--mouse-sensitivity or -s introduces a dead range.

[Gibitwit]

I was excited to see that --mouse-sensitivity now works. But it adds a hefty dead range to mouse movement. The view will not move at all without a good amount of movement of the mouse in a short time. This makes --mouse-sensitivity option useless. I suppose it does not actually scale mouse tracking but throws out counts instead.

Please eliminate the dead range.

[hfc2x]

What game are you playing? What did you set the sensitivity to? Those details are kind of too important to be left out lol.

[Gibitwit]

Shadow Warrior 3. Gamescope sensitivity set to 0.922. For Shadow Warrior 3 I also had to use Force grab cursor.

Star Wars Battlefront 2 2005 with Gamescope sensitivity set to 0.8586.

[anone-zone]

It would seem that zwp_relative_pointer_v1_send_relative_motion does nothing with dx_unaccel and dy_unaccel values less than 1.

Maybe not a proper fix but could hack together accumulation by doing something like:

--- a/src/wlserver.cpp
+++ b/src/wlserver.cpp
@@ -2335,2 +2335,4 @@ static bool wlserver_apply_constraint( double *dx, double *dy )

+static double rem_x = 0.0;
+static double rem_y = 0.0;
 void wlserver_mousemotion( double dx, double dy, uint32_t time )
@@ -2342,2 +2344,10 @@ void wlserver_mousemotion( double dx, double dy, uint32_t time )

+       dx += rem_x;
+       dy += rem_y;
+       rem_x = fmod(dx,  1.0);
+       rem_y = fmod(dy,  1.0);
+       dx -= rem_x;
+       dy -= rem_y;
+       if (abs(dx) <= 0 && abs(dy) <= 0) return;
+
        wlserver_perform_rel_pointer_motion( dx, dy );

Also #1371 maybe related.

[tobiasjakobi]

@laurirasanen Is this behaviour documented somewhere? I had a look at the spec, but the only thing I can see there is that the deltas are transmitted as fixed point. So it has of course happen, that e.g. if dx is very small (as double), then after the fixed point conversion it becomes zero. But that would mean dx needs to be smaller than 1.0 / 256.0. Is this usually the case here?

[anone-zone]

@tobiasjakobi Not sure, but I can repro the issue if the double value is less than 1.0.

Tested with on Plasma X11 with:

• gamescope -s 0.5 --force-grab-cursor --backend sdl.

And on Plasma Wayland with:

• gamescope -s 0.5 --force-grab-cursor --backend sdl
• gamescope -s 0.5 --force-grab-cursor --backend wayland

All running a 3D application that uses SDL2 for windowing if that matters.

The double values aren't so small that the conversion would break. Doing wl_fixed_to_double(wl_fixed_from_double(x)) still produces reasonable values within 1/512 of the original.

[anone-zone]

Seems that the issue happens specifically with applications that use SDL_SetRelativeMouseMode(SDL_TRUE).

In relative mode SDL_MOUSEMOTION events wont be triggered in the application unless one of the deltas in wlserver_mousemotion is >= 1.0.

Made a minimal repro here: https://github.com/anone-zone/sdl-gamescope-1521

Running the repro in gamescope with a patch for printing the sent values:

--- a/src/wlserver.cpp
+++ b/src/wlserver.cpp
@@ -2339,6 +2339,7 @@ void wlserver_mousemotion( double dx, double dy, uint32_t time )
 
        dx *= g_mouseSensitivity;
        dy *= g_mouseSensitivity;
+       printf("dx %f dy %f\n", dx, dy);

Will log, for example:

dx 0.200000 dy 0.600000
dx 0.100000 dy 0.600000
dx 0.200000 dy 0.700000
dx 0.200000 dy 0.700000
dx 0.300000 dy 0.800000
dx 0.300000 dy 0.700000
dx 0.300000 dy 0.800000
dx 0.500000 dy 1.000000
SDL_MOUSEMOTION 0 1
dx 0.500000 dy 1.100000
SDL_MOUSEMOTION 0 1
dx 0.600000 dy 1.000000
SDL_MOUSEMOTION 0 1
dx 0.800000 dy 1.300000
SDL_MOUSEMOTION 0 1
dx 0.700000 dy 1.300000
SDL_MOUSEMOTION 0 1
dx 0.900000 dy 1.300000
SDL_MOUSEMOTION 0 1
dx 0.900000 dy 1.300000
SDL_MOUSEMOTION 0 1
dx 1.100000 dy 1.500000
SDL_MOUSEMOTION 1 1
dx 1.200000 dy 1.700000
SDL_MOUSEMOTION 1 1
dx 1.200000 dy 1.500000
SDL_MOUSEMOTION 1 1
dx 1.800000 dy 2.100000
SDL_MOUSEMOTION 1 2

[tobiasjakobi]

@laurirasanen I've checked some of the games that I have mouse issues with, and Wolfenstein 2 appears to have the exact same problem: #1560 (comment)

Applying this integer coordinate buffering workaround makes things working there.

[2meito]

@tobiasjakobi any chance for a pull request?

[tobiasjakobi]

@2meito This is hack/workaround. It's not a proper solution.

[VoodaGod]

• i adapted @nullprop's patch as quoted below to apply to master here: https://gist.github.com/VoodaGod/e4ac5363bd98b86cf7ab739101f852e5 (let me know if it doesn't makes sense, it seems to work as expected)
• if you use the gamescope-git AUR package, here is a patch to the PKGBUILD to apply this patch: https://gist.github.com/VoodaGod/ae342280c368827a0e3315cda68e6f56

It would seem that zwp_relative_pointer_v1_send_relative_motion does nothing with dx_unaccel and dy_unaccel values less than 1.

Maybe not a proper fix but could hack together accumulation by doing something like:

--- a/src/wlserver.cpp
+++ b/src/wlserver.cpp
@@ -2335,2 +2335,4 @@ static bool wlserver_apply_constraint( double *dx, double *dy )

+static double rem_x = 0.0;
+static double rem_y = 0.0;
 void wlserver_mousemotion( double dx, double dy, uint32_t time )
@@ -2342,2 +2344,10 @@ void wlserver_mousemotion( double dx, double dy, uint32_t time )

+       dx += rem_x;
+       dy += rem_y;
+       rem_x = fmod(dx,  1.0);
+       rem_y = fmod(dy,  1.0);
+       dx -= rem_x;
+       dy -= rem_y;
+       if (abs(dx) <= 0 && abs(dy) <= 0) return;
+
        wlserver_perform_rel_pointer_motion( dx, dy );

Also #1371 maybe related.

[EmilianC]

Same exact issue here. Any chance for main branch fix?

[anone-zone]

This behavior is fixed in SDL3 since the mouse motion event uses floats now: https://wiki.libsdl.org/SDL3/README-migration#sdl_eventsh

Mouse events use floating point values for mouse coordinates and relative motion values. You can get sub-pixel motion depending on the platform and display scaling.

sdl2-compat already handles accumulation but only if the renderer is used with specific flags:
https://github.com/libsdl-org/sdl2-compat/blob/0db0a745e1b6c31c37b95020b239548cee29cc43/src/sdl2_compat.c#L2307-L2308

Not sure if this is an issue for gamescope or sdl2-compat. @misyltoad or @slouken might have insight.