[Matplotlib-users] matplotlib.backends.backend_gtk3cairo memory leak

[David]

Hi, I seem to have a memory leak while generating a 'live' plot display. This
wasn't the case for GTK2, but the example below is consuming ~800k/second
(Matplotlib 1.4.3, PyGI aio-3.14.0_rev18, Windows 7 x64, python 3.4.3). I
have checked the garbage collector but it doesn't show anything interesting
(no massive incrementing count of uncollected items). Anyway, I would be
very grateful if somebody could confirm and/or fix this (or tell me what I'm
doing wrong).Many thanksDavidCode below:
from gi.repository import Gtk, Gdk, GLibfrom matplotlib.figure import
Figure# Tell matplotlib to use a GTK canvas for drawing#from
matplotlib.backends.backend_gtk3agg import FigureCanvasGTK3Agg as
FigureCanvasfrom matplotlib.backends.backend_gtk3cairo import
FigureCanvasGTK3Cairo as FigureCanvas# Application Classclass
pyMatPlotLibTest(object): def update_gui(self): y = [self.index] *
1024 self.index += 1 if self.index > 1024: self.index = 0
Gdk.threads_enter() self.line.set_ydata(y)
self.axes.set_title("%d" % self.index) self.canvas.draw()
Gdk.threads_leave() return True def __init__(self):
self.index = 0 self.x = range(1024) # Initialise the
threads system and allow threads to work with GTK GLib.threads_init()
# Draw scope self.figure = Figure(dpi=100) self.canvas =
FigureCanvas(self.figure) # a Gtk.DrawingArea
#self.widget.alignment_ScopeDisplay.add(self.canvas) # Draw initial
scope self.axes = self.figure.add_subplot(111) self.line, =
self.axes.plot(self.x, [0]* 1024) self.axes.set_title("None")
self.axes.set_xbound(0.0, 1024) self.axes.set_ybound(-16, 1040)
self.window_main = Gtk.Window(title="pyMatPlotLibTest")
self.window_main.connect("destroy", lambda x: Gtk.main_quit())
self.window_main.add(self.canvas) self.window_main.show_all()
# Ticker for the update of the input state monitoring
Gdk.threads_add_timeout(priority = GLib.PRIORITY_DEFAULT_IDLE,
interval = 10, # msec function =
self.update_gui) Gtk.main()if __name__ == "__main__": gui =
pyMatPlotLibTest()




--
View this message in context: http://matplotlib.1069221.n5.nabble.com/matplotlib-backends-backend-gtk3cairo-memory-leak-tp45614.html
Sent from the matplotlib - users mailing list archive at Nabble.com.

[Benjamin Root]

I take it that it doesn't happen using the GTK3Agg backend? What about the
threading portion? Does it happen if you take the threading out?

Ben Root

Post by David
Hi, I seem to have a memory leak while generating a 'live' plot display.
This wasn't the case for GTK2, but the example below is consuming
~800k/second (Matplotlib 1.4.3, PyGI aio-3.14.0_rev18, Windows 7 x64,
python 3.4.3). I have checked the garbage collector but it doesn't show
anything interesting (no massive incrementing count of uncollected items).
Anyway, I would be very grateful if somebody could confirm and/or fix this
from gi.repository import Gtk, Gdk, GLib
from matplotlib.figure import Figure
# Tell matplotlib to use a GTK canvas for drawing
#from matplotlib.backends.backend_gtk3agg import FigureCanvasGTK3Agg as FigureCanvas
from matplotlib.backends.backend_gtk3cairo import FigureCanvasGTK3Cairo as FigureCanvas
# Application Class
y = [self.index] * 1024
self.index += 1
if self.index > 1024: self.index = 0
Gdk.threads_enter()
self.line.set_ydata(y)
self.axes.set_title("%d" % self.index)
self.canvas.draw()
Gdk.threads_leave()
return True
self.index = 0
self.x = range(1024)
# Initialise the threads system and allow threads to work with GTK
GLib.threads_init()
# Draw scope
self.figure = Figure(dpi=100)
self.canvas = FigureCanvas(self.figure) # a Gtk.DrawingArea
#self.widget.alignment_ScopeDisplay.add(self.canvas)
# Draw initial scope
self.axes = self.figure.add_subplot(111)
self.line, = self.axes.plot(self.x, [0]* 1024)
self.axes.set_title("None")
self.axes.set_xbound(0.0, 1024)
self.axes.set_ybound(-16, 1040)
self.window_main = Gtk.Window(title="pyMatPlotLibTest")
self.window_main.connect("destroy", lambda x: Gtk.main_quit())
self.window_main.add(self.canvas)
self.window_main.show_all()
# Ticker for the update of the input state monitoring
Gdk.threads_add_timeout(priority = GLib.PRIORITY_DEFAULT_IDLE,
interval = 10, # msec
function = self.update_gui)
Gtk.main()
gui = pyMatPlotLibTest()
------------------------------
View this message in context: matplotlib.backends.backend_gtk3cairo
memory leak
<http://matplotlib.1069221.n5.nabble.com/matplotlib-backends-backend-gtk3cairo-memory-leak-tp45614.html>
Sent from the matplotlib - users mailing list archive
<http://matplotlib.1069221.n5.nabble.com/matplotlib-users-f3.html> at
Nabble.com.
------------------------------------------------------------------------------
One dashboard for servers and applications across Physical-Virtual-Cloud
Widest out-of-the-box monitoring support with 50+ applications
Performance metrics, stats and reports that give you Actionable Insights
Deep dive visibility with transaction tracing using APM Insight.
http://ad.doubleclick.net/ddm/clk/290420510;117567292;y
_______________________________________________
Matplotlib-users mailing list
https://lists.sourceforge.net/lists/listinfo/matplotlib-users

[David]

Post by Benjamin Root
I take it that it doesn't happen using the GTK3Agg backend?

GTK3Agg is unimplemented at the GTK3-end:

File
"c:\Python34\lib\site-packages\matplotlib\backends\backend_gtk3agg.py", line
69, in on_draw_event
buf, cairo.FORMAT_ARGB32, width, height)
NotImplementedError: Surface.create_for_data: Not Implemented yet.

Post by Benjamin Root
What about the threading portion? Does it happen if you take the threading
out?

I removed all calls to threads and swapped Gdk.threads_add_timeout to
Glib.timeout_add. This made little difference.

However if I comment the call to self.canvas.draw(), the python memory
utilisation sits at 30.8Mb (but the graph does not update of course).

Thanks

David




--
View this message in context: http://matplotlib.1069221.n5.nabble.com/matplotlib-backends-backend-gtk3cairo-memory-leak-tp45614p45616.html
Sent from the matplotlib - users mailing list archive at Nabble.com.

[Jens Nielsen]

IGtk3Agg should work in you use cairocffi instead of py(2/3)cairo. AFAIK
py(2/3)cairo is more or less unmaintained these days and that function has
never been implemented in a released version.

Best
Jens

Post by David

Post by Benjamin Root
I take it that it doesn't happen using the GTK3Agg backend?

File
"c:\Python34\lib\site-packages\matplotlib\backends\backend_gtk3agg.py", line
69, in on_draw_event
buf, cairo.FORMAT_ARGB32, width, height)
NotImplementedError: Surface.create_for_data: Not Implemented yet.

Post by Benjamin Root
What about the threading portion? Does it happen if you take the

threading

Post by Benjamin Root
out?

I removed all calls to threads and swapped Gdk.threads_add_timeout to
Glib.timeout_add. This made little difference.
However if I comment the call to self.canvas.draw(), the python memory
utilisation sits at 30.8Mb (but the graph does not update of course).
Thanks
David
--
http://matplotlib.1069221.n5.nabble.com/matplotlib-backends-backend-gtk3cairo-memory-leak-tp45614p45616.html
Sent from the matplotlib - users mailing list archive at Nabble.com.
------------------------------------------------------------------------------
One dashboard for servers and applications across Physical-Virtual-Cloud
Widest out-of-the-box monitoring support with 50+ applications
Performance metrics, stats and reports that give you Actionable Insights
Deep dive visibility with transaction tracing using APM Insight.
http://ad.doubleclick.net/ddm/clk/290420510;117567292;y
_______________________________________________
Matplotlib-users mailing list
https://lists.sourceforge.net/lists/listinfo/matplotlib-users

[David]

Post by Jens Nielsen
IGtk3Agg should work in you use cairocffi instead of py(2/3)cairo. AFAIK

py(2/3)cairo is more or less

Post by Jens Nielsen
unmaintained these days and that function has never been implemented in a
released version.

Yes you're right. This does work. However you need to do the following
steps:

Run: pip install cairocffi
Copy: C:\Python34\Lib\site-packages\gnome\libcairo-gobject-2.dll to
C:\Python34
Rename: libcairo-gobject-2.dll to libcairo-2.dll

Then in the python code, change the back-ends line to:
from matplotlib.backends.backend_gtk3agg import FigureCanvasGTK3Agg as
FigureCanvas

Incidentally, if you're trying the script, modify the line from the first
script example I sent -- the second has a horrible threads issue (due to the
missing call to threads_init).

Regards

David



--
View this message in context: http://matplotlib.1069221.n5.nabble.com/matplotlib-backends-backend-gtk3cairo-memory-leak-tp45614p45628.html
Sent from the matplotlib - users mailing list archive at Nabble.com.

------------------------------------------------------------------------------

[David]

To clarify, this fixes the memory leak issue.



--
View this message in context: http://matplotlib.1069221.n5.nabble.com/matplotlib-backends-backend-gtk3cairo-memory-leak-tp45614p45629.html
Sent from the matplotlib - users mailing list archive at Nabble.com.

------------------------------------------------------------------------------

[David Hughes]

Thanks, however GTK3Agg is unimplemented in at the GTK3-end:

File "c:\Python34\lib\site-packages\matplotlib\backends\backend_gtk3agg.py", line 69, in on_draw_event
buf, cairo.FORMAT_ARGB32, width, height)
NotImplementedError: Surface.create_for_data: Not Implemented yet.

Regards

David

From: ***@gmail.com [mailto:***@gmail.com] On Behalf Of Benjamin Root
Sent: 26 May 2015 14:53
To: David Hughes
Cc: Matplotlib Users
Subject: Re: [Matplotlib-users] matplotlib.backends.backend_gtk3cairo memory leak

I take it that it doesn't happen using the GTK3Agg backend? What about the threading portion? Does it happen if you take the threading out?
Ben Root

On Tue, May 26, 2015 at 8:23 AM, David <***@rapiscansystems.com<mailto:***@rapiscansystems.com>> wrote:
Hi, I seem to have a memory leak while generating a 'live' plot display. This wasn't the case for GTK2, but the example below is consuming ~800k/second (Matplotlib 1.4.3, PyGI aio-3.14.0_rev18, Windows 7 x64, python 3.4.3). I have checked the garbage collector but it doesn't show anything interesting (no massive incrementing count of uncollected items). Anyway, I would be very grateful if somebody could confirm and/or fix this (or tell me what I'm doing wrong). Many thanks David Code below:

from gi.repository import Gtk, Gdk, GLib





from matplotlib.figure import Figure

# Tell matplotlib to use a GTK canvas for drawing

#from matplotlib.backends.backend_gtk3agg import FigureCanvasGTK3Agg as FigureCanvas

from matplotlib.backends.backend_gtk3cairo import FigureCanvasGTK3Cairo as FigureCanvas





# Application Class

class pyMatPlotLibTest(object):



def update_gui(self):

y = [self.index] * 1024



self.index += 1

if self.index > 1024: self.index = 0



Gdk.threads_enter()

self.line.set_ydata(y)

self.axes.set_title("%d" % self.index)

self.canvas.draw()

Gdk.threads_leave()



return True



def __init__(self):

self.index = 0

self.x = range(1024)



# Initialise the threads system and allow threads to work with GTK

GLib.threads_init()



# Draw scope

self.figure = Figure(dpi=100)

self.canvas = FigureCanvas(self.figure) # a Gtk.DrawingArea

#self.widget.alignment_ScopeDisplay.add(self.canvas)



# Draw initial scope

self.axes = self.figure.add_subplot(111)

self.line, = self.axes.plot(self.x, [0]* 1024)

self.axes.set_title("None")

self.axes.set_xbound(0.0, 1024)

self.axes.set_ybound(-16, 1040)



self.window_main = Gtk.Window(title="pyMatPlotLibTest")

self.window_main.connect("destroy", lambda x: Gtk.main_quit())

self.window_main.add(self.canvas)

self.window_main.show_all()



# Ticker for the update of the input state monitoring

Gdk.threads_add_timeout(priority = GLib.PRIORITY_DEFAULT_IDLE,

interval = 10, # msec

function = self.update_gui)

Gtk.main()





if __name__ == "__main__":

gui = pyMatPlotLibTest()

________________________________
View this message in context: matplotlib.backends.backend_gtk3cairo memory leak<http://matplotlib.1069221.n5.nabble.com/matplotlib-backends-backend-gtk3cairo-memory-leak-tp45614.html>
Sent from the matplotlib - users mailing list archive<http://matplotlib.1069221.n5.nabble.com/matplotlib-users-f3.html> at Nabble.com.

------------------------------------------------------------------------------
One dashboard for servers and applications across Physical-Virtual-Cloud
Widest out-of-the-box monitoring support with 50+ applications
Performance metrics, stats and reports that give you Actionable Insights
Deep dive visibility with transaction tracing using APM Insight.
http://ad.doubleclick.net/ddm/clk/290420510;117567292;y
_______________________________________________
Matplotlib-users mailing list
Matplotlib-***@lists.sourceforge.net<mailto:Matplotlib-***@lists.sourceforge.net>
https://lists.sourceforge.net/lists/listinfo/matplotlib-users

[David Hughes]

I removed all calls to threads and swapped Gdk.threads_add_timeout to Glib.timeout_add (See attached. However if I comment the call to self.canvas.draw(), the python memory utilisation sits at 30.8Mb.


from gi.repository import Gtk, Gdk, GLib


from matplotlib.figure import Figure
# Tell matplotlib to use a GTK canvas for drawing
#from matplotlib.backends.backend_gtk3agg import FigureCanvasGTK3Agg as FigureCanvas
from matplotlib.backends.backend_gtk3cairo import FigureCanvasGTK3Cairo as FigureCanvas


# Application Class
class pyMatPlotLibTest(object):

def update_gui(self):
y = [self.index] * 1024

self.index += 1
if self.index > 1024: self.index = 0

#Gdk.threads_enter()
self.line.set_ydata(y)
self.axes.set_title("%d" % self.index)
self.canvas.draw()
#Gdk.threads_leave()

return True

def __init__(self):
self.index = 0
self.x = range(1024)

# Initialise the threads system and allow threads to work with GTK
#GLib.threads_init()

# Draw scope
self.figure = Figure(dpi=100)
self.canvas = FigureCanvas(self.figure) # a Gtk.DrawingArea
#self.widget.alignment_ScopeDisplay.add(self.canvas)

# Draw initial scope
self.axes = self.figure.add_subplot(111)
self.line, = self.axes.plot(self.x, [0]* 1024)
self.axes.set_title("None")
self.axes.set_xbound(0.0, 1024)
self.axes.set_ybound(-16, 1040)

self.window_main = Gtk.Window(title="pyMatPlotLibTest")
self.window_main.connect("destroy", lambda x: Gtk.main_quit())
self.window_main.add(self.canvas)
self.window_main.show_all()

# Ticker for the update of the input state monitoring
GLib.timeout_add(10, self.update_gui)
#Gdk.threads_add_timeout(priority = GLib.PRIORITY_DEFAULT_IDLE,
# interval = 10, # msec
# function = self.update_gui)
Gtk.main()


if __name__ == "__main__":
gui = pyMatPlotLibTest()

From: ***@gmail.com [mailto:***@gmail.com] On Behalf Of Benjamin Root
Sent: 26 May 2015 14:53
To: David Hughes
Cc: Matplotlib Users
Subject: Re: [Matplotlib-users] matplotlib.backends.backend_gtk3cairo memory leak

I take it that it doesn't happen using the GTK3Agg backend? What about the threading portion? Does it happen if you take the threading out?
Ben Root

On Tue, May 26, 2015 at 8:23 AM, David <***@rapiscansystems.com<mailto:***@rapiscansystems.com>> wrote:
Hi, I seem to have a memory leak while generating a 'live' plot display. This wasn't the case for GTK2, but the example below is consuming ~800k/second (Matplotlib 1.4.3, PyGI aio-3.14.0_rev18, Windows 7 x64, python 3.4.3). I have checked the garbage collector but it doesn't show anything interesting (no massive incrementing count of uncollected items). Anyway, I would be very grateful if somebody could confirm and/or fix this (or tell me what I'm doing wrong). Many thanks David Code below:

from gi.repository import Gtk, Gdk, GLib





from matplotlib.figure import Figure

# Tell matplotlib to use a GTK canvas for drawing

#from matplotlib.backends.backend_gtk3agg import FigureCanvasGTK3Agg as FigureCanvas

from matplotlib.backends.backend_gtk3cairo import FigureCanvasGTK3Cairo as FigureCanvas





# Application Class

class pyMatPlotLibTest(object):



def update_gui(self):

y = [self.index] * 1024



self.index += 1

if self.index > 1024: self.index = 0



Gdk.threads_enter()

self.line.set_ydata(y)

self.axes.set_title("%d" % self.index)

self.canvas.draw()

Gdk.threads_leave()



return True



def __init__(self):

self.index = 0

self.x = range(1024)



# Initialise the threads system and allow threads to work with GTK

GLib.threads_init()



# Draw scope

self.figure = Figure(dpi=100)

self.canvas = FigureCanvas(self.figure) # a Gtk.DrawingArea

#self.widget.alignment_ScopeDisplay.add(self.canvas)



# Draw initial scope

self.axes = self.figure.add_subplot(111)

self.line, = self.axes.plot(self.x, [0]* 1024)

self.axes.set_title("None")

self.axes.set_xbound(0.0, 1024)

self.axes.set_ybound(-16, 1040)



self.window_main = Gtk.Window(title="pyMatPlotLibTest")

self.window_main.connect("destroy", lambda x: Gtk.main_quit())

self.window_main.add(self.canvas)

self.window_main.show_all()



# Ticker for the update of the input state monitoring

Gdk.threads_add_timeout(priority = GLib.PRIORITY_DEFAULT_IDLE,

interval = 10, # msec

function = self.update_gui)

Gtk.main()





if __name__ == "__main__":

gui = pyMatPlotLibTest()

________________________________
View this message in context: matplotlib.backends.backend_gtk3cairo memory leak<http://matplotlib.1069221.n5.nabble.com/matplotlib-backends-backend-gtk3cairo-memory-leak-tp45614.html>
Sent from the matplotlib - users mailing list archive<http://matplotlib.1069221.n5.nabble.com/matplotlib-users-f3.html> at Nabble.com.

------------------------------------------------------------------------------
One dashboard for servers and applications across Physical-Virtual-Cloud
Widest out-of-the-box monitoring support with 50+ applications
Performance metrics, stats and reports that give you Actionable Insights
Deep dive visibility with transaction tracing using APM Insight.
http://ad.doubleclick.net/ddm/clk/290420510;117567292;y
_______________________________________________
Matplotlib-users mailing list
Matplotlib-***@lists.sourceforge.net<mailto:Matplotlib-***@lists.sourceforge.net>
https://lists.sourceforge.net/lists/listinfo/matplotlib-users