Re: 10.0-CURRENT #0 r249720: graphics/jasper: eval: _at_CC_at_: not found

On Mon, 2013-04-22 at 21:40 +0200, Dimitry Andric wrote:
> On Apr 22, 2013, at 20:47, O. Hartmann <ohartman_at_zedat.fu-berlin.de> wrote:
> > Trying to update port graphics/jasper on our FreeBSD 10.0-CURRENT boxes
> > (all r249720 or higher) fails on all(!) Core2Duo based systems (Intel
> > E8400 or Q6600) (CLANG 3.3) with the below shown error.
> ...
> > Interestingly, the port does compile on all boxes running the very same
> > FreeBSD 10.0-CURRENT with either Sandy-Bridge-E (i7-3939K) or Ivy Bridge
> > (i3-3220) CPUs. All systems have in common that I compile world and
> > kernel with option -O3 (CLANG 3.3), -march=native.
> ...
> > I guess I hit a miscompilation with either option -O3 -march=native on
> > "Penryn" architectures with the new CLANG 3.3 compiler?
> > 
> > (By the way, I also see this obscure "_at_CC_at_: not found" error in
> > compiling games/qstat and it hits also the Core2Duo boxes, but not any
> > CPU type beyond the C2D).
> 
> This is most likely the same issue you reported in another thread, here:
> 
> http://docs.freebsd.org/cgi/mid.cgi?1365877246.2093.20.camel
> 
> The configure script tries to use sed to replace _at_CC_at_ with a build-time setting, but the sed replacement fails, because of a problem in the regex functions in libc, when libc is compiled with -O3 or higher.  The exact cause of the problem is still unknown (likely a bug in the vectorizer, or less likely a bug in libc).
> 
> Please recompile and reinstall at least your libc with -O2 instead of -O3,  and retry the ports build afterwards.

You're right.
After decreasing from -O3 to -O2 the "problem" on C2D machines went
away.

> 
> In general, if you use non-standard optimization settings, always remove the customizations first, before reporting problems.

I should do, indeed, but isn't it interesting of the knowledge, that
this happens only on Core2Duo architecure CPUs, but not on more modern
designs?

Regards,

Oliver