Re: [PATCH 1/4] is_directory(): a generic helper function

From: Junio C Hamano <gitster@pobox.com>
Date: Sep 9, 2008, 14:51 UTC
Message-ID: <7vprnd8jad.fsf@gitster.siamese.dyndns.org>
In-Reply-To: <20080909123207.GA25799@coredump.intra.peff.net>

Jeff King <peff@peff.net> writes:

Show 12 quoted lines

> On Tue, Sep 09, 2008 at 12:31:39PM +0200, Johannes Sixt wrote:
>
>> Junio C Hamano schrieb:
>> > +/*
>> > + * Do not use this for inspecting *tracked* content.  When path is a
>> > + * symlink to a directory, we do not want to say it is a directory.
>> 
>> I though stat(2) checks the thing that a symlink points to. Then either
>> this comment is not correct or you want to use lstat(2), no?
>
> I assume he meant "do not use this for tracked content, because in that
> case...". But I had to read it twice to make sense.

Sorry, I can't type, but you assumed correctly. When we are tracking a path "a/b" where "b" is a submodule, we want to notice the change made by the user to rmdir "a/b" and replace it with a symbolic link that points elsewhere, so we shouldn't be using is_directory() in such a codepath. Worse yet, we also want to catch the case where you "rm -fr a" and make a symlink that points at a directory that may (or may not) have a subdirectory "b". If we were tracking "a/b", then we have to say that the directory "a/b" does not exist anymore. It shows that is_directory() is a wrong function to use while inspecting the tracked content (aka worktree).