Sub-pixel motions are lost or amplified on X11 if application uses absolute motion and pointer warp about sdl HOT 7 CLOSED

This is unlikely to ever be fixed in SDL2, as mouse coordinates are treated everywhere in the library as integers. SDL3 uses floats for pointer coordinates to avoid losing subpixel precision, including for warp destination coordinates.

SDL2 also doesn't implement setting the pointer position hint via the confinement protocol in the Wayland driver, so it won't appear to be affected because any warp functionality is a no-op.

from sdl.

Does this happen with SDL3, using the latest code in main?

from sdl.

Can I use SDL3 as a drop in in a proprietary game that currently uses SDL2? Otherwise, can you recommend some demo application that I can test SDL3 with?

Edit: It would have to be an application that uses absolute motion + warp to emulate relative motion. I believe that relative motion works as expected since it uses raw motion data that usually has no fractional component.

Edit: I tried sdl2-compat and in X11 mode the pointer is unable to move at all even if I move the mouse quickly so that the delta should be >=1. Wayland continues to work as expected.

from sdl.

Conceptually, I would think a warp would clear out any fractional mouse motion when it resets the position. I'm curious why this works on Wayland?

from sdl.

It appears that the wayland driver automatically switches into relative motion mode if the application tries to warp the cursor and the cursor is invisible.

from sdl.

That makes sense. I wonder if we should do that for other drivers?

from sdl.

Additionally, it looks like XWayland converts absolute pointer coordinates to integers internally: https://gitlab.freedesktop.org/xorg/xserver/-/blob/master/hw/xwayland/xwayland-input.c#L660

With SDL3 under a real X session, subpixel pointer coordinates are passed through. On XWayland, I'm only seeing integers.

That makes sense. I wonder if we should do that for other drivers?

It wouldn't be hard to move the activation logic to the mouse layer, and the Wayland SDL_HINT_VIDEO_WAYLAND_EMULATE_MOUSE_WARP hint could be repurposed for the general case. The one 'gotcha' case to watch out for is when SDL tries to warp the pointer during fullscreen transitions, as this has inadvertently activated the relative mode emulation when it shouldn't if the client has the system cursor hidden in favor of drawing their own software cursor.

e: Running testrelative with SDL_MOUSE_RELATIVE_MODE_WARP=1 shows just how badly the rounding affects the movement under XWayland.

from sdl.