch is that it's the FoundationStream teardown sequence:

 func cleanup() {
    if let stream = inputStream {
      stream.delegate = nil
      CFReadStreamSetDispatchQueue(stream, nil)
      stream.close()
    }

stream.delegate is an unowned(unsafe) var.

There are several codepaths where cleanup is not called on com.apollographql.websocket. When this happens, delegate can be set to nil as the CFReadStream is attempting to invoke a callback on the deallocated delegate.

[calvincestari]

Hi @aaronbarsky - you mentioned this being consistent; do you know the conditions or sequence of events that leads to the crash. Being able to have a reproduction case would be immensely helpful to resolving this.

[aaronbarsky]

By consistent, I mean that the call stack is exactly the same in each case. It happens about 1/500 sessions.

The challenge is that the start of the crash is

CoreFoundation _inputStreamCallbackFunc

which makes it quite tough to write a unit test or minimally reproducible example.