xdg-activation-v1: new protocol implementation

[emersion]

See https://gitlab.freedesktop.org/wayland/wayland-protocols/-/merge_requests/50

[dnkl]

Getting a crash when I set a surface on the token, but nothing else:

struct xdg_activation_token_v1 *token = xdg_activation_v1_get_activation_token(term->wl->xdg_activation);
xassert(token != NULL);
xdg_activation_token_v1_add_listener(token, &activation_token_listener, term);
xdg_activation_token_v1_set_surface(token, term->window->surface);
xdg_activation_token_v1_commit(token);

Backtrace:

(gdb) bt full
#0  token_handle_commit (client=0x55555636c620, token_resource=0x555556397870)
    at ../../subprojects/wlroots/types/wlr_xdg_activation_v1.c:75
        token = 0x555556392770
        token_str = "4a0e176c7b01ce178bbaa2da310b0323\000"
        __PRETTY_FUNCTION__ = "token_handle_commit"
#1  0x00007ffff6f90acd in  () at /usr/lib/libffi.so.7

(gdb) p *token
$1 = {activation = 0x555555ebb710, surface = 0x5555563825e0, seat = 0x0, serial = 0, app_id = 0x0, link = {prev = 0x555556392798, 
    next = 0x555556392798}, token = 0x0, resource = 0x0, seat_destroy = {link = {prev = 0x5555563927b8, next = 0x5555563927b8}, 
    notify = 0x0}, surface_destroy = {link = {prev = 0x555556382b18, next = 0x555556382988}, 
    notify = 0x7ffff7f707aa <token_handle_surface_destroy>}}

	if (token->surface != token->seat->keyboard_state.focused_surface) {
		wlr_log(WLR_DEBUG, "Rejecting token commit request: "
			"surface doesn't have keyboard focus");
		goto error;
	}

The reason I did this was because without setting a surface, I got:

00:00:11.373 [DEBUG] [wlr] [types/wlr_xdg_activation_v1.c:71] Rejecting token commit request: surface has been destroyed

I was under the impression setting the surface was optional? What am I missing?

/**
 * @ingroup iface_xdg_activation_token_v1
 *
 * The requesting client can specify a surface to associate the token
 * being created with it.
 *
 * Must be triggered before commit. This information is optional.
 */

[emersion]

Should be fixed now. I've tested with foot's xdg-activation branch and it seems to work as expected.

Setting the surface is optional, but is recommended so that the compositor can better track where the request comes from. A compositor might reject requests without a surface coming from clients which have surfaces opened, for instance.

[dnkl]

Thanks, works for me as well now. I've updated foot's xdg-activation branch to set the token surface to the top-level surface, and everything appears to (still) be working as expected.

[emersion]

Compositors should probably:

1. Invalidate all tokens on (keyboard?) focus change, assuming the user has moved on to something else and focus shouldn't be transferred. Except it may still be desirable to retain tokens to mark toplevels as urgent.
2. Immediately destroy new tokens for serials older than the current focus. With the same caveats as above.

[djpohly]

I'm trying to implement urgency in dwl using foot as a testcase, but I'm getting a null dereference in wlr_xdg_activation_v1.c line 75:

if (token->surface != token->seat->keyboard_state.focused_surface) {

with token->seat being NULL. A few lines later, the code does check to see if (token->seat != NULL) - maybe a few lines need to be reordered?

[dnkl]

It looks kind of like it (the xdg-activation) branch has been reverted to the original version, where Sway also crashed.

677d6a2 is the last version I tested (and found working) with foot. It looks vastly different.

[emersion]

Hm, sorry, I probably messed up a rebase. I restored the branch.

[emersion]

Added a token timeout to avoid these dangling token issues.

[kennylevinsen]

LGTM.