Bug report

Bug summary

When a plot is made with theqt5 backend on and then the backend is changed toinline, the active plot window(s) are closed. This makes it very difficult to switch cleanly between backends within a notebook. The two mediocre workarounds I can see are:

• always specify the backend before plotting (since you can't automatically "switch back")
• useplt.show(block=True) and switch back toinline after the user is done with the GUI figure. This has the downside of locking the Notebook execution while the figure is live.

Code for reproduction

The linked gist shows what I'd like to accomplish which is a context manager that enables matplotlib plotting in a GUI window from within a Notebook that is otherwise using theinline backend. Basically I want the notebook to use inline (for a variety of reasons), but I occasionally wish I could interact with the data in a separate figure.
https://gist.github.com/flutefreak7/65d824358122360911e2d4c43085007a

As a side note, easy switching betweeninline andnotebook/widget backends would also scratch part of this itch, but the interactive notebook backends still don't enable full screen usage or easily throwing a plot on another monitor.ipyvolume has full screen figured out, so that seems doable.

Here's the context manager I wish worked:

# Paste your code here@contextmanagerdefwindow(block=False):%matplotlibqt5plt.ioff()yieldplt.show()# The switch back to inline closes the qt5 plotplt.ion()%matplotlibinlinewithwindow():plt.plot([1,3,2])

Actual outcome

The outcome of the above code is that a plot window flashes into existence for a split second, then is closed when the%matplotlib inline call is processed.

Expected outcome

It would be great if plots created with the qt5 backend could stay visible while other plots with the inline backend were also being created. If use%gui qt (to establish a reliable event loop) and create a bunch of Qt windows by other means, they live concurrently with the Notebook as long as the kernel is alive. I'd like matplotlib GUI figures to be able to live on regardless of the current backend.

Matplotlib version

• Operating system:
• Matplotlib version: 3.1.1
• Matplotlib backend (print(matplotlib.get_backend())):inline andqt5agg
• Python version: 3.7.3
• Jupyter version (if applicable):

jupyter                           1.0.0jupyter-client                    5.3.4jupyter-console                   6.0.0jupyter-contrib-core              0.3.3jupyter-contrib-nbextensions      0.5.1jupyter-core                      4.6.0jupyter-highlight-selected-word   0.2.0jupyter-latex-envs                1.4.6jupyter-nbextensions-configurator 0.4.1jupyterlab                        1.0.5