pointer moving, clicking doesn't work on headless backend about cage HOT 2 OPEN

I have tested it using weston-clickdot with WAYLAND_DEBUG=1 environment variable to try to understand what's happening.
Without any modification on wlrctl, I can observe wl_pointer.button event is not sent by the compositor (Cage) to the client (clickdot). If I add a call to wl_display_dispatch() or wl_display_roundtrip() (or just a simple sleep) after zwlr_virtual_pointer_manager_v1_create_virtual_pointer() in wlrctl, the event is well distributed.

I am clearly not a Wayland nor wlroots expert at all, but as I understand it in headless case, no pointer device exists before wlrctl creates a virtual one. Once created, the client (clickdot) get a notification of capabilities update for wl_seat and retrieves a pointer object using wl_seat.get_pointer. It seems that wlroots send pointer events only to client which has requested a pointer object yet. That's where the quick and dirty sleep patch I have tested in wlrctl can make a difference, allowing the client (clickdot) to have time to request a pointer object.

I can't see a reliable way to determine if client is ready to receive pointer events to propose a patch for wlrctl but it is quite interesting.

from cage.

I added -c to the grim command to show the mouse pointer. There is no pointer.

I noticed wlrctl creates a virtual pointer and then destroys it. I modified wlrctl and added some sleep time before it destroys the pointer, and took a screenshot while wlrctl was running and discovered that there is pointer.

Is there a way to keep the pointer after wlrctl exits?

from cage.