Files
msd-core/tests
sim 2134d97396 test(#3103): expect git's separators, not the ones this machine happens to use
The Windows shard failed on the worktree-root assertion:

  actual   C:/Users/runneradmin/AppData/Local/Temp/gsd-wt-info-Nn4hj6
  expected C:\Users\runneradmin\AppData\Local\Temp\gsd-wt-info-Nn4hj6

The value comes straight from `git rev-parse --show-toplevel`, and git reports
POSIX separators on every platform. The expected side was built with the native
realpath, so the test encoded the separator convention of the machine it was
written on. Normalising the expected side keeps the assertion exact — on POSIX
the replacement is a no-op, so nothing is weakened where it already passed.

This is the same assertion that was strengthened earlier today from a
typeof-string-and-non-empty shape check. Pinning the exact path was right; the
weaker version would have passed on Windows precisely because it asserted
almost nothing. Getting a real assertion wrong on one platform is the better
failure, and the Linux-only matrix could not see it — the platform shards
caught it, as they did twice in the previous wave.

Every other path comparison in the two new test files was swept for the same
mistake. The remaining ones are safe: the base-branch tests compare against
literal POSIX strings supplied to a mocked git, and the reap tests put both
sides through one canonicalising helper, so they cannot disagree on separators.
Drive-letter case can differ in principle at the fixed site; it did not here and
no case-folding was added on speculation.

Refs #3057

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-06 02:16:13 -04:00
..