pymex

pymex embeds the Python runtime (currently only works for version 2.7) into the MATLAB environment, using the MEX API. This allows MATLAB programs to leverage Python-based functionality by evaluating, importing and manipulating Python modules.

Currently, pymex should be thought of as alpha, as it may cause MATLAB to crash hard. Such crashes are most likely due to missing or excess Py_XDECREF calls, as such errors currently propagate outward as segfaults.

Note that pymex also is 64-bit only at the moment, due to hardcoding the size of Python pointers for marshalling to MATLAB data structures. This limitation can be removed, but is currently not a high priority.

Installation Instructions

The following instructions courtesy of Ian Hincks.

Linux/Unix

1. Make sure python-dev or the equivalent is installed on your computer. You will need Python.h.

2. Make sure g++ is installed. On Ubuntu:

   $ sudo apt-get install g++

3. Run rebuild_pymex.m in MATLAB.

Windows

1. Make sure you have Python 2.7 (64-bit) installed. Enthought Python Distribution 7.3 has been tested, but other distributions may also work.

2. Install Windows SDK 7.1 or any other compiler supported by MEX. You can check that you have setup MEX correctly by running:

   >> mex -setup
   

3. Run rebuild_pymex.m in MATLAB.

   >> rebuild_pymex

Known Issues

All Platforms

• There is a huge memory leak when boxing mxArray values as Python objects, rather than converting them. This is because there is currently no function written to unallocate the persistent array copied from the input to box_mxarray.

Linux

• py_import numpy fails due to the version of libgfortan.so.3.0.0 bundled with MATLAB conflicting with that used by NumPy. Can be fixed by symlinking the system libgfortran.so into MATLAB's installation dir. (Documentation forthcoming.)

Windows

• If built against Enthought Canopy, MATLAB will crash when any pymex function is called, due to a fatal error inside Py_Initialize() causing Python to call exit().