Fix initialization problem with useblit on SpanSelector instance creatio...

[cimarronm]

...n.

[cimarronm]

This fixes an initialization problem where useblit variable is used in function new_axes before it is initialized as seen below

  File "/Library/Frameworks/Python.framework/Versions/3.3/lib/python3.3/site-packages/matplotlib-1.3.x-py3.3-macosx-10.6-intel.egg/matplotlib/widgets.py", line 1077, in __init__
    self.new_axes(ax)
  File "/Library/Frameworks/Python.framework/Versions/3.3/lib/python3.3/site-packages/matplotlib-1.3.x-py3.3-macosx-10.6-intel.egg/matplotlib/widgets.py", line 1104, in new_axes
    if not self.useblit:
AttributeError: 'SpanSelector' object has no attribute 'useblit'

This sets the instance useblit to what was given at instance initialization and then disables it if the new canvas setting does not support blit in the new_axis function. I believe this was the intention in commit dd325759

[mdboom]

👍

[pelson]

Ah, I've just merged #1982 which I think was an alternative fix. @efiring & @cimarronm - is there a crossover here which we need to resolve? Otherwise, I think this PR can be closed.

Cheers,

[efiring]

@pelson, @cimarronm: Sorry, I didn't even think to scan the existing PRs and issues before generating a PR, so I completely missed this. I considered exactly the solution used in this PR, but decided on the alternative in #1982 because it seems simpler to just set the useblit on initialization and be done with it. Are there any circumstances under which this would be incorrect? Is it possible for a SpanSelector to be initialized with one canvas (via its ax argument) and then to have its new_axes() method called with a canvas from a different backend, with a different value of supports_blit? I couldn't imagine it; but if I'm wrong, then @cimarronm's solution is correct and mine is not.

[mdboom]

@cimarronm, @pelson, @efiring: I'm fine with @efiring's simpler solution for the same reason he states: I'm having trouble imagining a case where it matters. However, @cimarronm, if you have a case where this is superior, feel free to reopen and ping us.

Thanks for the contribution, and keep em coming!

[cimarronm]

@efiring, @mdboom, @pelson: I agree. @efiring's patch should work just fine as the backend shouldn't change so I am fine with it.