Re: [sigc] SEGV while using libsigc++ in an apparently legit way

[Régis Duchesne <hpreg vmware com>]

Murray,

Thanks for your reply.

> A simplified test case is attached. It does not need 3 base classes to
> crash.

Interesting. On my system, I came up with a minimal test case. If I 
removed a single line from it, then the problem would not show up. But I 
will try your test case.

> However, we'd rather have a libsigc++-only test case rather than one
> that uses gtkmm too, and the g++ developers would want one that did not
> involve even libsigc++.

I have already spent 2 hours to come up with this small test case and 
report the bug. I understand that it has dependencies, but quite frankly 
I think it will be easy to reproduce for anybody. I'm not planning to 
spend more time on making it even smaller. You are asking for a bit too 
much here.

> I left out the "-v -save-temps -O -g". They aren't necessary to make it
> crash.

-v and -save-temps are required when sending a bug report to gcc (this 
is bug number 20304 in gcc's bugzilla by the way).

On my system, -O is needed to reproduce the issue.

> It segfaults for me, on Fedora Core 3,
> [murrayc murrayclaptop ~]$ g++ --version
> g++ (GCC) 3.4.2 20041017 (Red Hat 3.4.2-6.fc3)
>
> The valgrind output is attached, from after pressing the Discard button.
> I used 
> valgrind --tool=memcheck --num-callers=20 ./a.out

I don't see the attachment.

> This probably belongs in bugzilla, where it can be explored more. You should give full details about your platform there.

In order to report this bug, I read the instructions at:

   http://libsigc.sourceforge.net/devel.shtml#mail

quoting:
"
There is a mailing list, libsigc-list gnome org for Libsigc++, to serve 
as a design discussion area as well as a place to report bugs.
"

Do you want me to report it in libsigc++'s bugzilla as well?

Thanks,
--
Regis "HPReg" Duchesne
VMware, Inc.