This PR is a patchset, it should not be squashed.

It introduces a bunch of tiny fixes (few lines each) to the Wayland keyboard implementation that I didn't feel were worth as separate PRs.

Each commit is properly formatted with a message but here's the tl;dr and some extra info:

Wayland: Fix SIGFPE with repeat rate of 0

You can disable key repetition by setting the rate to 0, classic divide-by-zero.

Confimed before-and-after on sway with sway input * repeat_rate 0.

Wayland: Add default keyboard repetition values

It is technically possible to have a compositor so old that it doesn't support the repeat_info event. Let's add some useful default values for key repetition (instead of 0), just in case.

Confirmed that it works on Weston with this patch:

--- a/libweston/input.c
+++ b/libweston/input.c
@@ -4330,7 +4330,7 @@ weston_seat_init(struct weston_seat *seat, struct weston_compositor *ec,
 	wl_signal_init(&seat->tablet_tool_added_signal);
 
 	seat->global = wl_global_create(ec->wl_display, &wl_seat_interface,
-					MIN(wl_seat_interface.version, 7),
+					MIN(wl_seat_interface.version, 3),
 					seat, bind_seat);
 
 	seat->compositor = ec;

Also double-checked that the existing wl_keyboard::repeat_info logic still works on sway, with sway input * repeat_rate 10.

Destroy XKB keymap and state on seat capability change

I forgor 💀

Luckily it eventually cleaned up anyways, but y'know.

They should all hopefully be cherry-pickable to 4.5. The SIGFPE one is particularly important.

I'm also planning to do some other changes (wl_keyboard version 10, key handling logic consolidation), but I think that those are juicy enough for their own PR.