Written by Claude Opus 5 in Claude Code, running through @rawsun007's account, per CONTRIBUTING "Prevent agent impersonation". I speak only for myself here.

In git, a # or ; outside quotes starts a comment — with or without a space before it, and whether or not the value is quoted. GitConfigParser cut only a ; that was preceded by whitespace in a value that did not start with ":

if vi in ("=", ":") and ";" in optval and not optval.strip().startswith('"'):
    pos = optval.find(";")
    if pos != -1 and optval[pos - 1].isspace():
        optval = optval[:pos]

Measured against git config -f <file> --get a.k on git 2.50.1:

The last row is the worst of them: because the value starts with " and does not end with one, the comment made it look like an unterminated quote, so it was read as the start of a multi-line value and the following lines were swallowed into it.

The fix is small because the correct scan is already in this file: is_line_continuation walks a value tracking quotes and backslash escapes. strip_inline_comment reuses that walk to cut the comment before the quote-structure branches, so all three branches see comment-free text, and the legacy ;-only block is gone.

A comment character inside quotes stays literal ("has # inside" → has # inside), and both are pinned.

Escape handling is deliberately untouched. My first attempt also routed every unquoted value through parse_value, which resolves \t, \n and friends — that broke every Windows job, because a temp path like C:\Temp\test_x\config2 came back with a tab in it. git rejects that line outright (fatal: bad config line 2), so unescaping it would have matched neither git nor the previous behaviour. Whether GitPython should follow git and reject invalid escapes in unquoted values is a separate question; this PR does not touch it.

Verification: test/test_config.py is 46 passed, 2 skipped, 14 subtests. Reverting only git/config.py fails 5 of the 8 new subtests, leaving the two quoted-literal cases and the already-working value ; comment green. The full test/ run is 231 passed / 6 failed / 663 errors both with and without this change — those are a fixture-setup problem in my checkout, identical on a clean tree, and none are in test_config.py.

Not touched: [a] k = inline, a key on the same line as the section header, which git accepts and this parser rejects with NoOptionError. That is section-header parsing rather than comment handling, so it looked like its own change — happy to send it separately if you want it.