Packages with optflags and/or debuginfo issues
Till Maas
opensource at till.name
Tue Jan 30 18:34:38 UTC 2007
On Tuesday 30 January 2007 18:48, Ville Skyttä wrote:
> On Tuesday 30 January 2007 16:31, Till Maas wrote:
> > On Monday 29 January 2007 22:06, Ville Skyttä wrote:
> >
> > What about -fomit-frame-pointer and -O3, that maybe breaks debuginfo,
> > too:
> > https://www.redhat.com/archives/fedora-extras-list/2006-July/msg00712.htm
> >l
>
> That would be a good addition to the Wiki page, but: does -O3 break it
> alone, or with -fomit-frame-pointer, or is it just -fomit-frame-pointer?
> What about other flags that could result in the same?
gcc man page:
| -fomit-frame-pointer
| Don’t keep the frame pointer in a register for functions that
| don’t need one. This avoids the instructions to save, set up and
| restore frame pointers; it also makes an extra register available
| in many functions. It also makes debugging impossible on some
| machines.
I don't really know whether or not -O3 breaks it but when I used gentoo I read
that some applications may break when using -O3, but better ask someone who
knows this. :-)
> > > no sources is also quite likely not honoring $RPM_OPT_FLAGS, and if so,
> > > also very likely to have been built without compiler generated security
> >
> > Do $RPM_OPT_FLAGS need to be used in LDFLAGS, too?
>
> I don't think I've ever seen it being used there. FWIW, unlike
> CFLAGS/CXXFLAGS/FFLAGS, %configure doesn't set LDFLAGS at all.
I do not really understand autoconf but it seems to me that LDFLAGS are
created using CFLAGS, e.g. in beryl-core:
| gcc -O2 -g -pipe -Wall -Wp,-D_FORTIFY_SOURCE=2 [...] -o .libs/beryl main.o
| privates.o [...]
line 633 in
http://buildsys.fedoraproject.org/logs/fedora-6-extras/26630-beryl-core-0.1.99.2-1.fc6/i386/build.log
> > What to do then? I once filled a bug about this for qemu and it was
> > closed with WONTFIX:
> >
> > https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=208026
>
> Maybe ask the maintainer to explain why the chosen flags are more important
> than working -debuginfo, and if they really are, why the packager is
> knowingly shipping a useless -debuginfo instead of explicitly disabling it
> (with a comment explaining why) in the specfile? Or forward to an
> appropriate committee to decide if there's no consensus?
It's speed vs security/debugging, afaik applications compiled
with -fomit-frame-pointer are (slightly?) faster that compiled without it and
I guess the security features slow down applications, too. I guess there is
not much to discuss, either speed or security/debugging is more important.
What would be the appropriate committee? fedora-packaging at redhat.com?
Regards,
Till
More information about the Fedora-maintainers
mailing list