Bug summary

I'm working on fixing several bugs with the macOS backend. I am trying to understand how show(), start_event_loop(), and stop_event_loop() should interact with each other.

I had several questions that I couldn't find in the documentation:

1. If show(block=True) is called, should stop_event_loop() make it return?
2. If start_event_loop() is called on a figure, should closing all windows make it return?
3. What should be the behavior if start_event_loop() or show(block=True) is called when an event loop is already running from show() or another start_event_loop()?
4. If stop_event_loop() is called without an event loop running, it appears to be a no-op across all backends. Should it instead throw an exception?

I wrote the attached code to test these. However, I am seeing different results from different backends. See charts in the "Actual Results" section.

Code for reproduction

import matplotlib.pyplot as plt
import matplotlib
import sys

example_index = int(sys.argv[1])

backend, mode = (
    ( "QtAgg",  "show" ),
    ( "TkAgg",  "show" ),
    ( "macosx", "show" ),
    ( "QtAgg",  "start_event_loop" ),
    ( "TkAgg",  "start_event_loop" ),
    ( "macosx", "start_event_loop" ),
)[ example_index ]

print(f"backend = '{backend}', mode = '{mode}'")


matplotlib.use(backend)

fig, ax = plt.subplots()
ax.set_title("Click anywhere to continue")
ax.plot([1, 2, 3])

clicked = []

def on_click(event):
    clicked.append((event.xdata, event.ydata))
    print("on_click")
    fig.canvas.stop_event_loop()

fig.canvas.mpl_connect('button_press_event', on_click)

fig, ax = plt.subplots()

if (mode == "show"):
    plt.show()
else:
    plt.show(block=False)
    fig.canvas.start_event_loop()

if (len(clicked) > 0):
    print(f"User clicked at: {clicked[0]}")
else:
    print(f"Event loop stopped without click")

Actual outcome

Does closing both windows stop the event loop?

Does clicking on Figure 1 stop the event loop?

Expected outcome

All backends should operate with the same behavior.

Additional information

No response

Operating system

No response

Matplotlib Version

Dev

Matplotlib Backend

No response

Python version

No response

Jupyter version

No response

Installation

None