Re: Massive testsuite failure on !(Linux || Solaris 8+)

From: Jonathan Nieder <jrnieder@gmail.com>
Date: Apr 26, 2010, 07:44 UTC
Message-ID: <20100426074411.GA31428@progeny.tock>
In-Reply-To: <20100316100917.GA27121@thor.il.thewrittenword.com>

Hi Gary,

Gary V. Vaughan wrote:

> I have full logs on all hosts, up until t9300 test 120, which hangs (I
> left it overnight to be sure since some of these machines are quite
> old and slow) the testsuite until SIGINT.  The logs are, obviously,
> huge so I won't post them here,

What tends be most useful is output from

  GIT_TEST_OPTS='-v -i' gmake test

This way you can see the actual output from the failing commands, and you only have to worry about the first test that fails (no doubt the later ones are similar). Certainly if any tests from t0000-basic.sh are failing, it is useless to worry about problems in later tests. Options for use when invoking tests are documented in t/README.

Also, if you would like to investigate, you might find it useful to use

  sh -x relevant-test.sh

Based on the list of failing tests, though, it seems that your test_cmp is not working correctly. You should make sure that the GIT_TEST_CMP environment variable is set to point to some command that will exit with status 0 if the two files are identical and with nonzero status if they differ. I imagine patches to the Makefile and t/Makefile to set this up appropriately for your platforms (without ruining the nice diff output for overone else) would be welcome.

Hope that helps, Jonathan