Semi-OT: IMHO subplot(m, n, p) is only marginally better than subplot(mnp). Figuring out what the p-th element always confuses me and is a lot of cognitive overhead. Unless we're making a point on using pyplot (which I think has subplot as the canonical way of adding multiple plots) we should switch to subplots().

@timhoffm I agree, but thats a far bigger job.

And the plt.subplot(m, n, p) grammar is very ingrained, and is at least well-defined.

Conversely subplot(mnp) is a really old grammar that causes genuine problems because it only applies to 9 or fewer rows and columns subplots.

Anybody using mnp or m, n, p with p > 9 must be a genius or a masochist. 😆 But I agree that we should do the change proposed here.

BTW, we could add p = i, j so that subplot(4, 4, (0, 3)) would be the top right edge. Would need to decide on 0-based vs. 1-based here, which is a bit unclear as the scalar p is 1-based, but I still would use numpy/matrix indexing. But that's for another time.

subplot(i, j, (k, l)) basically already exists, except that it's spelled subplot2grid((i, j), (k, l)) (zero-based indexing). I guess subplot2grid is so obscure though that also having the functionality in subplot may not be a bad idea? (not that I claim any understanding of what is good taste in a pyplot-centered workflow...).

And when you think you know all the miraculous ways you can create an Axes, there‘s always one more API ... 😩. I‘m really not sure we should change any of these API anymore unless we have a greater plan which and how these should be used (and maybe soft-deprecate some).

This is only referring to m, n, (i, j) not to the contents of this PR.

Well, we already have m, n, (start, stop), at least until it was broken ;-) Please see #17343

I would be happy to see the 3-digit form go--but we're still stuck with it in the axes_grid family, I see, which also has the idiosyncratic nrows_ncols kwarg.

I don't see any reasons not to merge this, but it's grown a bunch of conflicts now.

👍 on the PR. It's proably easier to redo this from scratch using regex serach&replace than handling all the conflicts.

I only relearn regexp every 6 months or so, so this may have to wait, or anyone else is welcome to take this up instead!

Most cases have been addressed via the above PRs.

Still open: More complex layouts that are not full single-cell grids. These can be handled using subplot_mosaic, but I'd like to wait until the single-line notation #18763 is in.

I'll go ahead and close this, but feel free to restore if you need it to track the issue...