Re: [RFC/PATCH] completion: avoid "words" as variable name for zsh portability

From: Jonathan Nieder <jrnieder@gmail.com>
Date: Apr 27, 2011, 09:11 UTC
Message-ID: <20110427091140.GB14849@elie>
In-Reply-To: <BANLkTinA5hfddqpGwOBjk+2oFDpwqORrSg@mail.gmail.com>

Felipe Contreras wrote:

> On Wed, Apr 27, 2011 at 9:40 AM, Jonathan Nieder <jrnieder@gmail.com> wrote:

>> Sorry, I got carried away (I am happy to see someone has made some
>> headway in investigating this old bug).  How about this?
>
> What's wrong with my patch?

As mentioned at http://thread.gmane.org/gmane.comp.version-control.git/172142/focus=172157 it breaks the tab completion in the common case that the user uses the standard bash completion library (usually provided at /etc/bash_completion) and git uses its _get_comp_words_by_ref function. You can test like so:

	% bash
	$ . /etc/bash_completion
	$ . contrib/completion/git-completion.bash
	$ git fetch origin <TAB>

I also made a small cosmetic change which is less important (sorry, I should have mentioned it before): the patch I sent spells out comp_words instead of writing cwords to avoid a false analogy between the array of all completion words (cwords) and the current word index (cword).

>> There is still a "for unknown reasons" in the above explanation.
>
> I'm asking zsh guys:
> http://www.zsh.org/mla/workers/2011/msg00515.html

Thanks. It looks like to get the semantics I expect from "local" in zsh, one needs to use "typeset -h" (which bash does not support, unfortunately). Probably it is best to steer clear of zsh's special variables anyway.