From: Junio C Hamano <gitster@pobox.com>
Date: Wed, 27 Aug 2008 05:01:03 GMT
Subject: Re: [PATCH v2] Support "core.excludesfile = ~/.gitignore"
Message-ID: <7vy72jrr00.fsf@gitster.siamese.dyndns.org>
In-Reply-To: <quack.20080826T2012.lthvdxn2ls4@roar.cs.berkeley.edu>

Karl Chen <quarl@cs.berkeley.edu> writes:

>>>>>> On 2008-08-26 17:25 PDT, Jeff King writes:
>
>     Jeff>   1. It should probably re-use path.c:user_path, as
>     Jeff>      Johannes mentioned.
>
> That function has the wrong interface for this task (requires
> extra strdup, imposes PATH_MAX, conflates all error conditions
> into returning NULL, also returns NULL if input doesn't have "~").
> Do you still think it should re-use that function?

On the other hand you have a pair of ugly "casting a const pointer down,
temporarily modifying what is const" in your version.

In any case, I think their point is that these two functions are meant to
do the same thing, and a unified interface would be desirable.

You do not necessarily have to use user_path() from path.c, but
user_path(), if its interface is coarser, could become a thin wrapper
around yours, don't you think?  Even better, perhaps the current callers
of user_path() may benefit from finer distinction among error conditions
(I didn't look, though).