[Debichem-devel] Bug#792051: chemps2: FTBFS on kFreeBSD: BLAS API not found.

Sebastian Wouters sebastianwouters at gmail.com
Sat Jul 11 22:04:14 UTC 2015


Hi Aaron,

I can add the dependency, but is there an easy way to test beforehand
whether it will help?

The library itself should in principle be sufficient as the c-functions are
declared external in the chemps2 source code itself:
https://github.com/SebWouters/CheMPS2/blob/master/CheMPS2/include/chemps2/Lapack.h

It also seems that the installation paths are identical:
https://packages.debian.org/sid/amd64/libatlas3-base/filelist
https://packages.debian.org/sid/kfreebsd-amd64/libatlas3-base/filelist
https://packages.debian.org/sid/kfreebsd-i386/libatlas3-base/filelist
So I don't understand why adding libatlas-base-dev would help. Are there
precendents of this issue / this fix for this issue?

It seems rather strange that FindBLAS.cmake doesn't find it, because it's
the same CMake package for amd64/i386 and kfreebsd-amd64/kfreebsd-i386. I
just don't understand yet how this bug can happen. But if only the
BLAS/LAPACK library is needed, and not the headers, I think this is
actually a bug of CMake and not of chemps2?

That being said, unless you strongly believe that switching to
libatlas-base-dev would help, there are known workarounds to fix it for
CMake / chemps2, e.g. passing the LAPACK libraries with
-DLAPACK_LIBRARIES=/usr/.... at the configuring stage:
https://github.com/SebWouters/CheMPS2/blob/master/CMakeLists.txt (lines
73-77). If Michael can print the content of that variable for sid just
after line 77 with

message(STATUS "LAPACK libs = ${LAPACK_LIBRARIES}")

then I (or Michael) can adjust the rules file accordingly. We can filter
DEB_HOST_MULTIARCH for kfreebsd ( which should be OK:
https://packages.debian.org/sid/kfreebsd-amd64/libhdf5-dev/filelist ;
https://packages.debian.org/sid/kfreebsd-i386/libhdf5-dev/filelist ) and if
it's in
DEB_HOST_MULTIARCH, add the line -DLAPACK_LIBRARIES=/usr/.... to the
configure specifications. What do you think?

Best wishes,
Sebastian
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.alioth.debian.org/pipermail/debichem-devel/attachments/20150711/0d677058/attachment-0003.html>


More information about the Debichem-devel mailing list