Open zimoun opened 8 years ago
Thanks for this contribution. I merged it in 25096c6. Hopefully this will help people get around confusing HDF5-related errors in the build process.
Alternatively, specifying the --without-hdf5
option to configure
will remove HDF5 from the build process entirely. This not particularly disabling to the code---the only reason for including HDF5 support is to allow the export of internal matrices and vectors for use by algorithm developers, so users who are just interested in solving scattering problems will generally not need HDF5 support.
this is relative, and hope will close, at least part of, bugs #66 #76.
These issues come from how
configure.ac
detects HDF5 header files and library. Especially, Debian does not store the header filehdf5.h
in/usr/include
but in/usr/include/hdf5/serial
. And the library is somewhere, e.g.,/usr/lib/x86_64-linux-gnu/hdf5/serial
. Moreover, the option--with-hdf5=<dir>
fixes then the flags at:CPPFLAGS=<dir>/include
andLDFLAGS=<dir>/lib
. However, this does not help to inform the configuration system to the both right locations.There are different ways, more or less flexible, to fix that:
configure.ac
(do not know exactly when
scuff-em
requires HDF5: compile time or linking time, etc. that's why it is maybe redundant)Everything compiles, just some warnings e.g. in
machcon.c
,UserDefinedMaterials.cc
,matheval.c
,pcubature.c
. And then,make check
fails. I do not know if the fail comes from wrong HDF5 locations (at run-time) or not. I can open another bug report about that. Just, in short:No idea yet if it is expected or not...
configure.ac
by adding 2 options:--with-hdf5-includedir
and--with-hdf5-libdir
. See PullRequest #90 (TravisCI failed because the origin commit failed too, see build #24, anyway) . Then for example,and there is some warnings. Then,
make check
fails too.Hope that help