[Bug]: compressed layout setting can be forgotten on second save · Issue #24954 · matplotlib/matplotlib · GitHub

Bug summary

I'm not sure whether this is really a bug or I'm just using an inconsistent combination of options. Under some specific circumstances (see below) compressed layout is not applied the second time a figure is saved.

Code for reproduction

import matplotlib
import matplotlib.pyplot as plt
import numpy as np

arr = np.arange(100).reshape((10, 10))

matplotlib.rcParams['figure.constrained_layout.use'] = True

fig, ax_dict = plt.subplot_mosaic('AB;AC', figsize=(6, 9), width_ratios=[3, 2],
                                  layout='compressed')

for key in ["B", "C"]:
    ax_dict[key].imshow(arr)
    
fig.savefig("test1.png", bbox_inches="tight")
fig.savefig("test2.png", bbox_inches="tight")

Actual outcome

test1.png

test2.png

Expected outcome

Both images should look like the first.

Additional information

If I do not set the rcParams, all is well. If I do not set bbox_inches="tight" in my calls to savefig, the images are identical (but I have too much white space top and bottom). Maybe there is a better option than bbox_inches="tight" when using compressed layout?

For context, my real example is a script that makes several figures. For most of them I want constrained layout, so I set that once in the rcParams for convenience. Only one figure needs "compressed", and I am saving twice because I want both a png and a pdf. Fixed it in my current example by just reverting the rcParams setting for the one figure.

Operating system

RHEL7

Matplotlib Version

3.6.2 and main

Matplotlib Backend

QtAgg

Python version

3.9 and 3.11

Jupyter version

No response

Installation

conda

Yeah we do some dancing around when we save with bbox inches - so this seems to get caught in that. I tried to track it down, but the figure-saving stack is full of context managers, and I can't see where the layout manager gets reset. Hopefully someone more cognizant of that part of the codebase can explain.

Thanks for looking @jklymak 🙂

I think it is set (temporarily) here;

matplotlib/lib/matplotlib/backend_bases.py

Lines 2356 to 2357 in 018c5ef

    
           # we have already done layout above, so turn it off: 
        
           stack.enter_context(self.figure._cm_set(layout_engine='none'))

It is, but I don't understand what _cm_set does to reset the layout engine after this. Somehow it is dropping the old layout engine and making a new one, and the new one doesn't know that the old one was a 'compressed' engine.

It calls get_{kwarg} and after running calls set({kwarg}={old value}). So here it calls oldvalue = figure.get_layout_engine() and figure.set(layout_engine=oldvalue). Is figure.set_layout_engine(figure.get_layout_engine()) working correctly?

I am way out of my depth here but

import matplotlib.pyplot as plt

plt.rcParams['figure.constrained_layout.use'] = True
fig = plt.figure(layout="compressed")

print(fig.get_layout_engine()._compress)
fig.set_layout_engine(fig.get_layout_engine())
print(fig.get_layout_engine()._compress)

fig.savefig('foo.png', bbox_inches='tight')
print(fig.get_layout_engine()._compress)

True
True
False

Without the rcParams line, fig.get_layout_engine() returns None after the savefig.

I think the problem is the call to adjust_bbox

matplotlib/lib/matplotlib/backend_bases.py

Lines 2349 to 2350 in 018c5ef

    
           restore_bbox = _tight_bbox.adjust_bbox( 
        
               self.figure, bbox_inches, self.figure.canvas.fixed_dpi)

which explicity calls

matplotlib/lib/matplotlib/_tight_bbox.py

Line 21 in a3011df

fig.set_layout_engine(None)

which will use the default constrained layout engine if the rcParams is set

matplotlib/lib/matplotlib/figure.py

Lines 2599 to 2610 in a3011df

    
           if layout is None: 
        
               if mpl.rcParams['figure.autolayout']: 
        
                   layout = 'tight' 
        
               elif mpl.rcParams['figure.constrained_layout.use']: 
        
                   layout = 'constrained' 
        
               else: 
        
                   self._layout_engine = None 
        
                   return 
        
           if layout == 'tight': 
        
               new_layout_engine = TightLayoutEngine(**kwargs) 
        
           elif layout == 'constrained': 
        
               new_layout_engine = ConstrainedLayoutEngine(**kwargs)

[Bug]: compressed layout setting can be forgotten on second save · Issue #24954 · matplotlib/matplotlib · GitHub

Repository navigation

[Bug]: compressed layout setting can be forgotten on second save #24954

Bug summary

Code for reproduction

Actual outcome

Expected outcome

Additional information

Operating system

Matplotlib Version

Matplotlib Backend

Python version

Jupyter version

Installation

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

	# we have already done layout above, so turn it off:
	stack.enter_context(self.figure._cm_set(layout_engine='none'))

	restore_bbox = _tight_bbox.adjust_bbox(
	self.figure, bbox_inches, self.figure.canvas.fixed_dpi)

	if layout is None:
	if mpl.rcParams['figure.autolayout']:
	layout = 'tight'
	elif mpl.rcParams['figure.constrained_layout.use']:
	layout = 'constrained'
	else:
	self._layout_engine = None
	return
	if layout == 'tight':
	new_layout_engine = TightLayoutEngine(**kwargs)
	elif layout == 'constrained':
	new_layout_engine = ConstrainedLayoutEngine(**kwargs)

[Bug]: compressed layout setting can be forgotten on second save · Issue #24954 · matplotlib/matplotlib · GitHub

Repository navigation

[Bug]: compressed layout setting can be forgotten on second save #24954

Description

Bug summary

Code for reproduction

Actual outcome

Expected outcome

Additional information

Operating system

Matplotlib Version

Matplotlib Backend

Python version

Jupyter version

Installation

Activity

jklymak commented on Jan 12, 2023

rcomer commented on Jan 12, 2023

QuLogic commented on Jan 13, 2023

jklymak commented on Jan 13, 2023

QuLogic commented on Jan 13, 2023

rcomer commented on Jan 13, 2023

rcomer commented on Jan 13, 2023

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions