The previous commit added a new worker thread. This thread might sit in
wait_for_event() indefinitely even after the X11Source was dropped. This
is because nothing guarantees that an X11 event will come in "soonish".
And only then would the thread notice that its main thread is gone.
This commit cleans that up by having X11Source explicitly wake up the
event thread and wait for it to exit in its Drop implementation.
Signed-off-by: Uli Schlachter <psychon@znc.in>
x11rb caches X11 events internally. This cache is used when waiting for
the reply of a request and an event is received first. This cache is
however also used when sending a request, because the X11 client may
never block and stop reading while writing a request. Doing so could
cause a deadlock with the X11 server.
Thus, this commit changes X11Source to spawn a thread internally. This
thread calls wait_for_event() in a loop and uses a calloop channel to
forward these events to the main loop. x11rb's RustConnection internally
uses a ConditionVariable to make sure this thread will wake up when
needed.
Signed-off-by: Uli Schlachter <psychon@znc.in>
expose output from output map and return a reference
in find
change scale with keyboard shortcut
scale input coordinates and render location
according to output_scale
scale buffers during rendering
adapt window locations on output scale
change to that the location will appear
to be stable
scale udev pointer location on output scale
change to make the pointer location appear stable
Use a UserDataMap to store the udev output id
Short names for udev outputs for easier output
scale configuration
Add debug logging to egl via the `EGL_KHR_debug` extension similar to our
support for `GL_KHR_debug`. Sadly unlike the GL-extensions the EGL-extension
does not let us provide any user-data for the callback, so we fall back to
using `slog-stdlog`, when available.
Since we only want to initialize this once, when libEGL gets loaded and
not per display, but we need to query the list of supported extensions
first, some initialization steps where moved around.
...if we are resizing. The configure message
is the result from telling the client that resizing
has stopped and therefore not include the resizing
state. The current state is safe to use because
AckConfigure will move the state to last_acked
and the current state will be set on the next surface commit.
move fullscreen output geometry calculation to
a helper function
rename with_outputs to with_client_outputs and
take client instead of generic resource
the positioning of outputs, tracks
the surfaces on the outputs and
sends enter/leave for the surfaces
Additionally the output_map will
handle re-location of windows if
a output mode is changed, an output is
added or removed.
The shell has been updated to use the
primary output for the initial placement.
The fullscreen/maximize requests will now
handle the shell part correctly. For real
fullscreen output the rendering has to be
changed.
The output name is considered unique
and an output can be retrieved from
the map by location, output or name.
This can be used as the base for HiDPI
handling.
Removes the notion of Seats from `InputBackend` and replaces it with
`DeviceAdded` and `DeviceRemoved` events allowing compositors to
descriminate between single devices.
A new abstraction to model multiple seats on a single input backend
may be be added separately at a later time.
This changes the state handling logic of wl_surface to automatically
track subsurface pending in the form of transactions. The role enum
(and the associated type parameter) is no more, and replaced by a
general-purpose typemap-like container on surfaces.
The new logic is introduced in the files:
- `src/wayland/compositor/cache.rs`
- `src/wayland/compositor/transaction.rs`
The rest of the PR is the fallout of these changes, as well as a few
trivial clippy fixes.
Victor Berger
Uli Schlachter
Christian Meissl
Poly
Victor Berger
Victor Brekenfeld
Victor Brekenfeld