XCB_DESTROY_NOTIFY & SIGTERM

[Uli Schlachter]

Hi,

Am 04.12.24 um 20:00 schrieb Steven J Abner:
> On Wed, Dec 4 2024 at 05:32:06 PM +0000, Uli Schlachter <psychon at znc.in> 
> wrote:
>> How come you only get XCB_DESTROY_NOTIFY for your window? Normally, 
>> one uses the WM_DELETE_WINDOW to get a notification and destroy the 
>> window oneself. That way, you can clean up your cairo resources before 
>> the window is gone.
> 
> On WM_DELETE_WINDOW:
>   /* check transients, appropriate branch */
> xcb_unmap_window(connection, window);
>   /* user save stuff, non-transient */
> xcb_destroy_window(connection, window);
> xcb_flush(connection);
> which yields a XCB_DESTROY_NOTIFY at which point:
> free/destroy data that was attached to destroy->window (event), from list.
> 
> But what you are implying is that cairo is wrong in that "The caller 
> owns the surface and should call cairo_surface_destroy() when done with 
> it.", and it's owned by xcb?
> Easy enuf to move after user save, I think?. Would create 2 locations 
> for resource deletion. Maybe a cairo ownership issue? Insights, on cairo 
> if any?

I am not sure about what you are implying with ownership, but (with my 
cairo hat on), I would claim that you have to (at least) 
cairo_surface_finish() before the drawable is destroyed. It might also 
be possible to just flush it, but I would not count on it.

Uli

P.S.: cairo_surface_destroy() implicitly calls cairo_surface_finish() 
when the reference count hits zero.
-- 
“I’m Olaf and I like warm hugs.”