cd /builddir/build/BUILD/ATLAS/ppc64_base ; ./xprobe_comp -v 0 -o atlconf.txt -O 1 -A 7 -Si nof77 0 -V 6 -Fa ic ' -g -Wa,--noexecstack -fPIC' -Fa sm ' -g -Wa,--noexecstack -fPIC' -Fa dm ' -g -Wa,--noexecstack -fPIC' -Fa sk ' -g -Wa,--noexecstack -fPIC' -Fa dk ' -g -Wa,--noexecstack -fPIC' -Fa xc ' -g -Wa,--noexecstack -fPIC' -Fa gc ' -g -Wa,--noexecstack -fPIC' -Fa if ' -g -Wa,--noexecstack -fPIC' -b 64 -d b /builddir/build/BUILD/ATLAS/ppc64_base > config1.out
make[2]: xlf: Command not found
make[2]: *** [IRunF77Comp] Error 127
===
extract from http://ppc.koji.fedoraproject.org/kojifiles/packages/atlas/3.10.1/12.fc21/data/logs/ppc64/build.log
The suggested patch in attachment is to use gfortan in place of xlf
Michel Normand
I still don't see the need for this patch. It is perfectly normal for ATLAS to try for a preferred compiler (in this case xlf), and then use a back-up compiler (in this case, gfortran) when it fails.
The report you sent seems to be successful, so what is the problem? Are the libraries not usuable, or are the fortran libs not built, or is gfortran not used eventually (I believe config should first try xlf, and when that fails, use gfortran).
?
Clint
I only wanted to avoid the two reported make error that lets assume there was a problem. May be there is a way to suffix with (ignored) as already done for other lines.
make[2]: xlf: Command not found
make[2]: *** [IRunF77Comp] Error 127
===
Please answer this question:
Did the original unpatched install produce good libraries, and use gfortran automatically or not?
I did not tried to use the generated atlas binaries/libraries.
At least the make check passed with some ignored errors and
atlas rpms have been generated as per previously reported build.log
I only noticed in this ticket that the build log has a specific
make[2]: *** [IRunF77Comp] Error 127
But I do not know enough the internals of atlas make files
to determine the consequences of this failure.
--
Michel Normand
OK, then there is no need for this patch, since configure is behaving exactly as I want. This patch would remove the ability to use xlf, which I don't want to do.
Thanks,
Clint