The grid seems to be turned off deliberately

I don't know if there is any specific reason for that?

Weird. It’s been like that for 14 years though, so may be hard to change?

Looks like pcolor() does the same, and has basically always done so.
It's probably actually not too hard to change that (and I think removing that line makes sense): one can check before the call to grid(False) whether the grid is on (which would only happen if explicitly set earlier in the code and then the user relying on pcolor(mesh)() unsetting it...) and if so emit a deprecationwarning asking to unset the grid first.

Could it be that this serves the purpose of overriding rcParams[axes.grid.*] settings for these plots? By default, a style might have grid lines, but they don't make sense for false-color plots.

Anyway should imshow() have the same behavior?

As a suggestion, maybe add another rcParams['axes.auto_disable'] ?

Could it be that this serves the purpose of overriding rcParams[axes.grid.*] settings for these plots? By default, a style might have grid lines, but they don't make sense for false-color plots.

Actually I occasionally do want grids on my images.

As a suggestion, maybe add another rcParams['axes.auto_disable'] ?

In that case I would suggest just disabling the grid yourself with grid(False) (or modifying the style to unset rcParams["axes.grid.*"]); I don't think every possible arrangement needs to be doable via rcParams.

Actually I occasionally do want grids on my images.

Agreed.

I don't think every possible arrangement needs to be doable via rcParams.

Agreed.

However, we need the possibility to have a style with gridded regular axes, but no grids on false-color plots. That's a reasonable combination of defaults. Moreover, that's how current styles are rendered, and we cannot request ggplot-style or seaborn-style users to add grid(False) to all their false-color plots manually.

Since an Axes does not know it will be filled with a false-color plot at creation time, I don't think there is a way around calling grid(False) within pcolormesh and similar. The only question is if it's called unconditionally, or if we add kwargs and/or rcParams to control it.

we cannot request ggplot-style or seaborn-style users to add grid(False) to all their false-color plots manually.

[...] if we add kwargs and/or rcParams to control it.

If we're going to make the user add a kwarg to each call I don't see how that's better than requesting that the user calls grid(False) first.

Also note that imshow() doesn't have that behavior, even though it's arguably in the same category of plots.

I don’t think any individual artists should set axes properties. If it weren’t for back compatibility concerns I’d vote to change this

If we're going to make the user add a kwarg to each call I don't see how that's better than requesting that the user calls grid(False) first.

For backward compatibility, the default would of course stay to deactivate the grid.

My point is that this can actually be changed with a proper deprecation transition (check whether the grid was set, and if so, warn that pcolormesh won't unset it for you anymore in the future).

I don’t think any individual artists should set axes properties.

It's the pcolormesh() method, not the artist. There's other precedence, e.g. imshow sets the aspect. I believe that some plotting functions may modify the axes by default to adapt it to the intention of the plot type.