Re: [PATCH 1/5] treewide: rename 'struct exclude' to 'struct path_pattern'

From: Junio C Hamano <gitster@pobox.com>
Date: Sep 5, 2019, 21:03 UTC
Message-ID: <xmqq4l1qpiaw.fsf@gitster-ct.c.googlers.com>
In-Reply-To: <20190905065518.GD21450@sigill.intra.peff.net>

Jeff King <peff@peff.net> writes:

Show 12 quoted lines

> I wonder if there's a name that could more clearly distinguish the two.
> Or if it's sufficient to just become Git jargon that "pathspec" is the
> command-line one and "path_pattern" is the file-based one (we're at
> least pretty consistent about the former already).
>
> I think one could also make an argument that the name collision is a
> sign that these two things should actually share both syntax and
> implementation, since we're exposing too similar-but-not-quite versions
> of the same idea to users. But given the compatibility issues, it's
> probably not worth changing the user facing parts at this point (and I
> also haven't thought too hard about it; there may be reasons why the two
> _should_ differ).

Hmph.  I did not realize there are so many differences X-<.

A pathspec is relative to $CWD, and there is a syntax, i.e. prefixing with ":(top)", to make it relative to the root level. An entry in a .gitignore file will never affect paths outside the directory the file appears in. And there should never be such a mechanism to allow it.

An entry without slash in .gitignore is a basename match, and there is a syntax i.e. prefixing with "/", to anchor it to a single directory. A pathspec without slash also can be a basename match (e.g. "*.c" matches "a/b.c" as well as "d.c"). A pathspec with a slash can be made to tail-match (e.g. "**/*.c" matches "a/b.c", "a/b/c.c", etc.) but I do not think of a way to make an entry with a slash in a .gitignore file a tail-match the same way. I do not think this is intended but merely a missing feature.

So, yes, eventually we may want to make them more similar, but I suspect that there are some things that should be in one but never be in the other.