From: Junio C Hamano <gitster@pobox.com>
Date: Sun, 06 Jun 2010 04:00:38 GMT
Subject: Re: What's cooking in git.git (Jun 2010, #01; Wed, 2)
Message-ID: <7vaar8yfmh.fsf@alter.siamese.dyndns.org>
In-Reply-To: <AANLkTimKwCDlW_uaCusZcmV2IttqORlUwSqkVnD3no6q@mail.gmail.com>

Sverre Rabbelier <srabbelier@gmail.com> writes:

> No addition to the test suite to catch this in the future?

If we had a test like this:

    test_expect_success 'run resolve' '
	... set up something to merge ...
        (
        	unset DIFF;
                git merge -s resolve ...
	)
    '

it would have caught this particular breakage, but

 (1) nobody could have anticipated before the problematic patch that $DIFF
     would be an exported variable that could cause trouble, so nobody
     could have written it before the $DIFF patch;

 (2) with the knowledge necessary to write the above test, the $DIFF patch
     wouldn't have made the misconversion, so it is unreasonable to say
     that the $DIFF patch should have come with the above test; and

 (3) having the above test in the test suite may catch exactly the same
     mistake in git-merge-one-file.sh, but that is not a guarantee that
     nobody will break another scripted Porcelain that is not triggered in
     the "resolve" codepath.

Other than doing "! git grep -e '$DIFF' 'git-\*.sh'" at the toplevel of
the source tree, I don't think it is possible to catch future bugs of this
sort.