'Figure' object has no attribute '_original_dpi' #9040
Comments
|
maybe related to #8052 |
|
What sha are you at? #9002 went in yesterday and may have fixed this. |
|
I just pulled from master (20 min ago) |
|
From |
|
I can not reproduce this, I assume it has something to do some subtle ordering in when things are created and when things are attempted to be drawn. Does applying 20:00 $ git diff
diff --git a/lib/matplotlib/backends/backend_qt5agg.py b/lib/matplotlib/backends/backend_qt5agg.py
index 78d9fdada..85954c003 100644
--- a/lib/matplotlib/backends/backend_qt5agg.py
+++ b/lib/matplotlib/backends/backend_qt5agg.py
@@ -176,7 +176,7 @@ class FigureCanvasQTAggBase(FigureCanvasAgg):
self.draw()
-class FigureCanvasQTAgg(FigureCanvasQTAggBase, FigureCanvasQT):
+class FigureCanvasQTAgg(FigureCanvasQT, FigureCanvasQTAggBase):
"""
The canvas the figure renders into. Calls the draw and print fig
methods, creates the renderers, etc.help? |
|
@tacaswell no, I change the order and it doesn't help. The I think |
|
@tacaswell if I make the change in Why it shows me the error in qt5? if it is qt4 in reality? |
|
Ah! Because the qt4 backend is a sub-class of the qt5 backend. The main differences between them are that qt5 is co-opperative and qt4 is not which mostly shows up in the |
import networkx
import matplotlib.backends.backend_qt4agg
def build_inheritance_graph(cls):
g = networkx.DiGraph()
todo = set([cls])
done = set()
while todo:
cls = todo.pop()
for b in cls.__bases__:
if b not in done:
todo.add(b)
g.add_edge(cls.__name__, b.__name__)
return g
g = build_inheritance_graph(matplotlib.backends.backend_qt5agg.FigureCanvas)
networkx.draw_networkx(g)will give you a visualization of the MI diagram from these classes (there are 3 diamonds, 1 from us, 1 from Qt and 1 from us + Qt back to |
def decorate_init(cls):
for b in cls.mro():
decorate(b)
def decorate(cls):
print(cls.__name__)
if cls is object:
return
if 'matplotlib' not in cls.__module__:
return
if '__init__' not in cls.__dict__:
return
orig_init = cls.__dict__['__init__']
def new_init(self, *args, **kwargs):
print(cls.__module__, cls.__name__)
return orig_init(self, *args, **kwargs)
cls.__init__ = new_init
decorate_init(matplotlib.backends.backend_qt4agg.FigureCanvas)Will trace through what get called. |
|
I have a patch that works but calls base class inits too many times.... |
in two places `super` needs to be used in the Qt5 case and frustrated in the Qt4 case. These are handled by the two `_fake_super_*` methods. The must be just private as we need to override in sub-classes. Closes matplotlib#9040
|
@tacaswell with the decorator |
|
#9052 solves this problem |
|
closed by #9052 |
|
In 2.1.0 I still see this problem, and with the decorator shown above I get From the path above, you may also see versions used, pyqt is 5.9 (the latest). |
|
How did you install matplotlib? Are you using pyqt4 or pyqt5? |
|
Sorry about the 5.9, specification. This is the print-out at installation: |
|
I just noticed that in fact it is using PyQt 5.9, as I initially wrote. If I do: So I guess there is a bug in the QT_VERSION_STR, I will try and report this to their mailing list. |
|
Ahh, no, of course. QT is the parent library, while PYQT is the python wrapper qt. |
|
In fact, I believe this to be a bug in while I guess it should read: What do you think? |
|
I think you're right, would you submit a PR for that fix? Not sure what's going wrong with your original problem, unfortunately. |
|
@zerothi Do you still have this problem with 2.1.1? |
|
Closing as we can not reproduce this with 2.1.1. Please ping to be re-opened if you can still reproduce this issue with the current stable release. |
With current master, running any example
python3 histogram_cumulative.pyI get
It opens the window but the image is black
If I force to use a GTK3 backend I dont get the error and the figure is fine
The text was updated successfully, but these errors were encountered: