Re: $PATH pollution and t9902-completion.sh

From: Jeff King <peff@peff.net>
Date: Dec 20, 2012, 14:55 UTC
Message-ID: <20121220145519.GB27211@sigill.intra.peff.net>
In-Reply-To: <20121217010538.GC3673@gmail.com>

On Mon, Dec 17, 2012 at 01:05:38AM +0000, Adam Spiers wrote:

Show 13 quoted lines

> t/t9902-completion.sh is currently failing for me because I happen to
> have a custom shell-script called git-check-email in ~/bin, which is
> on my $PATH.  This is different to a similar-looking case reported
> recently, which was due to an unclean working tree:
> 
>   http://thread.gmane.org/gmane.comp.version-control.git/208085
> 
> It's not unthinkable that in the future other tests could break for
> similar reasons.  Therefore it would be good to sanitize $PATH in the
> test framework so that it cannot destabilize tests, although I am
> struggling to think of a good way of doing this.  Naively stripping
> directories under $HOME would not protect against git "plugins" such
> as the above being installed into places like /usr/bin.  Thoughts?

I've run into this, too. I think sanitizing $PATH is the wrong approach. The real problem is that the test is overly picky. Right now it is failing because you happen to have "check-email" in your $PATH, but it will also need to be adjusted when a true "check-email" command is added to git.

I can think of two other options:

  1. Make the test input more specific (e.g., looking for "checkou").
     This doesn't eliminate the problem, but makes it less likely
     to occur.

  2. Loosen the test to look for the presence of "checkout", but not
     fail when other items are present. Bonus points if it makes sure
     that everything returned starts with "check".

I think (2) is the ideal solution in terms of behavior, but writing it may be more of a pain.

-Peff