[bunbeans2-arch]

What

tools/configure_seed.py reconfigures stdout and stderr to UTF-8, but tools/tests/test_configure_seed.py read them with text=True, which decodes with the locale encoding (cp1252 on a Western Windows install). When TEMP contains a non-ASCII character, test_unreadable_cache_is_reported_without_a_traceback and test_unwritable_seed_is_reported_without_a_traceback look for the path in stderr, find it as mojibake (e_é) and fail.

The six subprocess.run calls that start configure_seed now pass encoding="utf-8" instead of text=True, the same change #12993 makes for test_unwrap_self.py. The cmake calls in the round-trip tests are unchanged, since cmake does not write UTF-8 on its own. No tool code changes.

Tested

Windows 11 Pro 10.0.22631 (code page 1252), Python 3.13.5, MSYS2 UCRT64 GCC 16.1.0 and CMake on PATH. py -I tools/tests/test_configure_seed.py, whole file, 3 runs each:

TEMP and TMP	main 1edecd6	this PR
...\e_é (inside code page 1252)	2 failures, 3 of 3 runs	29 tests OK, 3 of 3 runs
normal profile TEMP	29 tests OK	29 tests OK
...\zh_Ж (outside the code page)	2 failures + 2 errors	2 errors, the same as main

The two errors under Ж are setUpClass of ConfigureSeedTests and ConfigureSeedRoundTripTests: cmake -G Ninja with MinGW GCC fails to configure a project under that path, the toolchain limit already noted in #13083. This PR does not change them.

• python tools/check_conventions.py --base origin/main: all checks passed.
• python tools/pr_overlap.py --branch fix/tests-configure-seed-utf8: no overlap with open pull requests. fix(tools): report configure_seed symbolic link loops as usage errors #12614 and feat(tools): preview configure seeds on stdout #12770 also edit test_configure_seed.py and merge cleanly with this branch; fix(tools): report configure_seed symbolic link loops as usage errors #12614 adds one more text=True call that starts configure_seed, which would want the same encoding="utf-8".
• Not run here: Linux and macOS. encoding="utf-8" matches what the tool writes there too; three of the six calls are in POSIX-only tests and were skipped on Windows.

Checklist

• Based on current main; no other open PR implements the same functions
• One topic per PR; follow-ups go in a new PR
• No comments in code except technical debt
• Unimplemented paths throw (NotImplemented_nid_no_patch); silent stubs are listed under Silent stubs in their library's technical debt file
• No notes, investigation .md files or images added to the repository (attach them to this PR)
• New third-party code is a submodule built from source
• Generic behaviour, not specific to one title
• Depends on: none
• Fixes: none
• Unblocks: none
• GPUs tested: n/a
• AI-assisted: yes (Claude Code)

🤖 Generated with Claude Code

[JJenkx]

Windows run of head 3ecb625ae1, merged onto main 1c6b316cf7 (clean merge, local only), compared with that main. Windows 11 Enterprise 10.0.22621 (ANSI code page 1252), Python 3.15.1 with -X utf8=0 and PYTHONUTF8=0 (3.15 defaults to UTF-8 mode, which would hide the locale decode), WinLibs GCC 15.2.0 UCRT r7, CMake 4.3.1, Ninja. python -m unittest tools/tests/test_configure_seed.py, with TEMP and TMP set to each directory:

TEMP	main	this PR
C:\x\tmp-ascii	OK, 29 run, 2 skipped	OK, 29 run, 2 skipped
C:\x\tëmp (inside code page 1252)	2 failures	OK, 29 run, 2 skipped
C:\x\тест (outside the code page)	6 errors, 22 run	3 errors, 22 run

That matches your table. The three errors left under тест all come from real cmake -G Ninja configures (CalledProcessError from cmake -S ...\тест\..., plus a UnicodeDecodeError in the reader thread for the text=True CMake output). That's the MinGW path limit from #13083, not the configure_seed decoding this PR fixes. I see three rather than your two: the extra one is ChecksParsingTests.test_multiline_help_from_real_cmake_survives_seed_round_trip, which also configures a real CMake project under TEMP.

Logs: one run per cell; the 13278-py-<side>-<temp>.log files are on this machine.

AI-assisted: yes (Claude Code).

[malekaifa]

Reviewed exact head 3ecb625 on Windows ACP 1252. With TEMP=TMP=C:\tmp\e_é, the two affected ConfigureSeedPathTests ( est_unreadable_cache_is_reported_without_a_traceback and est_unwritable_seed_is_reported_without_a_traceback) both pass on this head. Running the same tests from current main under the same TEMP makes both fail because the parent process decodes configure_seed's UTF-8 stderr with the locale encoding, producing a mismatched/mojibake path. Current-main and #12614 pairwise merge trees are clean, and git diff --check passes. Hosted eligibility is green; generic relinker/ASan lanes are still queued/in progress. Non-blocking integration note: #12614 adds one new subprocess call that starts configure_seed with text=True; if #12614 lands after this PR, that call should use encoding='utf-8' as well so the locale-decoding gap is not reintroduced. No blocking issue found in #13278 itself.

[bunbeans2-arch]

Thanks for the run. Re-ran under ...\zh_Ж with -v on current main 51c31ea and on this head, 3x each (Python 3.13.5, MSYS2 UCRT64 CMake 4.3.3 + GCC 16.1): here test_multiline_help_from_real_cmake_survives_seed_round_trip passes on both, so 2 errors remain on this PR (setUpClass of ConfigureSeedTests and ConfigureSeedRoundTripTests). Your third one from the WinLibs CMake run is the same cmake -G Ninja path limit under a non-code-page TEMP, so the count depends on the toolchain; this PR does not touch those cmake calls.

AI-assisted: yes (Claude Code)

[MarcZX10]

Reviewed — looks good

The UTF-8 decoding change matches the tool’s explicit UTF-8 output, and the focused test file passes at the reviewed head.

What I ran

Reviewed head 3ecb625ae1cf6cdb7bc4e68871e842f3ae529ca0 against base e471817c68d4e1b1e44b0ce369ed88bafe4ae8ed; the diff is limited to subprocess output decoding in tools/tests/test_configure_seed.py. Environment: Debian 12 x86-64, Python 3.11.2, Git 2.39.5. No build was needed.

git diff --check e471817c68d4e1b1e44b0ce369ed88bafe4ae8ed...3ecb625ae1cf6cdb7bc4e68871e842f3ae529ca0
python3 tools/check_conventions.py --base e471817c68d4e1b1e44b0ce369ed88bafe4ae8ed --head 3ecb625ae1cf6cdb7bc4e68871e842f3ae529ca0 --title 'fix(tests): decode configure_seed output as UTF-8' --body /tmp/anyps5-pr-13278-body.md
python3 -m unittest tools.tests.test_configure_seed

Results

Check	Result
Diff whitespace	Passed.
Repository conventions	Passed.
tools.tests.test_configure_seed	Passed: 29 tests, 1 skipped.
Hosted checks	Conventions, intake, labels, eligibility, Ubuntu relinker, Windows relinker, and relinker ASan passed. Two macOS relinker jobs remained queued at review time; Linux, Windows-specific, register-context, and write-watch jobs were skipped.

Unedited local output

All conventions checks passed.
.................s...........
----------------------------------------------------------------------
Ran 29 tests in 4.428s

OK (skipped=1)

Findings

No blocking or non-blocking findings in the reviewed test-only change. The tool reconfigures its output streams to UTF-8, and the changed subprocess calls now decode those streams using UTF-8.

Sources

• configure_seed.py sets stdout/stderr to UTF-8.
• Reviewed test changes.

Not verified

I did not run the tests on Windows, Linux CI, or macOS. The hosted macOS relinker jobs were queued; they do not exercise this Python test file.

[adriadam10]

Checked the six subprocess.run calls that start configure_seed: the tool reconfigures stdout/stderr to UTF-8 (tools/configure_seed.py:64-65), so decoding with encoding="utf-8" is correct; cmake calls are untouched. Windows ACP 1252 runs by @JJenkx and @malekaifa confirm both path tests fail on main and pass here; merged.