[ENH]: Add a QtPgfPng backend? about matplotlib HOT 8 OPEN

Seems reasonable. It looks like a majority of the work left writing a lot of docs?

from matplotlib.

Yes (including explaining in what context that would be useful). Perhaps also set whatever attributes are needed to mark this as not supporting blitting. Also decide whether we also want the same for the other GUI toolkits (it's just busy work I suspect...).

from matplotlib.

I'm not following why this is a desirable feature?

from matplotlib.

Interactive figures are great for exploration or fine tuning (a semi-common workflow for me is to work in an IPython shell using the explicit API to poke at a live figure and then doing %history and copy-pasting-editing that to save). Making this "just work", even with a super slow render, for people who want to output to pgf is small win (now I guess you could do fig.savefig(...) and hope your pdf viewer out-refreshes).

from matplotlib.

Fair, I just think the point of PGF is to write the plot in the PGF language for including in a (**)TeX document, so creating a very slow viewer for the compiled result seems out of scope. I appreciate that the code cost is low, but my concern is that it raises expectations and obscures the point of the PGF backend.

from matplotlib.

This only uses public API, maybe the right path is to put this in a new very thin repo and see if it gets any usage?

from matplotlib.

Sounds possible, or even as an example.

BTW I'm not strongly opposed at all, just making sure we aren't adding a feature that we will later regret.

from matplotlib.

I don't really mind either way; the main question is how to best help users confused by what parts support xelatex syntax and what parts don't (#5234, #28084).

from matplotlib.