Re: 3 failures on test t9100 (svn)

From: Jeff King <peff@peff.net>
Date: Mar 3, 2010, 17:23 UTC
Message-ID: <20100303172302.GA10168@coredump.intra.peff.net>
In-Reply-To: <20100303005804.GA17120@vidovic>

On Wed, Mar 03, 2010 at 01:58:04AM +0100, Nicolas Sebrecht wrote:

Show 8 quoted lines

> >   GIT_TEST_CMP='diff -u'
> >   test_cmp() {
> >     $GIT_TEST_CMP "$@"
> >   }
> > 
> > which _should_ split the "diff -u" on whitespace.
> 
> I've tried with these exact lines without more success.

That's weird. Can you test this in each of your shells:

  foo() {
    echo "in foo, arg=$1"
  }
  VARIABLE='foo bar'
  $VARIABLE

It should print "in foo, arg=bar". If it doesn't, then the shell is broken.

Also, can you confirm that you aren't setting SHELL_PATH in the Makefile (or in your environment) when building and testing git?

>   $ ls -ld $(which sh)
>   lrwxrwxrwx 1 root root 4 févr. 13  2009 /bin/sh -> bash
>   $ bash --version
>   GNU bash, version 3.2.39(1)-release (x86_64-pc-linux-gnu)

I have this same version of bash on my Debian stable box, and it passes the test I outlined above. So perhaps there is something more subtle going on.

Show 5 quoted lines

> That said, I did the svn tests because sometimes I have reproductible
> errors with 'git svn fetch' or 'git svn clone'. I'm not sure how both
> problems could be related... git-svn is a Perl script. The given message
> for this one is "error: died with signal 11" from what I remember (but
> I'm sure about the "signal 11" part).

I doubt they are related. This test_cmp problem is a problem in the test harness, but it may be masking actual problems by giving false negatives.

-Peff