Build error on Debian lenny amd64
Build error on Debian lenny amd64Posted by Adam Powell at December 21. 2007Greetings,
I'm building on Debian lenny (testing) amd64, and got an error so big I'm going to link it separately: http://lyre.mit.edu/~powell/salome-error
I'm afraid I don't know C++ well enough to weed through the 115k of error messages to get to the crux of it.
Thanks in advance for any help you can provide,
Re: Build error on Debian lenny amd64Posted by Adam Powell at January 01. 2008The crux of the problem appears to be bad gcc-openmpi interaction. A simple C++ program consisting of:
will fail in the same way using openmpi. I've sent a message to firstname.lastname@example.org to look into this.
Meanwhile, I've uncovered a bug in the configure process. HDF5 is often linked to MPI, so testing for HDF5 without using any MPI includes and libs is bound to fail. Instead, MPI should be tested before HDF5, and if found, the MPI includes and libs should be used when testing HDF5.
Salomé team, are you interested in investigating a fix, so this works with MPI-linked HDF5? I'd be glad to produce a patch to configure.ac and probably the HDF5 .m4 file.
Re: Build error on Debian lenny amd64Posted by Adam Powell at January 02. 2008I think I've fixed the HDF5-OpenMPI issue with the following one-line patch:
--- /usr/include/H5public.h.old 2008-01-01 16:55:04.000000000 -0500
+++ /usr/include/H5public.h 2008-01-01 16:53:37.000000000 -0500
@@ -50,6 +50,7 @@
+# define OMPI_SKIP_MPICXX
#ifndef MPI_FILE_NULL /*MPIO may be defined in mpi.h already */
I'm planning to suggest this for inclusion in the Debian hdf5 package.
Still interested in whether the Salomé team is interested in getting HDF5-MPI interaction working right... In the meantime, the workaround is to include MPI includes in CFLAGS and friends in the configure command line.
Powered by Ploneboard