Re: [PATCH v4 04/13] dir: fix pattern matching on dirs

On 11/2/2021 11:33 AM, Ævar Arnfjörð Bjarmason wrote:

We have t3070-wildmatch.sh testing various combinations of these, and
indeed this code resolves back to wildmatch().

But I think in this case this is due to dir.c's particular behavior of
splitting paths before feeding them to wildmatch, as it needs to match
things relative to the subdirectory.

Still, we've got a matrix of these in t3070-wildmatch.sh, which already
tests some (but apparently not all) cases where we need to create an
actual file on disk. These sorts of test blindspots are why I added that
in de8bada2bf6 (wildmatch test: create & test files on disk in addition
to in-memory, 2018-01-30).

Wouldn't it be better & more exhaustive here to change its test lines
like:


    match 0 0 1 1 foo/bar/baz 'bar'

To say:

    match 0 0 1 1 ? ? foo/bar/baz 'bar'

And just add to its match() function so that if we have a subject with a
slash, we mkdir up to that first slash (here: "mkdir foo"), and create a
.gitignore file therein with the "foo" directory with the "bar" content,
perhaps adding "/bar", "bar/" and "/bar/" variants implicitly.

Then create a "bar.txt" in the directory as well, and a
bar-otherdir/somefile.txt or whatever.

And fill in the "? ?" depending on whether those variants matched or
not...

Anyway, just an idea, but if you pursue that you should be able to get
much more exhaustive testing in this area that we've apparently had a
blindspot in.

Those tests are quite exhaustive, but the test script is pretty
inscrutable and the refactor you suggest is pretty major. I'd prefer
to keep to the focused test for the sake of fixing this in time for
the release.

Thanks,
-Stolee