prose_lint.py Decodes Git Output With the Locale Encoding, So a Windows Run Can Crash on a UTF-8 Diff

[ptr727]

Problem

prose_lint.py reads git output with text=True and no encoding, so Python decodes it with the locale's preferred encoding. On Windows, where Python's UTF-8 mode is off by default, that is the ANSI code page, commonly cp1252. Some UTF-8 byte sequences have no cp1252 mapping, so a diff carrying one raises UnicodeDecodeError. The except (subprocess.CalledProcessError, FileNotFoundError) around the call does not catch it, and the run ends in a traceback.

.github/actions/prose-gate/prose_lint.py at main (fb42028) has text=True without encoding= at lines 167, 201, 345, 504, 570, 599 and 615. Lines 167 and 201 are the diff-scope reads a --diff run always takes.

Evidence

This is unverified on Windows itself. The decode failure is shown directly:

>>> "Kyōto".encode("utf-8").decode("cp1252")
UnicodeDecodeError: 'charmap' codec can't decode byte 0x8d in position 3: character maps to <undefined>

A diff that adds or removes ō (UTF-8 C5 8D) therefore cannot be decoded under cp1252. CI on Linux runs in a UTF-8 locale and is unaffected.

Reach

A local run from Git Bash or PowerShell with a Windows CPython, which is the route a downstream repository's docs give for running the gate before a push.

Suggested fix

Pass encoding="utf-8" (and errors="replace" where a lossy read is acceptable) alongside text=True on each git call. Setting PYTHONUTF8=1 is a local workaround until then.

Found while documenting the local prose-gate invocation in ptr727/aiopurpleair.