[PATCH v2 0/1] MR7857: winewayland: Don't update constraints without wl_pointer.

[Attila Fidan (＠atticf)]

If ClipCursor is called while the seat doesn't have a pointer and the
call qualifies for locking the pointer, it would have tried to lock a
null wl_pointer.

For example, I'm launching a fullscreen application (which causes ClipCursor calls on startup) in a headless compositor before I launch my virtual keyboard/pointer client. While the pointer is deinitialized and there cannot be a focused hwnd, the cursor wl_surface will not be set, making it qualify for pointer locking.

--
  v2: winewayland: Require wl_pointer for pointer constraints.

https://gitlab.winehq.org/wine/wine/-/merge_requests/7857

[PATCH v2 1/1] winewayland: Require wl_pointer for pointer constraints.

[Attila Fidan]

From: Attila Fidan

If ClipCursor is called while the seat doesn't have a pointer and the
call qualifies for locking the pointer, it would have tried to lock a
null wl_pointer.
---
 dlls/winewayland.drv/wayland_pointer.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/dlls/winewayland.drv/wayland_pointer.c b/dlls/winewayland.drv/wayland_pointer.c
index 6c852292c1d..509301cada4 100644
--- a/dlls/winewayland.drv/wayland_pointer.c
+++ b/dlls/winewayland.drv/wayland_pointer.c
@@ -688,9 +688,10 @@ static void wayland_pointer_update_constraint(struct wl_surface *wl_surface,
     }

needs_lock = wl_surface && (((confine_rect || covers_vscreen) &&
-                 !pointer->cursor.wl_surface) || force_lock);
+                 !pointer->cursor.wl_surface) || force_lock) &&
+                 pointer->wl_pointer;
     needs_confine = wl_surface && confine_rect && pointer->cursor.wl_surface &&
-                 !force_lock;
+                 !force_lock && pointer->wl_pointer;

if (!needs_confine && pointer->zwp_confined_pointer_v1)
     {

[Attila Fidan (＠atticf)]

On Sat Apr 26 05:16:37 2025 +0000, Alexandros Frantzis wrote:
> Another alternative would be to encode the wl_pointer requirement in
> `needs_lock` and `needs_confine` in `wayland_pointer_update_constraint`.
> This has the extra benefit that it moves us towards better handling of
> dynamic removals, since we will then destroy any inert zwp constraint
> proxies without any special code paths (we are not completely there yet,
> since we don't guarantee reapplication of the cursor clip on wl_pointer
> removal, but it's a step in the right direction). Does such a change
> work for your use case?

Sure, I moved it there.

[PATCH v2 0/1] MR7857: winewayland: Don't update constraints without wl_pointer. - approved

[Alexandros Frantzis (＠afrantzis)]

This merge request was approved by Alexandros Frantzis.