bpo-18174: regrtest -R 3:3 checks for handle leak

[vstinner]

• Add _winapi.GetProcessHandleCount()
• regrtest now also checks for leak of Windows handles
• Add an unit test in test_regrtest

Co-Authored-By: Richard Oudkerk [email protected]

https://bugs.python.org/issue18174

[vstinner]

I'm running tests with "python -m test -j6 -R 2:3 -r" and it seems like some tests are failing randomly :-/

Example: "test_cmd_line leaked [1, 0, 0] handles, sum=1". Maybe regrtest is too strict.

[vstinner]

I ran "python -m test -j6 -R 2:3 -r" with the second commit "Make regrtest less strict". 10 tests failed:

• test__xxsubinterpreters: bug, "unknown interpreter ID"
• test_asyncio: memory block leak
• test_builtin: I failed to reproduce this failure
• test_bz2: leak of 100 handles, maybe the same bug than https://bugs.python.org/issue33916
• test_compileall: PermissionError
• test_concurrent_futures: I'm trying to reproduce that one, but I failed to reproduce the leak yet.
• test_lzma: https://bugs.python.org/issue33916
• test_multiprocessing_spawn: handle leak
• test_ntpath: if I recall correctly, it was a bug, not a leak
• test_script_helper: bug

Tests not always failed because of leaks, but I didn't have time yet to re-run these tests. I will do it later.

[vstinner]

test__xxsubinterpreters: bug, "unknown interpreter ID"

Unrelated to this PR: the test is fragile, see for example https://bugs.python.org/issue33868

test_asyncio: memory block leak

I'm investigating this issue, but there is already one open issue: https://bugs.python.org/issue32710

test_builtin: I failed to reproduce this failure

Unrelated to this PR, and again, I failed to reproduce the issue. Maybe a race condition.

test_bz2: leak of 100 handles, maybe the same bug than https://bugs.python.org/issue33916
test_lzma: https://bugs.python.org/issue33916

Fixed by my PR #7843 (not merged yet)

test_compileall: PermissionError

Unrelated to this PR, and again, I failed to reproduce the issue. Maybe a race condition.

test_concurrent_futures: I'm trying to reproduce that one, but I failed to reproduce the leak yet.

I confirm that I fail to reproduce the leak. Maybe a race condition.

test_multiprocessing_spawn: handle leak

I reproduced the leak. I'm running test.bisect on it.

test_ntpath: if I recall correctly, it was a bug, not a leak

Unrelated to this PR, and again, I failed to reproduce the issue. Maybe a race condition.

test_script_helper: bug

Oh. Running the test alone without -R 3:3 also fails.

[vstinner]

test_asyncio: memory block leak
I'm investigating this issue, but there is already one open issue: https://bugs.python.org/issue32710

Oh, I confirm that it's exactly the same bug: test.test_asyncio.test_events.ProactorEventLoopTests.test_sendfile_close_peer_in_middle_of_receiving leaks memory blocks (not handles!).

[vstinner]

test_multiprocessing_spawn: handle leak
I reproduced the leak. I'm running test.bisect on it.

I found a bug: https://bugs.python.org/issue33929

[vstinner]

test_script_helper: bug
Oh. Running the test alone without -R 3:3 also fails.

Ok, it's also a known issue: https://bugs.python.org/issue32942

[vstinner]

Oh, test_multiprocessing_spawn is still not good, even with my PR 7921:
https://bugs.python.org/issue33966
"test_multiprocessing_spawn.WithProcessesTestPool.test_traceback() leaks 4 handles on Windows"

[vstinner]

Oh, test_multiprocessing_spawn is still not good, even with my PR 7921: https://bugs.python.org/issue33966

I wrote a fix: PR 7966. Using this fix, test_multiprocessing_spawn now doesn't leak handles anymore!

[vstinner]

test_compileall: PermissionError

I created https://bugs.python.org/issue33977