[Plplot-devel] Current status of the python front-end of plplot

[Alan W. I. <ir...@be...>]

Earlier today I committed to CVS the new swig-based python interface that
has been kindly donated by Gary Bishop.

I have tested this new interface with pythondemos.py with both the single-
and double-precision version of the plplot library, and it seems to work
fine for all the xw??.py examples that are exercised by pythondemos.py. That
is a welcome change since the number-one issue for most would-be python
plplot users had been errors created when the old python interface which was
only compatible with double precision was inadvertently used with the
single-precision version of the plplot library.

Implementation details

swig version 1.3.11 or later is required for those using cvs (but not
tarball release users, see discussion below).  1.3.11 is the Debian woody
version you get with the swig1.3 package.  If you have no swig or just
swig-1.1.x you should download swig-1.3.13 from
http://www.swig.org/download.html.  To distinguish between double and single
precision, swig needs to be run with a command-line switch.  This is all
taken care of by the new configuration. The point is that
plplotcmodule.c_single and plplotcmodule.c_double are both produced at build
time, but plplotcmodule.c is symlinked to only one of those in the
configuration step depending on the --with-double configuration option.

I have gone with this approach since I don't know in advance whether our
tarball release users want to build the single or double-precision version
of the plplot library so I will provide both plplotcmodule.c_single and
plplotcmodule.c_double when I create a tarball release.  In this way we
won't be demanding all our tarball release users must have swig, and this
approach is also consistent with our prior decisions about some of our other
generated files such as configure.

swig-1.3.11 still has some issues with documentation strings so part of the
process of creating plplotcmodule.c_single and plplotcmodule.c_double
requires a python script (makedocstrings.py) to take care of the problem.
This is all handled automatically by the Makefile.

The Makefile automatically creates both the plplot_widget and plplotc
extension modules.  The former includes useful widget support functions, and
the latter makes the complete raw common C API of the plplot library
available to python (with redundant dimension information automatically
dropped from the argument lists).  Finally, there is also a new python
module called plplot (implemented with plplot.py) which is a user-friendly
python wrapper for the plplotc module.  This user-friendly version puts in
required python argument list variations for plcont, etc.

Note that the python modules have been renamed from pl and pyqt_pl to plplot
and plplot_widget.  There is a long tradition in python of having meaningful
module names so that is why I wanted plplot in both names.  Also, I replaced
pyqt with widget (in different order) because the widget support functions
in that module can be used not only by pyqt but also apparently by the
Tkinter widget. I decided on this new naming after some private discussion
with Geoffrey.  The change will undoubtedly be annoying to our python users,
but I thought so much else was being changed at this time in the way the
plplot modules are generated, that it was a good opportunity to change the
names as well.

There are a lot of minor issues I still have to clean up such as completing
the documentation strings, implementing some more user-friendly variations
in plplot.py, making sure the installed versions (as opposed to test
versions in plplot/tmp) work, etc.  Also, I have left
plplot/sys/win32/msdev/ completely untouched so there is undoubtedly work to
do there to adjust to the new module names and swig generation approach.
However, if Olof et al have any trouble with this they should consult with
Gary because he got all this working (although without name changes yet, and
apparently handling things somewhat differently than in
plplot/sys/win32/msdev) for his own windows environment before he donated
the software to us.

Despite these minor issues I have discussed, the Unix/Linux/MacOS X build of
the python modules in plplot/tmp should be in good shape both for single and
double precision, and I invite you to give it a test drive.

Alan

email: ir...@be...
phone: 250-727-2902	FAX: 250-721-7715
snail-mail:
Dr. Alan W. Irwin
Department of Physics and Astronomy,
University of Victoria, P.O. Box 3055,
Victoria, British Columbia, Canada, V8W 3P6
__________________________

Linux-powered astrophysics
__________________________