Re: Fwd: Possibly nicer pathspec syntax?

From: Linus Torvalds <torvalds@linux-foundation.org>
Date: Feb 8, 2017, 03:02 UTC
Message-ID: <CA+55aFwPLtuPciN1o_03CwkKqFWgZd_br9Q14qyr7a7N7mxTeA@mail.gmail.com>
In-Reply-To: <xmqqefz9xv0x.fsf@gitster.mtv.corp.google.com>

On Tue, Feb 7, 2017 at 6:42 PM, Junio C Hamano <gitster@pobox.com> wrote:

Show 9 quoted lines

>
>  1. I think some commands limit their operands to cwd and some work
>     on the whole tree when given no pathspec. I think the "no
>     positive?  then let's give you everything except these you
>     excluded" should base the definition of "everything" to that.
>     IOW, "cd t && git grep -e foo" shows everything in t/ directory,
>     so the default you would add would be "." for "grep"; "cd t &&
>     git diff HEAD~100 HEAD" would show everything, so you would give
>     ":(top)." for "diff".

No. The thing is, "git diff" is relative too - for path specifications. And the negative entries are pathspecs - and they act as relative ones.

IOW, that whole

  cd drivers
  git diff A..B -- net/

will actually show the diff for drivers/net - so the pathspec very much acts as relative to the cwd.

So no, absolute (ie ":(top)" or ":/") doesn't actually make sense for "diff" either, even though diff by default is absolute when not given a pathname at all.

But if you do

  cd drivers
  git diff A..B -- :^/arch

then suddenly an absolute positive root _does_ make sense,. because now the negative pathspec was absolute..

Odd? Yes it is. But the positive pathspec rules are what they are, and they are actually what I suspect everybody really wants. The existing negative ones match the rules for the positive ones.

So I suspect that the best thing is if the "implicit positive rule when there are no explicit ones" ends up matching the same semantics as the (explicit) negative entries have..

>  2. I am not sure what ctype.c change is about.  Care to elaborate?

I didn't see the need for it either until I made the rest of the patch, and it didn't work at all.

The pathspec.c code uses "if (is_pathspec_magic(..))" to test whether a character is a short magiic pathspec character. But '^' wasn't in that set, because it was already marked as being (only) in the regex set.

Does that whole is_pathspec_magic() thing make any sense when we have an array that specifies the special characters we react to? No it does not.

But it is what the code does, and I just made that code work.

>  3. I think our recent trend is to wean ourselves away from "an
>     empty element in pathspec means all paths match", and I think we
>     even have accepted a patch to emit a warning.  Doesn't the
>     warning trigger for the new code below?

It didn't trigger for me in my testing, I suspect the warning is at an earlier level when it walks through the argv[] array and fills in the pathspec arrays. But I didn't actually look for it.

                   Linus