output: store a weak ref in the global to avoid...

[cmeissl]

...a cycle keeping the output alive indefinitely

Should get rid of the last cycle in smithay as mentionen in #1585 and hopefully fix #1584

[cmeissl]

@ids1024 With this the check seems to work as expected in anvil (not saying we shouldn't think about an alternative)

[ids1024]

Ah. So when does global user data like this actually get freed? Only once every client that has had a registry with the global present has disconnected?

[Drakulix]

I don't think we should and can just return here. Any request to the output would otherwise trigger a panic, given we didn't initialize it's data.

The warning is fine (and a good idea), but I think we should just initialize the OutputUserData with Weak::new.

[cmeissl]

Ah. So when does global user data like this actually get freed? Only once every client that has had a registry with the global present has disconnected?

The global data itself is freed on remove_global. I actually started this before pulling in your changes, so looking at it again, the change here is unnecessary. Just tried it out and the primary_scanout_output in PrimaryScanoutOutput::update_from_render_element_states seems to work fine in anvil on master.

[ids1024]

Yeah, as long as this is freed immediately when remove_global is called, it makes a lot of sense for the global to hold a strong reference.