f20 gcc bug: what to do?

[Rich Mattes]

Hi all,

I'm trying to get some FTBFS bugs taken care of so that the Robotics spin
will compose and we can get it tested before release.  I'm running into a
problem with the "mrpt" package.  The build triggers an internal compiler
error in gcc on ARM (the other two archs work fine.)  I filed a bugzilla
report[1], but I don't know if anyone will get to it anytime soon.

Do you think I should just add an ExcludeArch to mrpt in f20+ until this
bug is resolved, or is someone able to look into it?  Is there anything I
can try to work around it (try different compiler flags, etc?)

Thanks,

Rich

[1] https://bugzilla.redhat.com/show_bug.cgi?id=1007586

[Andrew Haley]

On 09/17/2013 09:27 PM, Rich Mattes wrote:
> Do you think I should just add an ExcludeArch to mrpt in f20+ until this bug is resolved, or is someone able to look into it?  Is there anything I can try to work around it (try different compiler flags, etc?)

It's always worth trying -O0.

Andrew.

[Dan Horák]

On Wed, 18 Sep 2013 08:50:52 +0100
Andrew Haley  wrote:
> On 09/17/2013 09:27 PM, Rich Mattes wrote:
> > Do you think I should just add an ExcludeArch to mrpt in f20+ until
> > this bug is resolved, or is someone able to look into it?  Is there
> > anything I can try to work around it (try different compiler flags,
> > etc?)
>
> It's always worth trying -O0.

and reduce debuginfo size by changing from "-g" (equals to -g2) to
"-g1", you can see various tricks to reduce memory consumption in the
webkitgtk3 spec file [1]

Dan

[1]
http://pkgs.fedoraproject.org/cgit/webkitgtk3.git/tree/webkitgtk3.spec#n102

[Andrew Haley]

On 09/18/2013 09:04 AM, Dan Horák wrote:
> and reduce debuginfo size by changing from "-g" (equals to -g2) to
> "-g1",

Why would anyone do that?  It'll spoil the debug experience.

Andrew.

[Dan Horák]

On Wed, 18 Sep 2013 09:08:27 +0100
Andrew Haley  wrote:
> On 09/18/2013 09:04 AM, Dan Horák wrote:
> > and reduce debuginfo size by changing from "-g" (equals to -g2) to
> > "-g1",
>
> Why would anyone do that?  It'll spoil the debug experience.

you are right, but when on the 32-bit platforms the SUM(*.o) > 4 GB (or
> 2 GB on s390), the linker simply fails with OOM and it happens in

couple packages with C++ sources

Dan

[Rich Mattes]

On Wed, Sep 18, 2013 at 4:04 AM, Dan Horák  wrote:
> On Wed, 18 Sep 2013 08:50:52 +0100
> Andrew Haley  wrote:
> > On 09/17/2013 09:27 PM, Rich Mattes wrote:
> > > Do you think I should just add an ExcludeArch to mrpt in f20+ until
> > > this bug is resolved, or is someone able to look into it?  Is there
> > > anything I can try to work around it (try different compiler flags,
> > > etc?)
> >
> > It's always worth trying -O0.
>
> and reduce debuginfo size by changing from "-g" (equals to -g2) to
> "-g1", you can see various tricks to reduce memory consumption in the
> webkitgtk3 spec file [1]

Thanks for the feedback.  Based on this thread and the feedback I got in
the bug report, I replaced -O2 with -O1 to see if I could work around the
crash.  It looks like -O1 was enough to keep gcc from crashing, and the
build completed[1].

I guess the problem is still there though, so I'll keep the bug open.  It
doesn't look like the problem occured in the linker, the command line
provided looks like it's trying to create a .o file from a .cpp source.

Rich

[1] http://koji.fedoraproject.org/koji/taskinfo?taskID=5947644