gh-95027: Ensure test runner uses utf-8:surrogateescape for communicating with subprocesses #96669
Conversation
zooba
commented
Sep 7, 2022
•
zooba
commented
- Issue: Running tests in parallel on Windows quits too soon #95027
…municating with subprocesses
|
I tried to hand-apply this patch to installed 3.11.0rc1, but Lib/test/libregrtest/runtest_mp.py has changed too much. I tried copying the patched 3.12 libregrtest into 3.11, but that did not work either. I then ran the test with -j0 in updated main and got the same error as on the issue. Finally, I created a branch from this pr and ran the tests in paralled again and they ran to completion. So this patch appears to fix the issue I reported. (There were other problems -- see issue.) So from my viewpoint, it looks good. |
| env = dict(os.environ) | ||
| env['PYTHONIOENCODING'] = 'utf-8:surrogateescape' |
There was a problem hiding this comment.
I would prefer to test the Python "default" encoding, and so fix code which decodes test stdout instead.
There was a problem hiding this comment.
Python test suite worked well with any encoding for years without having to force an encoding, it was broken recently to support WASM.
There was a problem hiding this comment.
Then remove the encoding='utf-8' from the code that reads from the file.
But since this isn't testing, and is just IPC between test runners, I'd prefer to keep full fidelity in the data.
|
I wrote my PR #98492 Is this change still relevant? |
