warnings.filters was modified by test_check_c_globals · Issue #95349 · python/cpython · GitHub

When I run my own automated tests against main, I get the following test failure on macOS and Ubuntu.

Warning -- warnings.filters was modified by test_check_c_globals
Warning --   Before: (4544233792, [('error', None, <class 'BytesWarning'>, None, 0), ('default', None, <class 'Warning'>, None, 0), ('default', None, <class 'DeprecationWarning'>, '__main__', 0), ('ignore', None, <class 'DeprecationWarning'>, None, 0), ('ignore', None, <class 'PendingDeprecationWarning'>, None, 0), ('ignore', None, <class 'ImportWarning'>, None, 0), ('ignore', None, <class 'ResourceWarning'>, None, 0)], [('error', None, <class 'BytesWarning'>, None, 0), ('default', None, <class 'Warning'>, None, 0), ('default', None, <class 'DeprecationWarning'>, '__main__', 0), ('ignore', None, <class 'DeprecationWarning'>, None, 0), ('ignore', None, <class 'PendingDeprecationWarning'>, None, 0), ('ignore', None, <class 'ImportWarning'>, None, 0), ('ignore', None, <class 'ResourceWarning'>, None, 0)])
Warning --   After:  (4544233792, [('error', None, <class 'BytesWarning'>, None, 0), ('default', None, <class 'Warning'>, None, 0), ('default', None, <class 'DeprecationWarning'>, '__main__', 0), ('ignore', None, <class 'DeprecationWarning'>, None, 0), ('ignore', None, <class 'PendingDeprecationWarning'>, None, 0), ('ignore', None, <class 'ImportWarning'>, None, 0), ('ignore', None, <class 'ResourceWarning'>, None, 0)], [('error', None, <class 'BytesWarning'>, None, 0), ('default', None, <class 'Warning'>, None, 0), ('default', None, <class 'DeprecationWarning'>, '__main__', 0), ('ignore', None, <class 'DeprecationWarning'>, None, 0), ('ignore', None, <class 'PendingDeprecationWarning'>, None, 0), ('ignore', None, <class 'ImportWarning'>, None, 0), ('ignore', None, <class 'ResourceWarning'>, None, 0), ('ignore', None, <class 'pkg_resources.PEP440Warning'>, None, 0)])

We've previously added code to preserve warnings.filters around tests that import pkg_resources it seems. I'm not entirely sure where it's coming from in this case, as it doesn't repro on Windows (or apparently on our buildbots), but I guess we need the same preservation around it.

@ericsnowcurrently I think this may be yours?

I'll take a look.

I patched my own copy to have this as the test:

    def test_check_c_globals(self):
        with save_restore_warnings_filters():
            with test.test_tools.imports_under_tool('c-analyzer'):
                from cpython.__main__ import main
            try:
                main('check', {})
            except NotImplementedError:
                raise unittest.SkipTest('not supported on this host')

Solves the changed global, but I only assume the test is still valid.

Sorry for the delay. I'll try to take a look today.

Hmm, that test is disabled and the code doesn't look like yours. Are you based on some other branch?

https://github.com/python/cpython/blob/main/Lib/test/test_check_c_globals.py#L14-L19

Yeah, I didn't merge any changes. It's just in my internal fork.

The test may be disabled, but the import still happens, and that's apparently what leaks the change.

I'm unclear on why test_check_c_globals would be causing this problem. Neither it nor anything in Tools/c-analyzer messes with warnings. Perhaps it is something with test_tools? (I'm using a couple helpers from there.)

Is this a new failure?

FWIW, I don't get a failure when I run the test locally. How are you invoking the tests?

I think it's something in the c-analyzer tool triggering it. Considering the warning being filtered is from pkgutil, it might be loading distutils/setuptools or pip? But once I remove the import from the test module, it doesn't trigger anymore.

Maybe you've got something else in your environment that loads this stuff earlier? So then it doesn't look like the test is adding the extra warning filter because it's already there?

I'm basically just running normal CI, it's just not on GitHub Actions.

No, it's not quite normal. I configure with custom --prefix and make install before doing make buildbottest, so I'm not sure how different that makes things.

I think it's something in the c-analyzer tool triggering it. Considering the warning being filtered is from pkgutil, it might be loading distutils/setuptools or pip? But once I remove the import from the test module, it doesn't trigger anymore.

Yeah, I suspect this is a consequence of PEP 632 and the fact that we're still using parts of distutils in the c-analyzer (or at least were until 11 days ago).

Do you still see the problem with the latest main?

Regardless, I don't mind adding the fix you suggested if needed.

Maybe you've got something else in your environment that loads this stuff earlier? So then it doesn't look like the test is adding the extra warning filter because it's already there?

I bet that's it.

Do you still see the problem with the latest main?

My build from 7 hours ago failed in the same way. I wouldn't have thought it was coming from distutils though, unless it's coming from the distutils test suite?

FYI, here's what I see:

$ ./python -m test test_check_c_globals
0:00:00 load avg: 0.15 Run tests sequentially
0:00:00 load avg: 0.15 [1/1] test_check_c_globals
$CPYTHON_ROOOT/Tools/c-analyzer/c_parser/preprocessor/common.py:2: DeprecationWarning: The distutils package is deprecated and slated for removal in Python 3.12. Use setuptools or check PEP 632 for potential alternatives
  import distutils.ccompiler

== Tests result: SUCCESS ==

1 test OK.

Total duration: 278 ms
Tests result: SUCCESS

I do get the warning but it doesn't cause a failure.

I was able to reproduce the failure you originally described with an out-of-tree build and a custom --prefix:

mkdir ../build
cd ../build
../cpython/configure --cache-file=../python-config.cache --enable-ipv6 --prefix=$HOME/build/installed CFLAGS=-O0 --with-system-expat --with-system-ffi --with-pydebug
make -j8
make install
./python -E ../cpython/Tools/scripts/run_tests.py -j 1 -u all -W --slowest --fail-env-changed --timeout=1200  test_check_c_globals

There's a comment in the distutils test that explains the situation:

cpython/Lib/distutils/tests/__init__.py

Lines 30 to 34 in 63140b4

    
           # bpo-40055: Save/restore warnings filters to leave them unchanged. 
        
           # Importing tests imports docutils which imports pkg_resources 
        
           # which adds a warnings filter. 
        
           with save_restore_warnings_filters(): 
        
               __import__(modname)

The referenced issue, gh-84236 (AKA bpo-40055) is basically the same as this one. Looks like here it's only happening with the specific build steps.

My fix (gh-95837) should be good enough until we stop using distutils in the c-analyzer.

gh-95349: Hide a Distutils Warning Filter for test_check_c_globals (G…

pythongh-95349: Hide a Distutils Warning Filter for test_check_c_glob…

gh-95349: Hide a Distutils Warning Filter for test_check_c_globals (G…

Thanks for pointing this out, @zooba. It should be all fixed now. Would you mind verifying?

It looks good in my build overnight 👍

pythongh-95349: Hide a Distutils Warning Filter for test_check_c_glob…

gh-95349: Hide a Distutils Warning Filter for test_check_c_globals (G…

pythongh-95349: Hide a Distutils Warning Filter for test_check_c_glob…

warnings.filters was modified by test_check_c_globals · Issue #95349 · python/cpython · GitHub

Repository navigation

warnings.filters was modified by test_check_c_globals #95349

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

	# bpo-40055: Save/restore warnings filters to leave them unchanged.
	# Importing tests imports docutils which imports pkg_resources
	# which adds a warnings filter.
	with save_restore_warnings_filters():
	__import__(modname)

warnings.filters was modified by test_check_c_globals · Issue #95349 · python/cpython · GitHub

Repository navigation

warnings.filters was modified by test_check_c_globals #95349

Description

Activity

ericsnowcurrently commented on Jul 27, 2022

zooba commented on Jul 28, 2022

ericsnowcurrently commented on Aug 1, 2022

ericsnowcurrently commented on Aug 1, 2022

zooba commented on Aug 1, 2022

ericsnowcurrently commented on Aug 1, 2022

zooba commented on Aug 1, 2022

zooba commented on Aug 1, 2022

ericsnowcurrently commented on Aug 3, 2022

zooba commented on Aug 3, 2022

ericsnowcurrently commented on Aug 9, 2022

ericsnowcurrently commented on Aug 9, 2022

ericsnowcurrently commented on Aug 9, 2022

ericsnowcurrently commented on Aug 9, 2022

ericsnowcurrently commented on Aug 9, 2022

ericsnowcurrently commented on Aug 10, 2022

zooba commented on Aug 10, 2022

Metadata

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

Issue actions