test: add coverage for alternate version_path spellings

[MohammedAlkindi]

Closes #1366

@zzzeek asked on the issue for this to be "confirmed as fixed and maybe even a test added to confirm it" — this does both. I'm on Windows 11, so I could exercise the reported paths directly rather than reason about them.

Confirmed fixed

Built a multi-head project whose version locations contain spaces, seeded a head in each, then created a revision naming the same directory five different ways. On 1.19.2:

spelling of --version-path	result
exactly as configured	accepted
lower-case drive letter (c:\…)	accepted
upper-case drive letter (C:\…)	accepted
trailing separator	accepted
spaces anywhere in the path	accepted

So the original report — drive-letter capitalisation — is genuinely resolved by the move to pathlib, exactly as you expected. PureWindowsPath.__eq__ is case-insensitive, so .absolute() == .absolute() handles it.

The test

Nothing pinned that. MultiDirRevisionCommandTest gains one test asserting a revision lands in the configured directory when the path carries a trailing separator, and — where a drive letter exists — when its case is swapped. It is written to be portable rather than skipped: os.path.splitdrive returns an empty drive on POSIX, so only the trailing-separator spelling runs there.

To check it has teeth, I replaced the case-insensitive Path comparison in generate_revision() with a string comparison:

-            if vers_path.absolute() == norm_path:
+            if str(vers_path.absolute()) == str(norm_path):

Only the new test fails; the four existing tests in that class still pass. Reverting the whole comparison to raw strings fails the new test plus two existing ones. So it covers ground the suite did not.

Two spellings that still fail — deliberately not changed here

While testing I found the same error is still raised for two other spellings of a configured directory:

• a path routed through .. — …\migrations\zzz\..\versions before
• the 8.3 short name — C:\…\Temp\ALEMBI~3\MYPROJ~1\MIGRAT~1\VERSIO~1

Both are the same cause: Path.absolute() normalises neither, while Path.resolve() handles both (verified directly). I have not touched that, because switching to resolve() also starts following symlinks, which is a behaviour decision for you rather than something to slip into a test PR. Happy to open a separate issue or PR if you want it changed.

On the "spaces" reports in this thread

@Tayyab-H and @SalamM1 reported failures whenever the path contained spaces, which didn't match what I measured. It reproduces exactly when path_separator is absent from the config: alembic then falls back to legacy splitting on spaces and commas, which cuts any path containing a space in half. Same project, same path, only that key differing:

path_separator = os    -> revision created
(key removed)          -> FAILED: Path C:\...\my project\migrations\versions before
                          is not represented in current version locations

A DeprecationWarning is emitted, but DeprecationWarning is hidden by default outside __main__, so it never reaches the user and they see only the confusing path error. alembic init has written path_separator = os since 1.16.0, so this only affects configs predating that. That may be worth its own issue — say the word and I'll file it rather than widen this PR.

Testing

Windows 11, Python 3.13, alembic 1.19.2, SQLAlchemy 2.0.51.

• pytest tests/ — 1814 passed, 5 failed, 132 skipped. The 5 failures are all ScriptNamingTest timezone tests failing on a clean main checkout too (zoneinfo has no system tzdata on Windows); identical before and after this change, verified by stashing it and re-running. No new failures.
• tests/test_script_production.py alone: 39 passed on clean main, 40 with this change — same 5 pre-existing failures.
• black --check, flake8 and zimports --diff are clean on the changed file.
• I could not run the Linux or macOS jobs locally; the new test's POSIX path is the trailing-separator spelling only, and CI is the authority there.

[CaselIT]

thanks, looks ok, just a note regarding the changelog

[CaselIT]

I think it would make more sense to remove this changelog, since there are no code changes and from an user point of view the fact that we only added a test isn't really useful

[MohammedAlkindi]

Agreed, removed — no code change, so there is nothing for a user to act on. The PR is now just the test.

pytest tests/test_script_production.py::MultiDirRevisionCommandTest -> 5 passed after the removal.

[MohammedAlkindi]

This has been approved since 2026-08-11 with all seven CI checks green, but gerrit_review has stayed PENDING the whole time. Is there something needed from me on the Gerrit side, or did the change not get mirrored across?

[CaselIT]

Sorry i forgot to move it to gerri

[sqla-tester]

OK, this is sqla-tester setting up my work on behalf of CaselIT to try to get revision 2d1f834 of this pull request into gerrit so we can run tests and reviews and stuff

[sqla-tester]

New Gerrit review created for change 2d1f834: https://gerrit.sqlalchemy.org/c/sqlalchemy/alembic/+/6909