[Debichem-devel] Bug#767498: Bug#767498: aces3: FTBFS on several architectures

[Michael Banck]

On Sat, Nov 01, 2014 at 01:01:42PM +0100, Michael Banck wrote:
> So I will look about patching it out in the build system, rather, which
> should help the other buildds as well.

Unfortunately, the F12 technology appears to be coded in pretty deeply.
I had hoped that due to the way ACESIII is coded, it would be restricted
to some specific SIAL (some bytecode language) codes which could be
ommitted.  However, it is also coupled to the integral routines, which
are at the core of the system.  So patching it out is unfortunately
unfeasable.

The next thing I noticed is that due to a bug or undocumented feature in
the Makefiles, $(FFLAGS) is not used when compiling that code.  As an
aside "that code" is also the only Fortran90 code in the project.  For
those source files, $(FFLAGS1) is used in the Makefile, which is never
defined, so no compiler flags at all are passed to make.  That means
that those files already get compiled without optimization.  So far, I
could not get it to link successfully on armhf with providing different
compiler options, I've asked in #debian-arm now.

The way forward appears to be to give-back aces3 and hope that it builds
at least on s390x (and possibly mips*, if it gets queued on a beefier
buildd), and then get the binary packages of the non-building arches
removed in time for the freeze.

Alas, it turns out -lmpi_cxx as added to GNULIBS in debian/rules will
make it FTBFS on s390x.  That's a shame, cause AFAICT after some
investigation it's superfluous anyway (I copied it from INSTALL).  Maybe
that could be fixed post-freeze by a small patch, while removing the
s390x binary from unstable for now.


Michael