Re: [PATCH] grep: provide pathspecs/patterns via file or stdin

[Denton Liu <liu.denton@gmail.com>]

Hi Junio,

On Fri, Nov 22, 2019 at 11:34:17AM +0900, Junio C Hamano wrote:
> Denton Liu <liu.denton@gmail.com> writes:
> 
> > The reason I ask is because (correct me if I'm wrong) a lot of other git
> > commands (like add, reset and checkout) don't seem to accept pathspecs
> > via stdin and could suffer the same problem. xargs seems like a more
> > general way of solving the problem of long command lines.
> 
> You contributors who are potentially throwing your own topics into
> the cauldron, please be paying a bit more attention to other topics
> already cooking in the pot.  I think am/pathspec-from-file wants to
> go in the general direction.

Interesting, I never caught this topic when it went over the list. I
guess I should read your What's Cooking emails more thoroughly instead
of just scanning for my own contributions.

> 
> There are things "xargs" is sufficient, and there are things that
> absolutely requires a single invocation of "git".  "grep" is a bit
> of both.
> 
>     $ git grep -e "$pattern" -- A B C
> 
> (where A, B and C are hundreds) can be split into three independent
> invocations of "git grep" via "xargs", essentially running
> 
>     $ git grep -e "$pattern" -- A
>     $ git grep -e "$pattern" -- B
>     $ git grep -e "$pattern" -- C
> 
> independently.

In the above, I was talking about the new --pathspecs-file option in
particular. So it looks like you agree with me that the new option
doesn't supercede xargs?

> 
> But
> 
>     $ git grep -e P1 -e P2 -e P3 -- A
> 
> (where each of "-e Pn" in reality may be "-e Pn1 -e Pn2 -e Pn3..."
> that has hundreds of patterns) cannot be split into separate
> invocations and keep the same meaning.
> 
>     $ git grep -e P1 -- A
>     $ git grep -e P2 -- A
>     $ git grep -e P3 -- A
> 
> may show the same lines, but (1) lines with both P1 and P2 would be
> shown duplicated, and (2) the order of the output would be different
> from a single invocation looking for all patterns at once.

We already have `-f` to handle this particular case, no?

> 
> Needless to say, the ability to combine patterns with --all-match,
> --and, etc., and negate them would mean the list of patterns must be
> split (even when it makes sense to do so) at the right places.
> 
>     $ git grep -e P1 --and -e P2
> 
> cannot be split into two or more invocations, for example.

Anyway, thanks for going into detail about this. It makes things a lot
more clear.

-Denton