More info: Need help, x86_64 build fails on buildsys while make mockbuild on my x86_64 works fine.

[Hans de Goede]

Dan Williams wrote:
> On Tue, 2005-11-15 at 12:39 +0100, Hans de Goede wrote:
>> He all,
>>
>> The subject pretty much says it all, for the logs see:
>> http://buildsys.fedoraproject.org/logs/fedora-development-extras/868-directfb-0.9.24-1.fc5/
> 
> So, I rebuilt directfb with a patch to remove the custom CFLAGS that the
> directfb configure/configure.in adds (-O3 -ffast-math -pipe) and the
> build went through, not sure if that's the cause of the error.  However,
> if you get gcc segfaults, the first thing you should do is to back down
> the optimization levels.  And seriously, if a package is overriding the
> RPM_OPT_FLAGS, then you need to patch it to _not_ do that.  -O2 is
> really the highest optimization people should need use.
> 
> Note that while it built fine, it hasn't gone through due to download
> issues with the buildsystem, possibly iptables problems.  Still working
> on that.
> 

This still doesn't explain why:
-it does build unmodified locally on my x86_64 both with make x86_64 and
  with make mockbuild, esp since the mockbuild should be identical to
  the buildsystem build.
-why it crashes out in different places, if this we're a simple case
  of triggering a gcc bug by using strange optflags, it should always
  crash at the same file.
I really believe hammer2 needs a thorough look.

Also -ffast-math is a quite normal optflag, it offers some IEEE float 
accuracy compatibility in exchange for speed, which for say an imaging 
library is a good thing. Actually for almost anything non-scientific &
non-bookkeeping -ffastmath is a good idea.

Regards,

Hans