Re: [PATCH] grep: add --max-count command line option

From: Junio C Hamano <gitster@pobox.com>
Date: May 16, 2022, 15:18 UTC
Message-ID: <xmqqee0t5wv5.fsf@gitster.g>
In-Reply-To: <e89577f8-8f52-bf09-15f3-c534bf1a6c64@cs.ucla.edu>

Paul Eggert <eggert@cs.ucla.edu> writes:

Show 13 quoted lines

> On 5/15/22 22:57, Junio C Hamano wrote:
>
>> It indeed is curious why GNU grep chose to immediately exit with 1
>> when "-m 0" was given,
>
> As I vaguely recall, if "-m 1" stops before "-m 2" does, then the idea
> was that it's reasonable for "-m 0" to stop before "-m 1" does, and
> the logical place to stop is right at the start, before any matches
> are found (i.e., exit with status 1).
>
> What would be more useful for 'grep -m 0' to do? (Sorry, I came into
> this conversation just now.) Perhaps GNU 'grep -m 0' should change, if 
> there's something better for it to do.

"grep -m 0" that declares a failure upfront because it is asked to stop before finding any match, combined with the fact that the command is expected to signal a failure after finding no matches, is an optimization that is mathmatically correct ;-)

It was asked as a part of discussion on a proposed patch to teach the same "-m <max-number-of-hits>" option to "git grep" what it ought to mean to give "-m 0". As we are too accustomed to the "last command line option wins" behaviour, I initially did not find the behaviour of the proposed patch, where 0 (or negative) stood for "unlimited", quite natural and useful (e.g. it allows overriding a hardcoded default option in aliases, "[alias] gg = grep -m 4"), and then was surprised by the "'-m 0' is an immediate failure" in GNU grep. I would call it mathematically pure and correct but of dubious utility.

Sorry for not providing enough context. Full discussion is seen at https://lore.kernel.org/git/pull.1264.git.git.1652361610103.gitgitgadget@gmail.com/

Show 9 quoted lines

>> What "git grep -m -1" should do?  IIRC, OPT_INTEGER is for signed
>> integer but the new .max_count member, as well as the existing
>> "count" that is compared with it, are of "unsigned" type.  Either
>> erroring out or treating it as unlimited is probably fine, but
>> whatever we do, we should document and have a test for it.
>
> 'grep -m -1' treats the count as being unlimited, but this isn't
> documented and (from the code) appears to be accidental. It'd make
> sense for it to be documented.

Thanks. The question was asked for the proposed addition to "git grep", but it is funny to see it apply equally well to GNU grep ;-).