From: Junio C Hamano <gitster@pobox.com>
Date: Wed, 09 May 2012 19:43:33 GMT
Subject: Re: [PATCH 08/19] completion: use $__git_dir instead of $(__gitdir)
Message-ID: <7vaa1hrvh6.fsf@alter.siamese.dyndns.org>
In-Reply-To: <1336524290-30023-9-git-send-email-szeder@ira.uka.de>

SZEDER Gábor <szeder@ira.uka.de> writes:

> +	__gitdir >/dev/null

If this becomes the only call site of __gitdir helper function (and that
was the way I read the log message), it would be sane to rename it to
a more descriptive __setup_dash_dash_git_dir function and lose the need to
redirect its output, no?

> @@ -962,7 +967,8 @@ __git_aliases ()
>  # __git_aliased_command requires 1 argument
>  __git_aliased_command ()
>  {
> -	local word cmdline=$(git --git-dir="$(__gitdir)" \
> +	__gitdir >/dev/null
> +	local word cmdline=$(git --git-dir="$__git_dir" \
>  		config --get "alias.$1")
>  	for word in $cmdline; do
>  		case "$word" in

Now this worries me.  The way I read 07/19 was that the local __git_dir=""
declarations in __git_ps1 and __git were what protected this whole
machinery to protect us against surprises from user doing "cd" between
interactive commands, but you have the same __gitdir call to set up the
global $__git_dir variable there, without the initialization to "".

Having to have a call to __gitdir seems to indicate to me that you cannot
assume that the other initialization sites may not have been called before
we get to this point.  Then why is 'local __git_dir=""' unneeded here?