From: Petr Onderka <gsvick@gmail.com>
Date: Mon, 16 Aug 2010 16:51:54 GMT
Subject: Re: [PATCH/RFC] Add global and system-wide gitattributes
Message-ID: <AANLkTi=2tRjGicxJxnJ3xccCGxcP3bLUy8u0O7q85D=R@mail.gmail.com>
In-Reply-To: <7vzkwsvllz.fsf@alter.siamese.dyndns.org>

Hi,

On Thu, Aug 12, 2010 at 00:19, Junio C Hamano <gitster@pobox.com> wrote:
> Have you read the comment at the top of prepare-attr-stack?  This patch
> feels triply wrong:
>
>  - The attribute stack is arranged to have higher precedence file near the
>   top ($GIT_DIR/info/attributes used to be the highest).  The above
>   addition means that ~/.gitattributes from user's home trumps what is in
>   a particular repository.  That is backwards.  You may work on more than
>   one projects and have more than one repositories.  What you share among
>   them personally will go to ~/.gitattributes, while a setting specific
>   to a particular repository goes to $GIT_DIR/info/attributes and the
>   latter needs to be able to override the former.
>
>  - Same thing for git_attr_system() being at the end, which means you set
>   up your own $GIT_DIR/info/attributes (or ~/.gitattributes) carefully
>   but that can be broken by a selfish sysadmin who puts stuff that is
>   only useful to him in /etc/gitattributes, which is not what you want.
>
>  - Whenever we enter a new directory (either recursing into, or coming
>   back up), prepare_attr_stack() is called to pop the attributes records
>   from now-exited directories and push the attributes records from
>   directories we are about to descend into.  The current code knows that
>   the topmost element on the stack is special ($GIT_DIR/info/attributes)
>   and first pops it, adjust the stack for elements that came from the
>   directory hierarcy, and then pushes that back.  I don't see any code in
>   the patch to do the equivalent for these two new attribute sources.

Yeah, I realize now that I got the precedence wrong, the corrected
patch is in the following email.

But to your third point, I read the comment and the code worked,
because I put all the attributes from the 3 files into one attr_stack
struct (the one at the top of the stack). That's why I changed
read_attr_from_file too. Of course this is irrelevant now that those
attributes will be at the bottom of the stack.

Petr Onderka