mmtk1.1b1 on DEC AlphaStation

Konrad Hinsen hinsen@cnrs-orleans.fr
Wed, 15 Jul 1998 20:38:05 +0200 (DFT)


> I used whatever version of NumPy which was contained in the
> python-for-science.tar.gz as returned by the "extended" Python
> distribution link on the mmtk home page.  I don't see any version

That's 1.1, and thus before the latest DEC patches. I should
update this package...

> numbers in the NumPy *.py files.  Is there an easy way to tell?  I'll

Unfortunately not!

> go ahead and try building NumPy from their most recent source
> distributions (looks like NumPy-1.0b3 and NumPyLib-1.0b2) see if they
> work on the DEC AlphaStation.

No, the most recent distribution is 1.3, available from LLNL
in the package LLNLPython4.tar.gz. All the 'b' releases are
very, very old.

> numbers as I type instead of core dumping.  This lends credence to my
> suspicions that there are some some pointer-related problems with the
> C code on the Alpha.  The Alpha's 64 bit points have caused me more
> than a few problems porting stuff in the past.  If I can resolve the

Me too! In fact, MMTK version 1.0 did not really work on Alphas; with
1.1 I fixed many 64 bit problems.

> One final note.  I enabled NetCDF support, but had to manually copy
> netcdfmodule.h to the installed Python include directory before mmtk
> would compile.  I assume this is a Python problem rather than an mmtk
> problem?

Hmmm. netcdfmodule.h should end up there automatically, if you use the
python_for_science distribution. You must make sure however that
netcdf.h (from the NetCDF distribution) is somewhere where the
compiler can find it. The NetCDF installation procedure normally
copies it to /usr/local/include, but not all compilers look there by
default.

Installation problems are a real mess, and I wish someone would solve
that problem in a portable and general way. As it is, the much
proclaimed software reuse creates an enormous burden on final users,
who have to install dozens of packages before they can do anything.

BTW, what do people on the list think about binary distributions of
Python/MMTK, at least for machines that have some package management
system? The big problem is having all those machines, of course. I
could produce RPM files for Linux and bff files for AIX 4.3. If others
could help out with other system, maybe a nice collection could be set
up.

Konrad.
-- 
-------------------------------------------------------------------------------
Konrad Hinsen                            | E-Mail: hinsen@cnrs-orleans.fr
Centre de Biophysique Moleculaire (CNRS) | Tel.: +33-2.38.25.55.69
Rue Charles Sadron                       | Fax:  +33-2.38.63.15.17
45071 Orleans Cedex 2                    | Deutsch/Esperanto/English/
France                                   | Nederlands/Francais
-------------------------------------------------------------------------------