Exit tests hang indefinitely on Ubuntu 20.04

[jmschonfeld]

Description

When adding an exit test to a test suite, that test hangs indefinitely when run on Ubuntu 20.04. The test runs successfully and completes in ~0.3s on Ubuntu 22.04 and on all of the macOS versions that I've tried.

Reproduction

1. Download the latest nightly-focal swift docker image
2. Create a project that includes the following unit test

@Test func testExits() {
    await #expect(processExitsWith: .failure) {
        fatalError("I crashed")
    }
}

3. Build and run the test (swift test)
4. Notice that the test hangs indefinitely after ◇ Test testExits() started.

Expected behavior

Exit tests complete successfully on all supported platforms

Environment

↳ Testing Library Version: 6.3-dev (d6111205ce07f17)
↳ Target Platform: aarch64-unknown-linux-gnu

Swift version 6.2-dev (LLVM 5e7837820b1c886, Swift d5ef256b402bff7)
Target: aarch64-unknown-linux-gnu
Build config: +assertions

Linux nightly-focal 6.1.68 #1 SMP Fri Nov 22 18:05:17 UTC 2024 aarch64 aarch64 aarch64 GNU/Linux

Additional information

This is blocking Foundation from adopting exit tests as swift toolchain CI runs on Ubuntu 20.04

[jmschonfeld]

Here is a sample of the parent and child process while it is hanging:

root@nightly-focal:/Repos# ps -ef | grep swift-foundation
root      4895  4846  0 20:43 pts/0    00:00:00 /Repos/Public/swift-foundation/.build/aarch64-unknown-linux-gnu/debug/swift-foundationPackageTests.xctest --filter testExits --testing-library swift-testing
root      4904  4895  0 20:43 pts/0    00:00:00 /Repos/Public/swift-foundation/.build/aarch64-unknown-linux-gnu/debug/swift-foundationPackageTests.xctest --testing-library swift-testing
root      4915  4032  0 20:43 pts/1    00:00:00 grep --color=auto swift-foundation
root@nightly-focal:/Repos# eu-stack -id -p 4895          
PID 4895 - process
TID 4895:
#0  0x0000ffff8f7f916c __GI___sigsuspend
#1  0x0000ffff90519558 _dispatch_sigsuspend
#2  0x0000ffff90516e70 _dispatch_sig_thread
#3  0x0000ffff8fb5c3c0 __nptl_deallocate_tsd
#4  0x0000ffff8f7e5e54 __libc_start_main
#5  0x0000aaaae46d6d50 $x
#6  0x0000aaaae46d6d50 $x
eu-stack: dwfl_thread_getframes tid 4895 at 0xaaaae46d6d4f in /Repos/Public/swift-foundation/.build/aarch64-unknown-linux-gnu/debug/swift-foundationPackageTests.xctest: no matching address range
TID 4899:
#0  0x0000ffff8f888dc0 __GI___libc_read
#1  0x0000ffff8f888dc0 __GI___libc_read
#2  0x0000ffff8f83618c _IO_new_file_underflow
#3  0x0000ffff8f835298 __GI__IO_file_xsgetn
#4  0x0000ffff8f8292a0 __GI__IO_fread
#5  0x0000ffff8fd33c10 $s7Testing10FileHandleV9readToEndSays5UInt8VGyKF
#6  0x0000ffff8fcbf74c $s7Testing8ExitTestV15_processRecords33_95876674CE05CFF8D0FD5F064E685A97LL15fromBackChannelyAA10FileHandleV_tFZTf4nd_n
#7  0x0000ffff8fcbd2ac $s7Testing8ExitTestV20handlerForEntryPointAC6ResultVACYaYbKcyFZ6resultL_yAfCYaYbKFAFScgyyAFzYbcSgs5Error_pGzYaKXEfU0_AHyYacfU6_TY0_
#8  0x0000ffff90f69fe8 swift::runJobInEstablishedExecutorContext(swift::Job*)
#9  0x0000ffff90f6ad38 swift_job_run
#10 0x0000ffff9050fc70 _dispatch_continuation_pop
#11 0x0000ffff9050faac _dispatch_async_redirect_invoke
#12 0x0000ffff9051930c _dispatch_worker_thread
#13 0x0000ffff8fb5c624 start_thread
#14 0x0000ffff8f89666c thread_start
TID 4900:
#0  0x0000ffff8f896794 __GI_epoll_pwait
#1  0x0000ffff90520bd4 _dispatch_event_loop_drain
#2  0x0000ffff9051602c _dispatch_mgr_invoke
#3  0x0000ffff90515fa0 _dispatch_mgr_thread
#4  0x0000ffff9051930c _dispatch_worker_thread
#5  0x0000ffff8fb5c624 start_thread
#6  0x0000ffff8f89666c thread_start
TID 4905:
#0  0x0000ffff8f868d28 __waitid
#1  0x0000ffff8fcc1f2c $s7Testing17_createWaitThread33_F6EF36D824EB298AB893AA9AF261F249LLytvpfiyyXEfU_15waitForAnyChildL_yyF
#2  0x0000ffff8fcc2214 $s7Testing17_createWaitThread33_F6EF36D824EB298AB893AA9AF261F249LLytvpfiyyXEfU_SvSgADcfU_To
#3  0x0000ffff8fb5c624 start_thread
#4  0x0000ffff8f89666c thread_start
root@nightly-focal:/Repos# eu-stack -id -p 4904
PID 4904 - process
TID 4904:
#0  0x0000ffffb5d6016c __GI___sigsuspend
#1  0x0000ffffb6a73558 _dispatch_sigsuspend
#2  0x0000ffffb6a70e70 _dispatch_sig_thread
#3  0x0000ffffb60c33c0 __nptl_deallocate_tsd
#4  0x0000ffffb5d4ce54 __libc_start_main
#5  0x0000aaaad2cd6d50 $x
#6  0x0000aaaad2cd6d50 $x
eu-stack: dwfl_thread_getframes tid 4904 at 0xaaaad2cd6d4f in /Repos/Public/swift-foundation/.build/aarch64-unknown-linux-gnu/debug/swift-foundationPackageTests.xctest: no matching address range
TID 4907:
#0  0x0000ffffb5defdc0 __GI___libc_read
#1  0x0000ffffb5defdc0 __GI___libc_read
#2  0x0000ffffb5d9d18c _IO_new_file_underflow
#3  0x0000ffffb5d9c298 __GI__IO_file_xsgetn
#4  0x0000ffffb5d902a0 __GI__IO_fread
#5  0x0000ffffb692cc10 $s7Testing10FileHandleV9readToEndSays5UInt8VGyKF
#6  0x0000ffffb68b3f78 $s7Testing8ExitTestV34_decodeCapturedValuesForEntryPoint33_95876674CE05CFF8D0FD5F064E685A97LLyyKF
#7  0x0000ffffb68b3e74 $s7Testing8ExitTestV30findInEnvironmentForEntryPointACSgyFZyACzYaYbKcfU2_yyYaKXEfU_TY0_
#8  0x0000ffffb6dc3fe8 swift::runJobInEstablishedExecutorContext(swift::Job*)
#9  0x0000ffffb6dc4d38 swift_job_run
#10 0x0000ffffb6a69c70 _dispatch_continuation_pop
#11 0x0000ffffb6a69aac _dispatch_async_redirect_invoke
#12 0x0000ffffb6a7330c _dispatch_worker_thread
#13 0x0000ffffb60c3624 start_thread
#14 0x0000ffffb5dfd66c thread_start
TID 4908:
#0  0x0000ffffb5dfd794 __GI_epoll_pwait
#1  0x0000ffffb6a7abd4 _dispatch_event_loop_drain
#2  0x0000ffffb6a7002c _dispatch_mgr_invoke
#3  0x0000ffffb6a6ffa0 _dispatch_mgr_thread
#4  0x0000ffffb6a7330c _dispatch_worker_thread
#5  0x0000ffffb60c3624 start_thread
#6  0x0000ffffb5dfd66c thread_start

[jmschonfeld]

Also attaching the output of strace -f /Repos/Public/swift-foundation/.build/aarch64-unknown-linux-gnu/debug/swift-foundationPackageTests.xctest --filter testExits --testing-library swift-testing. This is all of the output until the end where it just starts repeating the epoll_pwait calls repeatedly/indefinitely

strace.txt

[jmschonfeld]

Also confirmed that this behaves as expected on Ubuntu 24.04

[grynspan]

We create the two pipes here:

[pid  5005] pipe2([6, 7], 0)            = 0
[pid  5005] fcntl(6, F_GETFL)           = 0 (flags O_RDONLY)
[pid  5005] fcntl(7, F_GETFL)           = 0x1 (flags O_WRONLY)
[pid  5005] pipe2([8, 9], 0)            = 0
[pid  5005] fcntl(8, F_GETFL)           = 0 (flags O_RDONLY)
[pid  5005] fcntl(9, F_GETFL)           = 0x1 (flags O_WRONLY)

8/9 will be used as the pipe from parent to child.

Parent process cloned here:

[pid  5005] clone(child_stack=0xffffa0ca7000, flags=CLONE_VM|CLONE_VFORK|SIGCHLDstrace: Process 5010 attached
 <unfinished ...>

Child process exec'ed here:

[pid  5010] execve("/Repos/Public/swift-foundation/.build/aarch64-unknown-linux-gnu/debug/swift-foundationPackageTests.xctest", ["/Repos/Public/swift-foundation/."..., "--testing-library", "swift-testing"], 0xffff9000a580 /* 13 vars */ <unfinished ...>

The parent process closed FDs 8/9 here:

[pid  5005] close(8 <unfinished ...>
[pid  5010] <... readlinkat resumed>"/Repos/Public/swift-foundation/."..., 4096) = 105
[pid  5005] <... close resumed>)        = 0
[pid  5010] mmap(NULL, 8192, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0 <unfinished ...>
[pid  5005] close(9 <unfinished ...>
[pid  5010] <... mmap resumed>)         = 0xffffb4db5000
[pid  5005] <... close resumed>)        = 0

Since this test doesn't capture any state (experimental feature), there are no writes to FD 9. PID 5010 creates various child threads, one of which is "PID" 5013 on which the exit test starts:

[pid  5013] fcntl(8, F_GETFL)           = 0 (flags O_RDONLY)
[pid  5013] fstat(8, {st_mode=S_IFIFO|0600, st_size=0, ...}) = 0
[pid  5013] fstat(8, {st_mode=S_IFIFO|0600, st_size=0, ...}) = 0
[pid  5013] read(8,  <unfinished ...>

That call hangs, indicating that the OS perhaps doesn't know that the other end of the pipe was closed?