• Sway version: 1.10-dev-be840f73 (Sep 11 2024, branch 'master')
• Debug log: sway.log (nothing useful)
• Description: There seems to be a regression of Fix swaylock showing transient cursor after locked #7690 (fixing Mouse cursor visible in swaylock swaylock#333) after the port to the scene graph API. To reproduce: launch sway (default config), launch swaylock (no arguments), observe visible cursor until it is moved. The cursor_rebase_all call in lock.c no longer triggers pointer enter.

This happens because at the time of the cursor_rebase_all call, while the wlr_surface for the lock surface has its client buffer initialized correctly, the wlr_scene_surface's scene buffer is null, so when sway/input/cursor.c traverses the scene tree it doesn't hit the lock surface.

The cursor rebase happens when the lock surface role's commit callback calls wlr_surface_map, triggering sway/lock.c's surface map callback. The scene buffer is initialized immediately after that, from the surface commit signal, by wlroots/types/scene/surface.c:204.

I don't know if this is a bug in sway or wlroots. Two small fixes which work but may be incorrect:

• Listen for surface commit rather than surface map in sway/lock.c for focus / cursor rebasing.
• Move the surface_reconfigure call from wlroots/types/scene/surface.c:204 to a surface map callback.

The change in lock.c is tiny and seems completely harmless; I'm happy to make it a PR if appropriate.