Bug report

Bug description:

test_sampling_with_collapsed_export failed in the ASan job of my PR gh-160296 (the change there is in a multiprocessing test and is unrelated):

FAIL: test_sampling_with_collapsed_export (test.test_profiling.test_sampling_profiler.test_integration.TestSampleProfilerIntegration.test_sampling_with_collapsed_export)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/home/runner/work/cpython/cpython/Lib/test/test_profiling/test_sampling_profiler/test_integration.py", line 512, in test_sampling_with_collapsed_export
    self.assertIn(":", part)
    ~~~~~~~~~~~~~^^^^^^^^^^^
AssertionError: ':' not found in '<unreadable frame>'

https://github.com/python/cpython/actions/runs/39832455069/job/120012213367

Since gh-154194, _remote_debugging returns a degraded frame (file "~", no location, function "<unreadable frame>") when it cannot read a code object, instead of failing the sample. CollapsedStackCollector turns the None location into line 0, and export() writes frames with file "~" and line 0 as just the function name, so the collapsed line contains <unreadable frame> with no colon. The test expects every part to look like file:func:line, so it fails whenever a sample hits an unreadable frame. I have only seen it on the ASan build so far.

gh-154195 updated test_binary_format.py for the new frame, but not this test.

Reproducer, without a target process:

import os
import tempfile
from profiling.sampling.stack_collector import CollapsedStackCollector

collector = CollapsedStackCollector(1000)
# The first (innermost) frame is what _remote_debugging returns
# when it cannot read a code object (gh-154194).
frames = [
    ("~", None, "<unreadable frame>", None),
    ("/tmp/script.py", (10, 10, 0, 1), "main", None),
]
collector.process_frames(frames, thread_id=1)

fd, path = tempfile.mkstemp(suffix=".txt")
os.close(fd)
collector.export(path)
with open(path) as f:
    line = f.read().strip()
os.unlink(path)

print(line)
stack_trace = line.rsplit(" ", 1)[0]
for part in stack_trace.split(";"):
    print(repr(part), ":" in part)

Output:

Collapsed stack output written to /tmp/tmp....txt
tid:1;script.py:main:10;<unreadable frame> 1
'tid:1' True
'script.py:main:10' True
'<unreadable frame>' False

<GC> frames (on by default in sample()) take the same "~" branch in export(), so they would fail this assertion too. The test could accept these special <...> frames instead of requiring a colon. I can open a PR if that approach is fine.

The same test and the gh-154194 change are also in 3.15.

CPython versions tested on:

CPython main branch

Operating systems tested on:

Linux, macOS