We should probably talk about where all these tests are going? mpl_toolkits has, in the past, been discussed as under-tested and under-maintained. If we want to maintain it, why not move into the main library?

I have not really understood the role of mpl_toolkits. Is it correct that it is possible to "install" other packages under mpl_toolkits and that some do?`

Primarily just trying to increase coverage and learn a bit more of the code base... (And sometimes, untested code actually turns out to be incorrect, like in this case.)

But one could for sure, I guess, claim that for 4.0 one moves the useful parts of it to the main library?

💯 your work here has been great and, if we plan to keep these routines around, is well overdue. But if the plan is to obsolete a fair bit of this, then I didn't want you to feel you had wasted effort.

It would be great to revisit this on a call soon. I guess id recommend that if you find parts of the tool kit useful, consider if there are straightforward ways to bring into the main library. For instance we did that with inset_axes a couple of years ago, and 3.6 will have layout="compressed" which is meant to do the same things as image_grid.

I do not really have that much of a threshold for doing this. Something I play around with when I lack energy for proper things... It will probably take some time to deprecate etc anyway.

Yes but if we codify the behaviour with tests, it becomes even harder to change and or deprecate.