We tried to synchronize layer shell popups with the parent layer shell
on commits, but this is subtly wrong because we would only update
the position for one layer shell that was committed, but not any other
layer that might be affected. By moving handling to the scene descriptor
we can iterate all popups and ensure they are synchronized.
```
Program terminated with signal SIGSEGV, Segmentation fault.
144 struct wlr_layer_surface_v1 *layer_surface = surface->layer_surface;
[Current thread is 1 (Thread 0x7f1f7c5b3ac0 (LWP 2473))]
(gdb) bt
```
Add a NULL check in `find_mapped_layer_by_client` like the one in `arrange_surface`.
Instead of doing this roundabout thing where we get the surface from the
view, let's instead get it from the `wlr_surface_state` that we already
track in `handle_commit`. This makes the NULL state impossible which is
what the old `get_geometry` is checking for and generally cleans
things up a little bit.
Also don't check if the geometry x/y changed, those will always
be 0 for xwayland.
Since output layout is destroyed when the wayland display is destroyed
we run into a destroy listener order problem: Either the display starts
destroying the outputs first, in which case we're good: The existing
handling will clean up. However, things go wrong if the display decides
to destroy the output layout first. In this case, sway will hold
invalid references to the output layout as part of each output so that
when it finally goes to destroy them, sway will dereference destroyed
output layout bits.
Ref: https://github.com/swaywm/sway/pull/6844#issuecomment-1843599513
3d5ae9813d added logic to change the
underlying wlr_toplevel size for floating containers, but it does it
even if the container has no actual coordinates yet. This doesn't really
make sense to update the toplevel size in this case since there's many
things that could affect the initial coordinates (sway commands,
fullscreen state, etc.). Skip this by doing a crude check to see if the
current container state has any width.
If a floating client resizes itself, sway updates several of its
internal dimensions to match but not wlr_toplevel. This means that the
next time wlroots sends a toplevel configure event, it can have wrong
coordinates that resize the client back to its old size. To fix this,
let's just use wlr_xdg_toplevel_set_size so the wlr_toplevel has the
same dimensions as sway.
Exactly the same as 0183b9d35d0ce750588e700e536b7d2e367b0d0a but the
logic is onlly applied to xdg_shell and not xwayland.
With recent wlroots changes, backends which don't support output
modes can now support being disabled.
We were always marking mode-less outputs as disabled. Stop doing
that, check whether the output takes up some space in the layout
instead.
