Re: [PATCH] Add two grep config options

From: Michael J Gruber <git@drmicha.warpmail.net>
Date: Mar 28, 2011, 07:24 UTC
Message-ID: <4D9037AA.9090601@drmicha.warpmail.net>
In-Reply-To: <7vlj024wal.fsf@alter.siamese.dyndns.org>

Junio C Hamano venit, vidit, dixit 26.03.2011 00:25:

Show 35 quoted lines

> Joe Ratterman <jratt0@gmail.com> writes:
> 
>> grep.extended-regexp: Enabling this boolean option has the same effect
>> as adding "-E" to all "git grep " instantiations.  This can be
>> disabled by specifying "--no-extended-regexp" on a particular call.
>>
>> grep.line-numbers: Enabling this boolean option has the same effect as
>> adding "-n" to all "git grep " instantiations.
>>
>> Signed-off-by: Joe Ratterman <jratt0@gmail.com>
> 
> Thanks.
> 
> Things to consider:
> 
>  - Apply this patch on top of "master",run "git shortlog v1.7.4..HEAD",
>    store the output somewhere, and imagine reading that 2 months from now.
>    Does a single line in the output about this patch sufficiently tell you
>    what it was about?
> 
>  - Configuration variables are spelled without hyphens between words (you
>    can see "gui.commitmsgwidth" in the context of the patch you sent and
>    notice that it is not "gui.commit-msg-width").
> 
>  - This will break scripts people have written, knowing that they can rely
>    on "grep" they wrote without giving "-E" from their command line will
>    use BRE, and force them to update the script with --no-extended-regexp
>    for no good reason.  Worse yet, there isn't even --no-line-numbers
>    supported to defeat grep.linenumbers configuration to protect such
>    scripts.
> 
>    I understand that some people would feel that the convenience would
>    outweigh the risk of script breakage in this particular case, and I am
>    sympathetic to the cause, but I still have to point it out.  Is there
>    anything we can do to mitigate the risk somehow?

This comes up again and again, and I feel that rather than adding config options one by one, we should either allow aliases for standard commands and/or setting default options depending on the mode (ui use vs. scripting use), so to say a companion to "git -c n=v" which allows

git config ui.grep "-E -n"

I.e. just like "git -c n=v <cmd>" sets up pseudo config before running cmd, our wrapper could augment argv from "ui.<cmd>".

We could safeguard scripts from this by

- checking istty and
- checking env for GIT_PLUMBING

and setting the latter in git-sh-setup.sh. After a long migration phase, we could skip the first (fragile) check.

Michael