The list_distinfo_files method does not fulfills its mission.  I changed the test file to be more stupid (and therefore not use high-level functions that we can’t trust) when building the expected list and discovered that list_distinfo_files returned too many files!

I would need someone to test the attached patch on Windows to see if I need to replace '/' in self.path or not (the code before the XXX comment).

With the patch applied, I still get failures in test_old_record_extensions
and test_record_basic, so it doesn't seem to work.

Thanks.  Can you test again after removing these two lines:

        # XXX see first if we need to convert at all
        prefix = self.path

Ahem, I read your message too fast.

> With the patch applied, I still get failures in test_old_record_extensions
> and test_record_basic, so it doesn't seem to work.
These are unrelated, I’m only interested in test_database for this patch.

> These are unrelated, I’m only interested in test_database for this patch.

I see. I'll re-run the tests this evening, and see whether that failure
crops up.

The tests using build_ext now pass \o/  I would love for someone with Windows and a 3.3 clone to test this patch (and if it does not fix, try with the two lines mentioned in msg142773 removed).

I've tested the attached patch against 3.3 on Windows XP, and it seems to
fix the test_database failures. There were some merge conflicts when I
applied the patch (because some of the docstring and comment changes had
already been committed in fixes from the other issue), so I've uploaded
the exact diff of what I tested with.

I tested both with and without the change you suggested in msg142773 -
the tests pass in both cases.

> I've tested the attached patch against 3.3 on Windows XP, and it
> seems to fix the test_database failures.
Thanks.

> I've uploaded the exact diff of what I tested with.
Your diff matches the one I get after updating my clone, we’re good.

> I tested both with and without the change you suggested in msg142773 -
> the tests pass in both cases.
I have to have another look at the code to make sure this makes sense and take the right approach.

I have added tests to make sure the return value (depending on the local parameter) is correct.  Please test when you have a free slot.  If it fails, the usual line after the XXX comment should be deleted and the test re-run.

The attached patch seems to work as-is.  That is, just testing for `self.path` as the prefix.  On Windows, at least, the paths in RECORD are always absolute.

Further changes will be necessary, of course, once changes for alternative paths (--prefix, --home, and so on) are incorporated.

New changeset c6d52971dd2a by Éric Araujo in branch 'default':
Fix packaging.database.Distribution.list_distinfo_files (#12785).
http://hg.python.org/cpython/rev/c6d52971dd2a

Backported to distutils2 as 5e49c9bc43eb.