[Gnuplot-py-users] Changes in CVS and backwards compatibility

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

Hi all,

today I got a report from an ipython user who saw some breakage of IPython's 
gnuplot support.  He thought he was using Gnuplot.py 1.6, but it turned out he 
had a CVS-updated copy.  I have a few questions:

- Is the __version__ in CVS still 1.6?  If so, could it be tagged to reflect 
that this is not 'stock' 1.6 anymore?  It would help in pinning down these 
problems, since I know that IPython lives happily with Gnuplot.py 1.5 and 1.6.

- The reported problem generated the following traceback:

/home/python/PYTHON/lib/python2.3/site-packages/IPython/Gnuplot2.py
      43
      44 # Fix the File class to add the 'index' option
---> 45 class File(Gnuplot_ori.File):
      46
      47     _option_list = Gnuplot_ori.File._option_list.copy()

/home/python/PYTHON/lib/python2.3/site-packages/IPython/Gnuplot2.py in
File()
      45 class File(Gnuplot_ori.File):
      46
---> 47     _option_list = Gnuplot_ori.File._option_list.copy()
      48     _option_list.update({
      49         'index' : lambda self, index:
self.set_option_index(index),

AttributeError: 'function' object has no attribute '_option_list'


Is this change fundamental, or merely accidental?  If it is at all reasonable, 
I was wondering if the current evolution of Gnuplot.py could maintain 
backwards compatibility for the sake of users who have old code (such as 
IPython, in fairly wide deployment already through debian, gentoo & Fink).  I 
can always update ipython for a new release to protect against this, but that 
won't help much existing users if they update their Gnuplot.py and their 
existing code breaks (which is already happening).

I understand perfectly that sometimes, backwards compatibility just has to be 
broken in order to advance with a proper design.  If that is the case here, 
would it be possible to label the new Gnuplot.py version 2.0 or something like 
that?  This would be a strong indicator of 'significant changes', and I could 
either that ipython only works with versions 1.x.  Perhaps (since I know many 
of the new changes supercede my ipython hacks), I can even make my 
ipython-gnuplot layer be much thinner in the presence of 2.0.

Thanks in advance for any feedback.

Regards,

Fernando.