bpo-1635741: test_embed cheks that Python does not leak #31555
Conversation
Lib/test/test_cmd_line.py
Outdated
| # because of bugs in C extensions. This test is only about checking | ||
| # the showrefcount feature. | ||
| self.assertRegex(err, br'^\[-?\d+ refs, \d+ blocks\]') | ||
| self.assertRegex(err, br'^\[\d+ refs, \d+ blocks\]') |
There was a problem hiding this comment.
FYI, from my macOS
test test_cmd_line failed -- Traceback (most recent call last):
File "/Users/user/oss/cpython/Lib/test/test_cmd_line.py", line 124, in test_showrefcount
self.assertRegex(err, br'^\[\d+ refs, \d+ blocks\]')
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
AssertionError: Regex didn't match: b'^\\[\\d+ refs, \\d+ blocks\\]' not found in b'[-1 refs, 0 blocks]\n'
There was a problem hiding this comment.
Oh crap. Ok, let's tolerate negative ref count for now. Checking that refs <= 0 and blocks == 0 is already a step forward!
There was a problem hiding this comment.
I didn't account for negative refcount on other platforms as I only saw output on Linux where it is zero.
There was a problem hiding this comment.
For the record,
https://github.com/python/cpython/runs/5322493186?check_suite_focus=true
Ubuntu is also failed with the same reason.
FAIL: test_showrefcount (test.test_cmd_line.CmdLineTest)
----------------------------------------------------------------------
Traceback (most recent call last):
File "/home/runner/work/cpython/cpython-ro-srcdir/Lib/test/test_cmd_line.py", line 124, in test_showrefcount
self.assertRegex(err, br'^\[\d+ refs, \d+ blocks\]')
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
AssertionError: Regex didn't match: b'^\\[\\d+ refs, \\d+ blocks\\]' not found in b'[-1 refs, 0 blocks]\n'
There was a problem hiding this comment.
The test itself looks good to me (I suggested to use a test.support helper, if you want fewer lines), but EmbeddingTestsMixin (where this test lives) does not run at all on out-of-tree-builds (at least on my MacBook):
$ ./python.exe -m test test_embed -m test_no_memleak -v
Raised RLIMIT_NOFILE: 256 -> 1024
== CPython 3.11.0a5+ (heads/test_no_memleak-dirty:e6ca72c9e4, Feb 24 2022, 20:21:11) [Clang 13.0.0 (clang-1300.0.29.30)]
== macOS-12.2.1-x86_64-i386-64bit little-endian
== cwd: /Users/erlendaasland/src/cpython-build/build/test_python_68548æ
== CPU count: 16
== encodings: locale=UTF-8, FS=utf-8
0:00:00 load avg: 1.87 Run tests sequentially
0:00:00 load avg: 1.87 [1/1] test_embed
test_no_memleak (test.test_embed.MiscTests) ... skipped "'/Users/erlendaasland/src/cpython-build/Programs/_testembed' doesn't exist"
EmbeddingTestsMixin is fenced out by its setUp function:
if exepath != expecteddir or not os.path.exists(exe):
self.skipTest("%r doesn't exist" % exe)For OOT builds, exepath != expecteddir resolves to True and not os.path.exists(exe) resolves to False.
Hm, looks like this is a deliberate choice. See f4b1244 / GH-29063.
| cmd = [sys.executable, "-I", "-X", "showrefcount", "-c", "pass"] | ||
| proc = subprocess.run(cmd, | ||
| stdout=subprocess.PIPE, | ||
| stderr=subprocess.STDOUT, | ||
| text=True) | ||
| self.assertEqual(proc.returncode, 0) | ||
| out = proc.stdout.rstrip() |
There was a problem hiding this comment.
Why not this?
| cmd = [sys.executable, "-I", "-X", "showrefcount", "-c", "pass"] | |
| proc = subprocess.run(cmd, | |
| stdout=subprocess.PIPE, | |
| stderr=subprocess.STDOUT, | |
| text=True) | |
| self.assertEqual(proc.returncode, 0) | |
| out = proc.stdout.rstrip() | |
| rc, _, out = assert_python_ok("-I", "-X", "showrefcount", "-c", "pass") | |
| self.assertEqual(rc, 0) | |
| out = out.decode() |
(You'd need from test.support.script_helper import assert_python_ok)
There was a problem hiding this comment.
assert_python_ok() adds arguments to the command line. Here I really want to control the exact command line. Sadly, Python still leaks some references depending on the exact command line.
That's unfortunate :-( test_no_memleak() doesn't use _testembed program. |
|
test_embed fails on AMD64 Windows10 3.x: Strange. The test didn't fail on the Windows CI jobs on this PR. |
|
I wrote #31560 to fix winreg leaks. |
|
Windows buildbots are still failing |
|
I know and I don't have time to fix the remaining leak, so I created https://bugs.python.org/issue46857 and #31589 |
https://bugs.python.org/issue1635741