Re: [matplotlib-devel] 1.1.1rc RuntimeError: underlying C/C++ object has been deleted

[Benjamin R. <ben...@ou...>]

On Wed, Mar 28, 2012 at 2:36 PM, John Hunter <jd...@gm...> wrote:

>
>
> On Wed, Mar 28, 2012 at 1:22 PM, Neal Becker <ndb...@gm...> wrote:
>
>>
>> > import matplotlib
>> > matplotlib.use('pdf')
>> >
>> > It looks like you are inadvertently importing the qt library in a
>> headless
>> > script run.
>> >
>> > JDH
>>
>> Thanks, but should that cause a (scary looking) error?  Or is there a real
>> problem?
>>
>>
>> It's just a clean up error in the qt destructors I  think.  You are
> basically in an unsupported use case: using a gui backend, but not raising
> the figures with show, so our initialization code doesn't get run properly,
> which means the clean up may not be properly configured.  Our work is hard
> enough supporting all the GUI toolkits across multiple operating systems --
> I don't know that we want to get into trying to support *unsupported* use
> cases.
>
> JDH
>
>
True that this is probably an unsupported use-case, but I don't see it as a
totally unreasonable one.  For example, in many of my scripts in one of my
projects, I have command-line options to determine if I am going to save
the figure to a file and another option to determine if I am going to show
the figure as well.  I could run these scripts with or without showing a
figure.  By the time I process the command-line arguments, I have already
imported matplotlib.  Of course, it isn't very difficult to recode it to
import and select a backend according to the command-line arguments, it
would be a more convoluted script that way.

Note, I have not noticed any issues with this approach while using TkAgg
and GTKAgg.  I guess it could probably be made into a low-priority TODO
item in PyQt4 to see if the destructor is being over-zealous.  Neal, your
best bet would be to file a wishlist item and tag it as such.  Again, low
priority, but maybe someone will notice something obvious at a later time.

Cheers!
Ben Root