Re: [PATCH 01/12] tests: add initial bash completion tests

From: Junio C Hamano <gitster@pobox.com>
Date: Apr 11, 2012, 23:49 UTC
Message-ID: <7v1untj0cl.fsf@alter.siamese.dyndns.org>
In-Reply-To: <CAMP44s2EunUabZdU23e+jYnBrWtBmhZeqt9yR+wV3X9HZhd7hA@mail.gmail.com>

Felipe Contreras <felipe.contreras@gmail.com> writes:

Show 8 quoted lines

> On Sun, Apr 8, 2012 at 7:25 AM, Junio C Hamano <gitster@pobox.com> wrote:
> ...
>> What shell do you use on your system as /bin/sh (or if you use SHELL_PATH
>> in the Makefile to override it, what do you use)?
>
> /bin/sh is a symlink to bash on my system, and although bash defines
> POSIXLY_CORRECT, the tests pass just fine. I even tried to run /bin/sh
> --posix.

It really depends on what's in contrib/git-completion.bash and what you test.

For example, if a (possibly future) version of completion script had lines like these (taken from 6486ca6d7):

	__some_function ()
        {
		...
                while read key value; do
                        ...
                done < <(git config -z --get-regexp ...
		...
	}

invoking "bash --posix" would have failed like this:

        $ bash --posix
        bash-4.1$ _a ()
        > {
        >   while read x; do
        >     echo "* $x"
        >   done < <(echo a; echo b)
        bash: syntax error near unexpected token `<'

Perhaps the "initial tests" you posted, together with the version of the completion script we happen to have near 'master', didn't have any construct that gives a visible failure like the above, but that is just by dumb luck and is not a reason for us to be complacent and not to be careful.

The early part of your re-roll (i.e. v2) seems to be done carefully to make sure we run the script in full-fledged bash, from a quick glance.

Thanks.