Re: 'git grep needle rev' attempts to access 'rev:.../.gitattributes' in the worktree

From: Nguyen Thai Ngoc Duy <pclouds@gmail.com>
Date: Oct 10, 2012, 05:45 UTC
Message-ID: <CACsJy8AoLvWc9NCkqX_gEiBwLZDDx7TA9_CEMmXbb=k8iDS7_Q@mail.gmail.com>
In-Reply-To: <7vwqyydh6v.fsf@alter.siamese.dyndns.org>

On Wed, Oct 10, 2012 at 12:33 PM, Junio C Hamano <gitster@pobox.com> wrote:

Show 19 quoted lines

> Nguyen Thai Ngoc Duy <pclouds@gmail.com> writes:
>
>> On Wed, Oct 10, 2012 at 1:59 AM, Junio C Hamano <gitster@pobox.com> wrote:
>>> Jeff King <peff@peff.net> writes:
>>>
>>>> I think we just need to have callers of grep_source_init provide us with
>>>> the actual pathname (or NULL, in the case of GREP_SOURCE_BUF). That is
>>>> where the information is lost.
>>>
>>> Yes.  I agree that is the right approach.
>>
>> Passing full path this way means prepare_attr_stack has to walk back
>> to top directory for every files (even in the same directory). If
>> .gitattributes are loaded while git-grep traverses the tree, then it
>> can preload attr once per directory. But Jeff's approach looks
>> simpler.
>
> Why can't you do both?  That is, to build a full path as you
> descend, and read per-directory .gitattributes as you go?

Hm... I need to check attr.c code but I think it means read .gitattributes and prepare attr stack in builtin/grep.c (where we traverse trees) and actually check the attribute in grep_source_load_driver(), much further down the call stack. I'm not sure how we can pass the prepared attr stack down to grep_source_load_driver().

-- 
Duy