Re: [PATCH] Don't search files with an unset "grep" attribute

From: Junio C Hamano <gitster@pobox.com>
Date: Jan 23, 2012, 18:33 UTC
Message-ID: <7vy5sy8e0y.fsf@alter.siamese.dyndns.org>
In-Reply-To: <4f1d2a8b.a2d8320a.50ec.576d@mx.google.com>

conrad.irwin@gmail.com writes:

Show 7 quoted lines

> ---8<---
>
> To set -grep on an file in .gitattributes will cause that file to be
> skipped completely while grepping. This can be used to reduce the number
> of false positives when your repository contains files that are
> uninteresting to you, for example test fixtures, dlls or minified source
> code.

Please reword this to describe the problem being solved first (why it needs to be solved, in what situation you cannot live without the feature), and then describe the approach you used to solve it.

Plain "grep" does this:

	$ grep world hello.*
	hello.c: printf("Hello, world.\n");
        Binary file hello.o matches

in order to avoid uselessly spewing garbage at you while reminding you that the command is not showing the whole truth and you _might_ want to look into the elided file if you wanted to with "grep -a world hello.o". Compared to this, it feels that the result your patch goes too far and hides the truth a bit too much to my taste. Maybe it is just me.

Perhaps you, or all participants of your particular project, usually do not want to see any grep hits from minified.js, but you may still want to be able to say "I want to dig deeper and make sure I have copyright strings in all files", for example. It is unclear how you envision to support such a use case building on top of this patch.

Your "attributes only" is not an acceptable solution in the longer run, even though it is a good first step (i.e. "attributes first and other enhancement later"). There should be an easy way to get the best of both worlds.

> The other approach considered was to allow an --exclude flag to grep at
> runtime, however that better serves the less common use-case of wanting
> to customise the list of files per-invocation rather than statically.

I doubt that it is justifiable to call per-invocation "the less common".

By the way, if the uninteresting ones are dll and minified.js, I wonder why it is insufficient to mark them binary, i.e. uninteresting for the purpose of all textual operations not just grep but also for diff.

I am *not* going to ask why they are treated as source and tracked by git to begin with.