From: Jeff King <peff@peff.net>
Date: Fri, 22 Feb 2008 05:42:29 GMT
Subject: Re: [PATCH 0/3] solaris test results
Message-ID: <20080222054229.GA32294@coredump.intra.peff.net>
In-Reply-To: <7vr6f5pnmz.fsf@gitster.siamese.dyndns.org>

On Thu, Feb 21, 2008 at 09:26:44PM -0800, Junio C Hamano wrote:

> Here is a possible solution.
> 
> This is a fairly mechanical substitution (grepping for "git diff -u"
> and "diff -u", and replacing them with test_compare_expect.

I think this approach is reasonable (though as I suggested in my
original message, I probably would have selected something a little
easier to type, like test_cmp).

> Test scripts can be run with --no-diff-u option, to use "git
> diff -u --no-index" for comparison instead.

I thought you were against using "git diff" in the test suite for this
purpose.  Using "diff -u" produces nice output in case of failure, but
even "cmp" can do the job of returning an exit code. Perhaps it makes
sense as a more robust fallback.

Perhaps the command line option should be --compare=cmp, which lets the
caller choose an appropriate fallback for their system (and which means
that you can use --compare="git diff" to make debugging more pleasant,
but confirm with --compare="cmp" before claiming breakage on the list).

But honestly, I do not care that much about this. I don't even usually
use Solaris; I happened to have access to a box, so I followed up on
Whit's bug report. It is probably not worth spending a lot of time
over-engineering a solution. I have no idea how common it is to run the
test suite on boxes without GNU diff.

-Peff