[SciPy-dev] gui_thread and chaco under Linux

[Prabhu Ramachandran]

>>>>> "PP" == Pearu Peterson <pearu at scipy.org> writes:

[snip]
    >> wrapped_class = gui_thread.register(some_wx_python_class)
    >>
    >> Then you create wrapped_class instances as you would create
    >> some_wx_python_class and it should work (in theory).

    PP> This will not work for chaco anymore because chaco executes
    PP> wxPython functions during chaco.wxplt import.

OK. :(

    >> I can't test this because I cannot run chaco anymore.  kiva no
    >> longer builds under gcc 2.95.4. :(

    PP> Yes. In fact, this triggered me to upgrade gcc on my debian
    PP> system.  It turned out to be quite painless, now I have
    PP>   gcc-2.95 gcc-3.2 gcc-3.3
    PP> nicely co-existing on my system.  You could try the same ^K
    PP> ;-)

Well, I'm running Woody (stable) so all I have is gcc-3.0 which I've
heard has bugs.  testing and unstable are too unstable for me at the
moment.

    >> Admittedly, the gui_thread approach is a hack and has
    >> limitations but it used to work pretty OK.  Eric was talking
    >> about a better approach that IIRC is similar to your approach
    >> (1).  As you notice its a non-trivial job and AFAIK no one has
    >> yet had a stab at it.

    PP> I took this challenge and here are intermediate results:

    >>>> from parallel_exec import ParallelExec pexec=ParallelExec() #
[snip]
    >>>> app.shutdown()

    PP> that is, I can already use chaco from my Linux prompt! And I
    PP> have made no changes to chaco for that. The next step is to
    PP> make

Nice.  This looks to be similar to the code written for Gtk by
someone.  IIRC it was available on ASPN.

    >>>> from chaco.wxplt import * plot([1,2])

    PP> safe...

I'd think you'd need to modify or wrap around wxPython itself to get
this to work that transparently or you'd need to do something similar
to gui_thread.  

Actually it looks like this could be easy to hack into wxPython.  Most
of wxPython's work is done inside of a call to apply.  So if we
quietly replace apply inside wxPython's namespace with a massaged
apply that calls the real apply in the right thread, I think we should
be all set.  This is a hack but if its this easy I guess Robin could
be convinced to add cleaner support for it.

  orig_apply = apply
  def apply(o, *args, **kwargs):
      pexec('orig_apply(o, *args, **kwargs)')


I don't know if this will work with pexec the way its implemented
currently but I think you get what I'm trying to say here.

cheers,
prabhu