Providing a Java encapsulation of these tools would be nice, especially Fortran. We also already use NestedVM for SQLite support, through this project: http://www.zentus.com/sqlitejdbc/

NestedVM on its own doesn't complete the chain. We still need the basic type support for multidimensional arrays. Ideally, by using Unidata we get an array type that then works with their support for NetCDF/HDF. Also, there's some overhead in using NestedVM.

But please try your experiments too. I personally plan to start looking at NumPy on Jython in earnest at or after PyCon. We need to get our 2.5 release out first!

- Jim

On Wed, Jan 21, 2009 at 3:05 PM, Arlindo da Silva <dasilva@alum.mit.edu> wrote:
On Mon, Dec 1, 2008 at 12:01 PM, Jim Baker <jbaker@zyasoft.com> wrote:
SciPy is not ported. We plan to work on NumPy first, once we release 2.5 very soon (in early 2009). Any such NumPy work will take advantage of optimizations we are planning for the advanced compiler and will out of the box support backing large arrays with NetCDF (and unifed with HDF and DAP), since we plan to use the Unidata project as our base (http://www.unidata.ucar.edu/). That could be quite interesting, I think.


Has anybody considered using NextedVM as an alternative to implementing Numpy and eventually Matplotlib/SciPy?

        http://nestedvm.ibex.org/

(In a nutshell, NextedVM compiles C code into Java byte code, using MIPS as an intermediate step.) I have had good luck compiling, NetCDF, HDF, cURL, and a whole bunch of other libraries into Java.  For example,

                   http://opengrads.org/devel/java/netcdf-vm.jar

In particular, the current NestedVM gcc 3.x tool chain allows you can compile Fortran 77 into Java byte code with g77; updating NestedVM to gcc 4.x may give us gfortran.  I am wondering if one could get f2py implemented this way.

           Arlindo

--
Arlindo da Silva
dasilva@alum.mit.edu



--
Jim Baker
jbaker@zyasoft.com