input/cursor: correctly transfer focus when using tablet pen

[Xyene]

Tentatively fixes #4819.

This commit ensures that seatop_motion is called to transfer focus when a window is selected via a pen. Previously, it would race with node_at_coords, and only properly transfer focus if its returned surface was NULL.

I've tested this locally, and it fixes the issues I was experiencing.

Calling seatop_motion directly here works as well, but my impression is that the confine handling inside cursor_motion should be done for tablet devices too. Applications see the following sequence of events:

[912231.664] wl_pointer@3.motion(7987491, 281.539062, 991.007812)
[912231.673] zwp_tablet_tool_v2@4278190083.motion(281.539062, 991.007812)
[912231.676] zwp_tablet_tool_v2@4278190083.frame(7987491)

In a GNOME session on the same application, the wl_pointer event is not present. I don't know enough to say which is the correct behaviour, or if it even matters. Sway infrequently omits the wl_pointer event, too.

[emersion]

Calling seatop_motion directly here works as well, but my impression is that the confine handling inside cursor_motion should be done for tablet devices too.

Yes. The wl_pointer.motion event is incorrect.

[Xyene]

Thanks! So, to confirm:

• We don't want wlr_relative_pointer_manager_v1_send_relative_motion to be called for tablet events.
• We do want wlr_region_confine to be called.
• We don't want wlr_cursor_move to be called. (Is this what triggers wl_pointer.motion?)

In that case, a follow-up question is: is

sway/sway/input/cursor.c

Lines 481 to 484 in fcd524b

	if (!surface || !wlr_surface_accepts_tablet_v2(tablet->tablet_v2, surface)) {
	wlr_tablet_v2_tablet_tool_notify_proximity_out(sway_tool->tablet_v2_tool);
	cursor_motion(cursor, time_msec, input_device->wlr_device, dx, dy, dx, dy);
	return;

correct? The branch is taken even if the surface supports tablet v2, if the surface returned by node_at_coords is NULL, and ends up sending wl_pointer.motion. That branch is how tablet focus seems to work right now (even if erratically).

[emersion]

and ends up sending wl_pointer.motion

Just like touch, we should use seat_set_focus to transfer focus.

[Xyene]

I've updated this PR to call seat_set_focus. I also added a check for focus_follows_mouse, which while not semantically correct (perhaps a new option should be better), matches i3 behavior.

[emersion]

Looks good, thanks!