[Matplotlib-users] Gtk (and specifically matplotlib) support in ipython

[Fernando P. <Fer...@co...>]

Hi all,

after help and discussions with John, I'm putting up AS ALPHA CODE, a small 
script which will load ipython in a multithreaded mode which allows it to run 
GTK gui apps, and in particular matplotlib with the GTK backends.

Please note that using this requires at the moment _both_ CVS ipython and CVS 
matplotlib, since John and I have worked in conjunction to make the necessary 
changes.

The code already works fairly well, with one significant caveat: any process 
started via os.system() (whether directly or implicitly using !cmd) will hang 
after a while.  I have currently no solution for this, but will welcome any 
ideas.  Note that the problem seems to only affect long-running GUI apps, 
simple '!grep foo *' type things appear to function just fine.  But something 
like '!gv foo.eps &' will open gv, and after about 30 seconds the gv instance 
stops responding to all keyboard/mouse input.

Debugging multithreaded code is _very_ tricky, but this tool may prove useful 
to many who want the interactive convenience of ipython combined with the 
better matplotlib backends, or other GTK apps.  All feedback from willing 
testers will be very welcome.

Note that those wanting to use matplotlib interactively, and willng to stick 
to the Tk backends, can already do so.  Tk does not require manual threading 
control, so it works perfectly OK with normal ipython.  John already posted in 
the past a 'pylab' ipython profile which loads and configures matplotlib for 
such use, I can repost it if necessary.

This will become part of ipython 0.6.3, but I want to try to resolve the 
os.system() deadlock first.  Since both John and I are complete ignoramuses on 
threading issues, help from anyone with knowledge on the topic is most welcome.

If you don't want to update to CVS ipython/matplotlib but think you may have 
an idea, you can use this small shell for tests:

http://amath.colorado.edu/faculty/fperez/tmp/pyint-gtk.py

It provides a basic multithreaded shell as a standalone python script (no 
matplotlib or ipython needed), and will show the os.system() deadlock problem.

Many thanks in advance for any feedback, and to John for all the help on this. 
  We (I assume I speak for both John and I here :) hope you all find it useful.

Best,

f