Re: [matplotlib-devel] wx woes

[John H. <jdh...@ac...>]

>>>>> "cfuller" == cfuller  <cf...@th...> writes:

    cfuller> wx won't start up on my linux machine, running Fedora
    cfuller> Core 2 and a compiled-from-source wxPython 2.5.1. There
    cfuller> was some incompatability with the new GTK libraries,
    cfuller> since the wxPython rpm was comppiled on RH9. I'm guessing
    cfuller> the wx backend is having similar troubles, although I
    cfuller> compiled it from source, naturally. That's a long
    cfuller> compile, BTW!  The wx backend works for me at work, under
    cfuller> Server 2003 and the same version of wx. It still leaks,
    cfuller> although I upgraded to the .60b installation binary from
    cfuller> John's website. Compiling distutil packages under windows
    cfuller> is something I've yet to master, whether with the Borland
    cfuller> compiler or the (not so new) Visual C++ Tooolkit
    cfuller> 2003. Visual Studio just isn't something I can justify,
    cfuller> even if the lab would pay for it. Especially with the VC6
    cfuller> vs VC7 funny business. At least Tk stopped leaking!

Could you send me the script that is leaking?

    cfuller> Since I have a wx that works, its not so bad. I'll still
    cfuller> work on the FC2 issues, but I can see about that toolbar
    cfuller> addition as well. I should get around to installing
    cfuller> windows again at home, I've been putting off an upgrade
    cfuller> to XP, and my old win2k pro install is useless, after a
    cfuller> mobo upgrade.

I believe this is the bug Fernando Perez was writing about which is
specific to recent releases of wx that use private GTK symbols that
are no longer present in the gtk libs in Fedora core 2.  Here is a
snip from an email he sent me earlier on the subject.  As far as I
know, there is nothing we can do about it on the matplotlib side,
except perhaps check the wx/wxpython list to see if it has been fixed
yet in CVS, and agitate for a fix if not.  

Perhaps Fernando or some other knowledgeable wx person can comment on
the appropriate workaround if there is one.

From: Fernando Perez <Fer...@co...>
Subject: Re: matplotlib, ipython and other comments
To: John Hunter <jdh...@ni...>
Date: Wed, 09 Jun 2004 14:29:54 -0600
Organization: Applied Mathematics, University of Colorado at Boulder

...snip ...

> I'll look into this later.  My experience with WX and WXAgg is that
> both work under linux, but WX is a bit slow and buggy.  Is the problem
> you are describing Fedora specific?  (Sorry I can't easily read these
> links now since URL cut-and-paste from my xterm on OSX laptop to my
> browser window doesn't work).

No, the bug is in current WX.  Here's a traceback:

In [6]: import wx
---------------------------------------------------------------------------
ImportError                               Traceback (most recent call last)

/home/fperez/<console>

/usr/lib/python2.3/site-packages/wx/__init__.py
      43 __revision__ = "$Revision: 1.1.2.4 $"[11:-2]
      44
---> 45 from wxPython import wx
      46
      47 _newnames = {}

/usr/lib/python2.3/site-packages/wxPython/__init__.py
      18 # Ensure the main extension module is loaded, in case the add-on modules
      19 # (such as utils,) are used standalone.
---> 20 import wxc
      21
      22 
#----------------------------------------------------------------------------

ImportError: /usr/lib/libwx_gtk2-2.4.so.0: undefined symbol: 
_gtk_accel_group_detach

Apparently the WX guys chose to use _gtk_* symbols which the GTK documentation 
_explicitly warned_ were private and could go away at any time.  Now, in the 
version of GTK shipped with Fedora they _did_ go away, so Wx broke.  This just 
needs to be fixed by the Wx team (maybe it already is in CVS, I'm using Wx as 
shipped with Fedora).  So don't worry about this, it will get fixed in time by 
those responsible.