test_perf_profiler: test_trampoline_works_with_forks() failed on Address Sanitizer CI

[vstinner]

Bug report

test_trampoline_works_with_forks (test.test_perf_profiler.TestPerfTrampoline.test_trampoline_works_with_forks) ... FAIL

FAIL: test_trampoline_works_with_forks (test.test_perf_profiler.TestPerfTrampoline.test_trampoline_works_with_forks)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/home/runner/work/cpython/cpython/Lib/test/test_perf_profiler.py", line 123, in test_trampoline_works_with_forks
    self.assertEqual(process.returncode, 0)
AssertionError: 139 != 0

It looks like a SIGSEGV crash (128 + SIGSEGV = 139 on Linux).

build: https://github.com/python/cpython/actions/runs/6237694188/job/16931875912?pr=108965

Passed when re-run:

0:09:17 load avg: 4.36 Re-running 1 failed tests in verbose mode in subprocesses
0:09:17 load avg: 4.36 Run 1 test in parallel using 1 worker process (timeout: 20 min, worker timeout: 25 min)
0:09:17 load avg: 4.36 [1/1] test_perf_profiler passed
Re-running test_perf_profiler in verbose mode (matching: test_trampoline_works_with_forks)
test_trampoline_works_with_forks (test.test_perf_profiler.TestPerfTrampoline.test_trampoline_works_with_forks) ... ok

----------------------------------------------------------------------
Ran 1 test in 0.083s

OK

cc @pablogsal

Linked PRs

• gh-109580: Skip test_perf_profiler on ASAN build #109584
• [3.12] gh-109580: Skip test_perf_profiler on ASAN build (GH-109584) #109585

[vstinner]

I cannot reproduce the issue locally on Fedora 38.

One option is to skip the test if Python is built with ASAN, as already done on TestPerfProfiler test case: skip the whole file.

[pablogsal]

I cannot reproduce the issue locally on Fedora 38.

One option is to skip the test if Python is built with ASAN, as already done on TestPerfProfiler test case: skip the whole file.

I'm fine skipping the file if is built with ASAN.

[vstinner]

@pablogsal:

I'm fine skipping the file if is built with ASAN.

Ok, I created PR #109584 for that.

I don't have the bandwidth to investigate why Python built with ASAN does crash on this Ubuntu Address Sanitizer job, but it doesn't when I run the test locally on my Fedora 38.