Re: [Matplotlib-users] FigureCanvasQTAgg clean up

[Mads I. <mad...@gm...>]

On 01/30/2013 08:02 PM, Benjamin Root wrote:
>
>
> On Wed, Jan 30, 2013 at 1:46 PM, Mads Ipsen <mad...@gm... 
> <mailto:mad...@gm...>> wrote:
>
>     On 01/30/2013 05:03 PM, Benjamin Root wrote:
>>
>>
>>     On Tue, Jan 29, 2013 at 5:12 PM, Mads Ipsen <mad...@gm...
>>     <mailto:mad...@gm...>> wrote:
>>
>>         Hi,
>>
>>         I spend some time writing up the question below on
>>         Stackoverflow which immediately was closed as a duplicate of
>>         other posts. To my best knowledge, these posts did not answer
>>         my questions - so I'll try my luck here instead:
>>
>>         I am using the qt backengine for setting up a QWidget that
>>         embeds a matplotlib scene. When the widget is closed it
>>         appears that many of the matplotlib objects that were part of
>>         the plot still are alive in Python space.
>>
>>         My question is basically the following: What actions should I
>>         take to clean up the figure and axes objects etc. that were
>>         part of the plot widget? The qt backend comes with a figure
>>         manager, but it appears a little unclear how it should be used.
>>
>>         I have attached a small unit test example that sets up a
>>         plot. When the plot appears, just close it, and the test will
>>         garbage collects the plot, and then display info of the
>>         matplotlib objects that are still alive in Python space.
>>         Clearly both the path of the plot, and several Bbox objects
>>         are still referenced.
>>
>>         Our current unit test suite contains almost 10000 GUI tests
>>         and its imperative that proper object space clean up is done
>>         after each test. Any help is much appreciated.
>>
>>         Best regards,
>>
>>         Mads
>>
>>
>>     Would "fig.clf()" do what you need?
>>
>>     Ben Root
>>
>     Thanks for the feedback.
>
>     The trick is to get this done automatically when the widget is
>     closed. If you look in backend_qt4.py in the constructor of
>     FigureCanvasQT you'll find
>
>       QtCore.QObject.connect(self, QtCore.SIGNAL('destroyed()'),
>     self.close_event)
>
>     which should perform the steps below when close_event is called
>
>       event = CloseEvent(s, self, guiEvent=guiEvent)
>       self.callbacks.process(s, event)
>
>     If I insert print statements, the signal is ignored with
>     matplotlib 1.1.0, but called with 1.2.0 (which uses a lambda
>     function in the above connection).But inspecting pythons objects
>     after the close() method is called on the widget, it seems that -
>     at least - the paths associated with plot actually still exist.
>
>     So I'm just a bit worried if there already is existing
>     functionality in the FigureCanvasQTAgg object that can do these
>     things for me. And if so, how to use them?
>
>     Best regards,
>
>     Mads
>
>
> Actually, now I am starting to recall a bug fix that came shortly 
> after the 1.2.0 release related to properly setting up the destruction 
> signal of the QT figure object.  Could you test your code-base with 
> the maintenance branch (1.2.x)? or master?
>
> Cheers!
> Ben Root
>
Hi,

I attach log files from running my setup 'plot.py' with 1.1.0, 1.2.0, 
and master. To my judgement - if an automated memory clean-up is handled 
by the FigureCanvasQTAgg class - I must be abusing the usage of the 
class, since all logs still show the presence of paths, labels, bbox's, etc.

Also notice that log-1.2.0 show the presence of a path with hundreds of 
data points, which are not present in the other logs.

Best regards,

Mads


-- 
+-----------------------------------------------------+
| Mads Ipsen                                          |
+----------------------+------------------------------+
| Gåsebæksvej 7, 4. tv |                              |
| DK-2500 Valby        | phone:          +45-29716388 |
| Denmark              | email:  mad...@gm... |
+----------------------+------------------------------+