Crash on HTTPBodyStreamDelegate

[victor-osc]

Description

We are having a crash on HTTPBodyStreamDelegate

The stack trace seems odd which may be happening because of the inline optimizations or some race condition.

• OpenAPIURLSession.BufferedStream._StateMachine._State.Initial.init(backPressureStrategy: OpenAPIURLSession.BufferedStream._InternalBackPressureStrategy, iteratorInitialized: Swift.Bool, onTermination: (@sendable () -> ())?) -> OpenAPIURLSession.BufferedStream._StateMachine._State.Initial
• OpenAPIURLSession.BufferedStream._StateMachine._State.SourceFinished.failure.setter : Swift.Error?
• OpenAPIURLSession.BufferedStream.Source.write<A where A == A1.Element, A1: Swift.Sequence>(contentsOf: A1) async throws -> ()
• OpenAPIURLSession.BufferedStream.Source.WriteResult.CallbackToken.id.getter : Swift.UInt
• OpenAPIURLSession.LockStorage.unlock() -> ()

Reproduction

We don't know yet the exact steps to reproduce the issue

Package version(s)

https://github.com/apple/swift-openapi-urlsession
version 1.0.2

Expected behavior

not crash

Environment

iOS 18.5

Additional information

No response

[czechboy0]

Transfered over to the right repo. cc @guoye-zhang @simonjbeaumont

[guoye-zhang]

Would you attach a complete crash report?

[ls-alexandre-perez]

We are experiencing a similar crash on our end. Attached is the stack trace of one of the crashes. I don't have a way to reproduce it, but it might be related to the cancellation of the task that the upload is part of.

SIGTRAP

Crashed: HTTPBodyStreamDelegate

0  App                            0x1981afc HTTPBodyOutputStreamBridge.State.wroteFinalChunk() + 210 (HTTPBodyOutputStreamBridge.swift:210)
1  App                            0x1981990 closure #2 in closure #1 in HTTPBodyOutputStreamBridge.startWriterTask() + 75 (HTTPBodyOutputStreamBridge.swift:75)
2  App                            0x607790 <deduplicated_symbol> + 4383307664
3  libdispatch.dylib              0x5b830 <redacted> + 24
4  libdispatch.dylib              0x5ca68 <redacted> + 12
5  libdispatch.dylib              0x9a04 <redacted> + 668
6  libdispatch.dylib              0xa440 <redacted> + 384
7  libdispatch.dylib              0x134d0 <redacted> + 256
8  libdispatch.dylib              0x12dec <redacted> + 528
9  libsystem_pthread.dylib        0xa10 _pthread_wqthread + 288
10 libsystem_pthread.dylib        0xab0 start_wqthread + 8

[aluco100]

Hey there,

I want to contribute here, could you tell me the steps to reproduce this crash?

Best

[madsodgaard]

We are also seeing this crash. @fortmarek I saw you closed your PR for a fix, does it work for you? If so, will you upstream it?

Stacktrace

* thread #37, queue = 'HTTPBodyStreamDelegate', stop reason = EXC_BREAKPOINT (code=1, subcode=0x197309f0c)
  * frame #0: 0x0000000197309f0c libswiftCore.dylib`_assertionFailure(_:_:file:line:flags:) + 156
    frame #1: 0x0000000114d5426c App.debug.dylib`HTTPBodyOutputStreamBridge.State.wroteFinalChunk() at HTTPBodyOutputStreamBridge.swift:211:17
    frame #2: 0x0000000114d53ddc App.debug.dylib`closure #2 in closure #1 in HTTPBodyOutputStreamBridge.startWriterTask() at HTTPBodyOutputStreamBridge.swift:75:47
    frame #4: 0x00000001801c07a8 libdispatch.dylib`_dispatch_call_block_and_release + 24
    frame #5: 0x00000001801db4b0 libdispatch.dylib`_dispatch_client_callout + 12
    frame #6: 0x00000001801c9c28 libdispatch.dylib`_dispatch_lane_serial_drain + 984
    frame #7: 0x00000001801ca6e8 libdispatch.dylib`_dispatch_lane_invoke + 396
    frame #8: 0x00000001801d5534 libdispatch.dylib`_dispatch_root_queue_drain_deferred_wlh + 288
    frame #9: 0x00000001801d4c74 libdispatch.dylib`_dispatch_workloop_worker_thread + 692
    frame #10: 0x000000010549ab8c libsystem_pthread.dylib`_pthread_wqthread + 288

[fortmarek]

I saw you closed your PR for a fix, does it work for you? If so, will you upstream it?

It didn't seem to have helped. What helped in our case was capping the concurrency of a concurrent map from unlimited to 100-1000. To be fair, capping the concurrency is possibly making the crash only less likely to occur, but we haven't seen this error since.

Unfortunately, without a reproducer, it's quite tricky to fix this one. If you take the PR I did for a spin and can confirm it at least fixes your instance of the crash, I'm happy to take it over the finish line.