I hoped to split this up into multiple commits, but the api
changes to `shell/workspace.rs` were to invasive to feasibly do this.
Here is a rough list of changes:
- Fullscreen windows aren't mapped to other layers anymore
- This they need their own logic for:
- Sending frames
- Dmabuf Feedback
- Primary outputs
- On commit handlers
- cursor tests
- They get their own unmap/remap logic
- They get a new restore state similar to minimized windows
- Refactored the minimized window state to reuse as much as possible
here
- They need to be part of focus stacks, which means adjusting them
to a new type `FocusTarget` as they previously only handled
`CosmicMapped`.
- Various shell handlers (minimize, move, menu) now have dedicated
logic for fullscreen surfaces
- This was partially necessary due to relying on CosmicSurface now,
partially because they should've had their own logic from the
start. E.g. the context menu is now reflecting the fullscreen
state
- Fullscreen windows may be rendered behind other windows now, when they
loose focus.
- This needed changes to input handling / rendering
It can still get unthrottled frame callbacks by not having any primary
scanout output, but presumably we want this. Alongside the code setting
it for cursor surfaces, and dragged windows.
Add more sophisticated code to handle the primary node disappearing.
Also overhaul the selection logic to respect our allow/deny-list and
prefer devices with built-in connectors before using the boot gpu.
This will also allow triggering a primary node switch at runtime
for debugging purposes in the future.
This is slightly simpler, if there's not some reason I'm missing to do
this as it was previously done. And in particular provides a cleaner
API (if we wanted to move this to Smithay; perhaps without the Cosmic
extension).
But it also should be more correct. Presumably if a client (unusually)
had multiple components with their own `ext_workspace_manager_v1`
instance, they should have their own queues, and
`ext_workspace_manager_v1::commit` should be independent.
Inevitably, there's a racy element to multiple components trying to
update the workspace state like this, but it should behave the same as
two clients with separate connections.
(This is different from `CompositorClientState`, since the commit queue
there is fundamentally tied to the client, and different components with
their own compositor instance way have related surfaces.)
Requires https://github.com/Smithay/smithay/pull/1676.
This changes two things:
* `Workspace::is_empty` no longer checks if there are activation tokens,
but a separate `Workspace::can_auto_remove` checks if the workspace is
empty and has no activation tokens.
- When we add workspace pinning, that can also be checked there.
* `Workspace` no longer contains a `pending_tokens` list that is updated
on `refresh`. Instead, `can_auto_remove` takes the xdg activation
state as an argument.
Since `Workspace::refresh` normally is run for focused workspaces, this
fixes allowing non-focused workspaces to be removed when an activation
token expires. It seems generally good to avoid tracking the activation
tokens in two places, and this is probably more efficient than needing
to refresh in more places.
By splitting this, we still don't remove an empty workspace if it has a
pending activation token, but we also don't add an empty workspace for
an activation token.
This mitigates the confusing behavior with activation tokens that aren't
used, but having to wait a few seconds in some cases before a workspace
is removed is still a little confusing. (We probably want `cosmic-term`
and `cosmic-workspace` to either consume the activation tokens they are
passed, or not be passed tokens when started by keybinding?)
Fixes https://github.com/pop-os/cosmic-comp/issues/1099.
This should fix an issue where output hotplug can sometimes cause
clients (including XWayland) to crash with a protocol error trying to
bind the output.
Using a timer doesn't seem ideal, but seems to be the correct way to do
this at present. Wlroots `wlr_global_destroy_safe` is basically the same
as this.
Adding a `LoopHandle` argument to `OutputConfigurationState::new` seems
awkward, but maybe better than a handler method for removing globals.
(`IdleNotifierState::new` also takes a `LoopHandle`). Perhaps Smithay
could provide some kind of helper for this.
If an individual window was being screencast, and that window was not
visible, for example because it was minimized, that window would only
be rendered every 995ms, which did not look good on the screencast.
Now, non-visible windows with active screencopy sessions, as well as
windows that are mapped on non-visible workspaces with active
screencopy sessions, are rendered every 1/60th of a second, which
matches the frame rate of the video produced by
xdg-desktop-portal-cosmic. In future, it might be good to let
screencopy clients suggest a redraw rate for captured windows.
`Output` in Smithay doesn't track if the output still exists, other than
based on whether or not it has strong references. Which doesn't seem to
be working correctly.
There may be leaked strong references to `Output`s somewhere, and maybe
Smithay should track if an output is still valid, generally when it is
exposed as a Wayland global
(https://github.com/Smithay/smithay/issues/1584). But a check like this
works for now.
Addresses https://github.com/pop-os/cosmic-comp/issues/985.
Previously, changing the resolution, scale, or orientation of a display
left tiling layouts that don't fill the screen and/or extend off of the
screen area, until an action like opening a window makes it recalculate
window positions.
Now this is done immediately when the output configuration changes.
Perhaps we should consider if we want a different animation for things
like rotating a screen, but the current behavior isn't too bad.
For floating layouts, `refresh` already remaps windows that are out of
bounds for the output, so this doesn't change that. Perhaps decreasing
the resolution (or moving to a lower resolution output) should try to
reduce the window size. But the current behavior generally seems okay.
Previously, screenshots and screencopy panicked if no DRM render nodes
are present.
Instead, create a `GlowRenderer` using llvmpipe, if available. This
should work as a fallback until pixman is integrated.
Avoids a little duplication (matching two variants, instead of three
backends).
The behavior, including errors and panics, should be unchanged for now.
Performance should also not be impacted.
This should help for adding llvmpipe rendering without a device node on
the kms backend, or for adding a variant for pixman.
