Uh oh!
There was an error while loading.Please reload this page.
- Notifications
You must be signed in to change notification settings - Fork34.3k
Description
Windows CI failure:test_itertools corrupts Unicode temp directory name undermbcs encoding, breaks subsequent tests
Summary
On Windows (official Azure CI), the CPython 3.15 test suite produces cascading failures originating fromtest_itertools. The test creates a temporary directory whose name contains a non-ASCII character. Because the Windows default filesystem encoding ismbcs, the Unicode byte sequence is misinterpreted and the directory name becomes corrupted on disk. The directory is never cleaned up, which causes all subsequent tests that create temporary directories to fail withFileNotFoundError.
CPython Version
Python 3.15.0a7 (main branch)Operating System / Platform
Windows (Azure Pipelines Microsoft-hosted runner)Filesystem Encoding
>>>importsys>>>sys.getfilesystemencoding()'mbcs'
CI Pipeline Context
Build step:
PCbuild\build.bat -e $(buildOpt)Test step:
PCbuild\rt.bat -q -uall -u-cpu -rwW --slowest --timeout=1200 -j0 \ --junit-xml="$(Build.BinariesDirectory)\test-results.xml" \ --tempdir="$(Build.BinariesDirectory)\test"Azure DevOps build log:
https://dev.azure.com/Python/cpython/_build/results?buildId=168864&view=logs&j=c8a71634-e5ec-54a0-3958-760f4148b765&t=ddcdae4e-111a-5c2a-2289-6b784c553924
Steps to Reproduce
- Check out the CPython
mainbranch. - Build on Windows:
PCbuild\build.bat -e <buildOpt> - Run the test suite:
PCbuild\rt.bat -q -uall -u-cpu -rwW --slowest --timeout=1200 ^ -j0 --tempdir="%TEMP%" - Observe a directory with a corrupted name (e.g.
packageæ/) left behind in the temp directory. - Observe that subsequent tests fail with
FileNotFoundError.
Expected Behavior
- All tests pass cleanly on Windows during CI builds.
- No test leaves behind artifacts that interfere with other tests.
- Test isolation is preserved — temporary directories are created and removed correctly.
- Unicode filenames are handled consistently regardless of platform filesystem encoding.
Actual Behavior
test_itertoolscreates a directory whose name contains a non-ASCII character (e.g.æ).- Under the Windows default codepage (
mbcs), the Unicode name is misinterpreted:æ(U+00E6) is stored asæ. - The corrupted directory isnot removed after
test_itertoolscompletes. - Subsequent tests (e.g.
test_class,test_pdb) fail when attempting to create temporary directories because the corrupted entry causesos_helper.temp_dirto raiseFileNotFoundError. - The runner reports a warning for
test_itertoolsand marks unrelated tests as failed, significantly increasing triage cost.
Relevant Log Excerpt
From Azure build 168864:
Warning -- files was modified by test_itertoolsAfter: ... 'packageæ/'test_class failed (env changed)FileNotFoundError: [WinError 3] unable to create temporary directory:'D:\a\1\s\build\test_python_9140æ'Root Cause (Analysis)
Two independent problems combine to produce this failure:
1. Encoding mismatch intest_itertools
The test constructs or receives a directory name containing a non-ASCII character. On Windows,sys.getfilesystemencoding() returns'mbcs' rather than'utf-8'. The UTF-8 byte sequence foræ (0xC3 0xA6) is reinterpreted under the ANSI code page as two separate characters, producing the visible corruptionæ. This is a violation of cross-platform Unicode path handling requirements.
2. Missing cleanup intest_itertools
The test creates the directory but does not register it for cleanup (e.g. viaaddCleanup,self.addCleanup, or a context manager such astempfile.TemporaryDirectory). The corrupted directory persists in the shared--tempdir location for the duration of the entire test run, causing every subsequent call toos_helper.temp_dir to fail withWinError 3.
Impact
| Area | Description | Severity |
|---|---|---|
| CI pipeline stability | All Windows builds fail aftertest_itertools runs | High |
| Test isolation | Subsequent tests inherit corrupted filesystem state | High |
| Unicode correctness | Non-ASCII filenames silently corrupted on Windows | Medium |
| Triage cost | Unrelated failures obscure the root cause | Medium |
| Local Windows dev | Developers running the full suite locally may be affected | Low |
Files Involved
Lib/test/test_itertools.py— primary location of the encoding issue and missing cleanupLib/test/support/os_helper.py— secondary impact point (temp_dircontext manager)
Linked Resources
- Azure DevOps build 168864:https://dev.azure.com/Python/cpython/_build/results?buildId=168864
- PEP 529 — Change Windows filesystem encoding to UTF-8:https://peps.python.org/pep-0529/
sys.getfilesystemencoding()docs:https://docs.python.org/3/library/sys.html#sys.getfilesystemencodingos.fsencode()docs:https://docs.python.org/3/library/os.html#os.fsencode