On 10/09/2013 03:45 PM, Rhys Ulerich wrote: >> I will take a stab at this and report back. It may be a couple of days >> as I don't personally have access to the system. > > I'm all thumbs at this. I do see the following suspect differences at > the tail of that config.log.trimmed I attached previously: > > NEXT_ERRNO_H='' > NEXT_FCNTL_H='"///bgsys/drivers/V1R4M2_200_2010-100508P/ppc/gnu-linux/lib/gcc/powerpc-bgp-linux/4.1.2/../../../../powerpc-bgp-linux/sys-include/bits/fcntl.h"' > NEXT_FLOAT_H='""' > ... > NEXT_SYSEXITS_H='""' > > You'll notice that NEXT_FLOAT_H has a quoted empty string instead of > an empty string. On this system, gnulib/float.h is produced but > gnulib/sysexit.h is not.
Your config log says you are using CC=bgxlc_r. Can you collect the following output: $ echo '#include <fcntl.h>' > foo.c $ bgxlc_r -E foo.c > fcntl.out $ echo '#include <float.h>' > foo.c $ bgxlc_r -E foo.c > float.out as well as a copy of your system's <float.h>? My guess is that your float.h is macro-only, and therefore gives us nothing to go by to tell where it lives. We already special-case AIX, using a different mode of '$CC -C -E' to make the preprocessed output more verbose; is there some compiler switch for bgxlc_r (maybe -C, maybe some other spelling) that makes the output more verbose? Basically, we are looking for some way of inspecting the preprocessor output to reverse-engineer the path to the file that actually got included. -- Eric Blake eblake redhat com +1-919-301-3266 Libvirt virtualization library http://libvirt.org
signature.asc
Description: OpenPGP digital signature
