From: Jeff King <peff@peff.net>
Date: Tue, 17 Mar 2009 08:28:45 GMT
Subject: Re: [PATCH] Tests: use test_cmp instead of diff where possible
Message-ID: <20090317082845.GH18475@coredump.intra.peff.net>
In-Reply-To: <1237124036-1348-1-git-send-email-vmiklos@frugalware.org>

On Sun, Mar 15, 2009 at 02:33:56PM +0100, Miklos Vajna wrote:

> I intentionally did not touch t5000 - using test_cmp -r works for me,
> since the default is diff -u, but that would break the setup of users
> where GIT_TEST_CMP is set to cmp.

Definitely "test_cmp -r" is a bad idea since we don't know what
underlies it. But I wonder how portable "diff -r" is. Wasn't it a point
of contention in the recent cvsimport tests that were added (and Michael
ended up writing a mini recursive differ using some shell commands)?

OTOH, I haven't seen any complaints about it, so perhaps it is fine to
leave it.

-Peff