I have a relevant piece of information from my experience with set_mode.

Whether a window sticks around through set_mode or not is consistent and platform dependent. IIRC on MacOS a new window is created on a resize, for example, but not on Windows.

Ok, I haven't really fleshed out this idea yet, but

set_mode is also the window resizing API. It shouldn't lose the contents of the original window surface.

I see this was kind of discussed. I think the fact that you are both testing on windows could be impacting this as well, because of different SDL_Window behavior across OSes.

The idea behind this PR is to add consistency. It should be a documented API detail thatset_mode clears a window and users are supposed to redraw.

As mentioned above, the behaviour from pygame 1.9.6 to pygame 2.1.2 on windows has been clearing the window. This is only breaking now as an affect of the SDL 2.0.22/2.0.20 update (on windows only)

I also just tested on Ubuntu 22.04 (WSL) and the behaviour is the same as windows on pygame 2.1.2 (SDL 2.0.16), and I also had a SDL 2.0.22 wheel lying around from my manylinux PR (the one that is slow) and this too works the same the rest.

I don't know the status of this on macs.

I can't imagine many situations where pygame could successfully manage a window resizing eventwithout the user redrawing.

Where do we put the pixels? Do we run a smoothscale on the display surface? Fill the new space with black pixels in the case of an enlargement? Cut off the surface when shrinking?

I think always clearing to black is the easiest behaviour at pygame's level - because we can't know exactly what the application developer wants to happen when their window is resized.

I just noticed this change is missing some docs, will add that rq

It doesn't seem there is consensus yet on this one.

set_mode is also the window resizing API. It shouldn't lose the contents of the original window surface.

Does resizing the window leave the old surface just on windows? (or other platforms too?)
Seems like that might be useful behaviour?

To me, it's 99% clear that set_mode should definitely clear the surface the first time set_mode is called.

Yeah, I also agree with clearing the background on every set_mode() call (so that users can redraw their stuff). I also tested 1.9.6 behaviour and it is inline with 2.1.2 and the screen is always black after a set_mode call (so this issue is definitely a 2.1.3 break)

Ok, since this is old behaviour; I agree that this patch should be merged. The new behaviour could be useful though, but perhaps that can be added with a flag.

If there's no objections from@Starbuck5 I'd like to merge this for 2.1.3.