Re: git-svn died of signal 11 (was "3 failures on test t9100 (svn)")

From: Jeff King <peff@peff.net>
Date: Mar 3, 2010, 17:31 UTC
Message-ID: <20100303173131.GB10168@coredump.intra.peff.net>
In-Reply-To: <20100303092213.GA12909@vidovic.wsb>

On Wed, Mar 03, 2010 at 10:22:13AM +0100, Nicolas Sebrecht wrote:

Show 10 quoted lines

>   $ cd okawix
>   $ cat .git/config
>   <...>
>   [svn-remote "svn"]
>     url = https://okawix.svn.sourceforge.net/svnroot/okawix
>     fetch = :refs/remotes/git-svn
> 
>   $ git svn fetch
> [...]
>   error: git-svn died of signal 11

I was able to clone this repository just fine with "git svn clone" (which should internally just be doing that same fetch). But since it is a segfault, and since you are having problems with some repositories and not others, it may be a problem with uninitialized data or similar.

And the fact that you are getting a perl program to segfault means it is probably not the perl program's fault, but a problem with perl itself or an XS module loaded by perl (like, say, the SVN module).

Do you have valgrind installed? Have you tried running

  valgrind perl /usr/libexec/git-core/git-svn fetch

(you may need to tweak the libexeec path depending on your git install). That should at least correctly identify the source of the segfault (you can try it with gdb, too, but often segfaults are caused by memory problems that happened much earlier, so valgrind will give you a more accurate answer).

Show 7 quoted lines

> > I'll do more investigations at work tomorrow. I've already tried to add
> > the "-d" flag in git-svn and it didn't help neither: git looked like in
> > an infinite loop. Odd, odd, odd.
> 
> Adding the -d option to the shebang of git-svn doesn't help. I've
> stopped the process after 24 minutes running.  I have no more
> information with the -w option set.

What's it doing for 24 minutes? Is it actually chewing on cpu, or is it blocking on I/O? Have you tried running it under strace?

-Peff