Added a test for the cycle_window_in_direction function. The function is
tested by creating 3 windows and cycling in both the next a previous
direction. The test will ensure that the windows were cycled by checking
the window index to ensure it is the expected window.
Created a test for cycling the window by index. This test is similar to
the other cycle window test and performs the same steps as that one,
except it uses the test_cycle_container_index_in_direction function
instead.
This commit removes some workarounds on the `update_focused_workspace`
function that were there to fix issues related to some bugs on the
`workspace.restore()` function.
This commit fixes the bugs on the `restore` function instead. The
`update_focused_workspace` function should be used only to update a
workspace layout.
This commit ensures that when the user switches to the flooating layer
any new spawned window will be spawned floating as if the float override
was set.
This results in a workflow which makes it easier to spawn floating
windows on the fly.
This effectively reverts commit ff2aa5e51a.
This fixes the issue where borders moved to other monitors would
sometimes be removed by their initial monitor because this data wasn't
being updated.
This effectively reverts commit 5919f88b38.
Added a test for moving a workspace from one monitor to another. The
test will ensure that after calling the function, the current monitor
index is the one where the workspace was transfered to. Test also will
check the count to ensure the new workspace was added to the new monitor
and that the old monitor has one less.
Added a test for swapping monitor workspaces. The test will create two
monitors one with a workspace that has a container containing multiple
windows and another with one. The test calls the swap_monitor_workspaces
function and then checks to see if the two workspaces were successfully
swapped.
Added a test to test toggling the tiling state. After each switch the
test will test the current state to ensure that it is the expected
state.
It's very annoying and unclear when enabling the float override to open
the next window as floating to have that window appear immediately over
an existing tile.
This commit sets an explicit behaviour to center windows spawned while
the float override is active.
This commit addresses a border rendering issue when moving a window from
a higher-indexed monitor to a lower-indexed monitor.
Previously, we would do a single render pass across all monitors in
order of their indexes, destroying borders no longer needed, and
creating new borders for new windows.
This resulted in the window being moved to the lower-indexed monitor
still existing in the global border cache when that monitor's borders
were updated, but then being removed when the borders of the origin,
higher-indexed monitor were updated.
With the changes in this commit, if we encounter a situation like this,
an additional render pass will be executed to ensure that the window
will have a corresponding border created on the destination
lower-indexed monitor.
This commit ensures that a workspace float override will also be applied
when the user switches to the floating layer, and removed when the user
switches to the tiling layer. This results in a workflow which makes it
easier to spawn floating windows on the fly.
This commit ensures that if a user switches from a workspace with
managed windows to a workspace without any managed windows but only
floating windows, the focused window from the previous workspace will
lose focus as it should, and the first floating window on the new
workspace will gain focus as it should.
Added a test for removing a window from a container. The test checks to
ensure a new container is created and that the removed window from the
original container is added to the new container. Also ensures that the
old container has one less container after the function is called.
Added a test for removing a workspace. The test keeps track of the
number of workspaces and the current workspace index.
Added test for swapping containers from different monitors. The test
will create two monitors and containers with different amount of
windows. Test will verify that the swap is successful by checking that
the number of windows in the container matches the number of windows it
was created with in the previous monitor. The same will be done with the
other container.
This commit allows either the single canonical applications.json file,
or multiple files which adhere to the asc scheme to be given to the
app_specific_configuration_path config option.
I thought I had already implemented this earlier, but evidently I
hadn't.
This will be useful for people who want to maintain their own
independent set of asc rules, as they can be kept in a dedicated file
which won't be overwritten by the fetch-asc command.
resolve#736
Use the preferred id set by the user as the key when caching monitors.
This way if a user is having issues with the device_id on their system
they can use the serial_number_id on display_index_preferences.
If instead a user is having issues with the serial_number_id they can
make sure to use the device_id on display_index_preferences. Komorebi
will use the preferred one as a key when caching the monitors.
This PR also includes a change of the DISPLAY_INDEX_PREFERENCES from a
Mutex to RwLock.
I believe the RwLock is better since it allows multiple readers at the
same time, while the mutex blocks on all calls even if you just want to
read or compare some value.
For display index preferences (same thing applies to other existing
mutexes) most times we access it is a read-only access. We only ever
change it when applying a config or when a command is used to change it.
Every other use is a read-only use so it doesn't make sense to block on
those situations
This commit fixes an issue where the move/resize functions for floating
windows weren't properly taking into account the coordinates of
secondary monitors and were only working correctly on the main monitor
where top/left was 0/0.
Implemented drop to ensure that the socket is deleted whenever the
function goes out of scope. This will ensure that if the test fails, the
test socket file will still be removed.
Added a test for cycling to the next and previous container in a
workspace.
Added a test for transfering a window from Monitor 1 to Monitor 2
Added a test for transfering a container to another montior. Also
updated the transfer window test to transfer the window back to the
original container. Both tests will check both monitors to ensure the
expected number of containers are in both.
Previously if we had a stack on a monocle container and tried to cycle
stack or move the window within the stack or even using the focus stack
window from a bar it would focus the wrong window and temporarely show
that wrong window. This commit fixes this.
Created a test that creates the WM instance and ensures the instance is
running. The test creates a custom socket and then cleans up the socket
file after completion.
Created a test that creates a WM instance, monitor instance, and
workpace. The tests checks to ensure that the expected workspace is
focused properly.
Included recommended fixes to ensure that the focus_workspace function
is used correctly and that the test accurately checks the workspaces
length, current workspace index, and switching to an existing workspace.
This commit adds a variation of the cycle-workspace command which will
attempt to focus the next empty workspace in the given direction (with
the usual wraparound). If there are no empty workspaces available, this
command will do nothing.
This commit refactors the `border_manager` with the following changes:
- Rework the way the borders are created and their pointer is sent to
the window message handler. Now we also store the same pointer as a
`Box<Border>` on the `border_manager`'s `BORDER_STATE`. This means
that the borders we have are exactly the same ones that the border
window itself is accessing so we can now store the border's info
inside the `Border` struct and all of it will be accessible by the
border window as well. This makes it so the "ACTUAL" border struct is
the one created on the thread of the `Border::create()` function and
when that thread finishes (after the border window closes) it will
handle the drop of the border itself. However this means we need to be
careful with our own stored `Box<Border>` since it will point to the
same memory we can't let the compiler dropped them as usual or
otherwise it would create heap corruption errors. So this commit
creates a special function called `destroy_border()` to actually close
the window border without dropping the border, since it will be later
dropped by the thread that created it.
- Remove `BORDERS_MONITORS`, `FOCUS_STATE` and `RENDER_TARGETS` arc
mutexes, since now this info is stored on the border itself.
- Change the `BORDER_STATE` to now map an id (container or window) to a
`Box<Border>`.
- Change the `WINDOWS_BORDERS` to now map a window hwnd to a border id.
- Create new struct `BorderInfo` which as the border window hwnd and the
border kind. This struct is what is now returned by the function
`window_border()` which checks if some window as a border attached to
it and if it does it returns this info. There is no need to clone the
entire border. If in the future we need more info we can just add it
to this struct.
- Change the way we clear the `BORDER_STATE`. Like mentioned before we
need to be sure we don't drop the `Box<Border>` when removing it, so
now we use the `.drain` function to remove all the borders as an
iterator and we call the `destroy_border()` on each border we are
removing.
- We now check if a border's `tracking_hwnd` has changed and if it does
we simply update it instead of destroying the border and create a new
one.
- Create function `delete_border` so that we can remove a border
properly from outside the `border_manager`.
- Create function `hide_border` which spawns a new thread that searches
if a window hwnd has a border attached to it and if it does it hides
said border window. This function is called on every `window.hide()`.
- Create function `show_border` which spawns a new thread that searches
if a window hwnd has a border attached to it and if it does it restores
said border window. This function is called on every `window.restore()`.
- This commit also changes the previous `window.hide()` and
`window.restore()` functions to be named:
- `window.hide_with_border(hide_border: bool)`: this is the same
function as before but adds a check at the end in case `hide_border`
is true it calls `border_manager::hide_border()`. A new function was
created with the same name as before `window.hide()` which by
default calls this new function with `hide_border = true`.
- `window.restore_with_border(restore_border: bool)`: this is the same
function as before but adds a check at the end in case `hide_border`
is true it calls `border_manager::hide_border()`. A new function was
created with the same name as before `window.hide()` which by
default calls this new function with `hide_border = true`.
- This commit creates a new function on `Container` called
`load_focused_window_ignore_borders()` which performs the same as
`load_focused_window()` but it ignores the borders when hiding and
restoring the windows. This function, along with the
`hide_with_border(false)` and `restore_with_border(false)` are used on
all functions related to changing focus on a stack since if we let the
borders be hidden and restored when cycling or changing focus on a
stack the border would flicker slightly, this prevents that. Ignore
borders when clicking on the stackbar as well.
(P.S. there might still be other places that I forgot to use these new
functions, but if that is the case then what will happen is a simple
flicker of the stack border...)
- The `remove_window` from `Workspace` needs to call the
`border_manager::delete_border()` so that wew make sure we remove that
windows's border window as well if it exists. This is essential when
enforcing workspace rules, otherwise the border would be left behind.
- Lastly, but not least, now that we hide the borders windows along with
their tracking window, we no longer remove the borders when swapping
workspaces or when toggling monocle, etc. Instead we keep all borders
of all workspaces cached and simply hide them. They are only removed
when their tracking window is closed or cloaked on a stack (since on a
stack we only keep one border for all the entire stack container).
This means that when changing between workspaces we no longer see the
borders showing up delayed after the windows show up. Now both the
window and it's border show up as if they are one and the same.
This commit makes all schemars::JsonSchema derives optional. After
analyzing the output of cargo build timings and llvm-lines, it was clear
that the majority of the 2m+ incremental dev build times was taken up by
codegen, and the majority of it by schemars.
Developers can now run cargo commands with --no-default-features to
disable schemars::JsonSchema codegen, and all justfile commands have
been updated to take this flag by default, with the exception of the
jsonschema target, which will compile with all derives required to
export the various jsonschema files.
Incremental dev build times for komorebi.exe on my machine are now at
around ~18s, while clean dev build times for the entire workspace are at
around ~1m.
This commit ensures that the resize dimensions will be reserved for
other monitors and workspaces when the
toggle-window-based-work-area-offset command is used.
Previously the stacking logic would sometimes change the focused
container without actually changing focus to said container.
This resulted in the stack showing up with an unfocused border even
though we had focus on one windows belonging to the stack (just not the
right one).
Also before it wasn't possible to stack windows on some directions
when we were already on a stack.
This commit fixes both issues.
This commit fixes multiple issues with the borders which were resulting
in multiple borders being created and not completely destroyed which
meant that the amount of borders in memory kept increasing indefinitely
the more we used komorebi. To do so this commit does the following:
- Clear all the maps on `destroy_all_borders`.
- Create function `remove_border` which should always be used when we
want to remove a border since this function destroys the border window
and removes all its related data and clones from all the maps.
- Create function `remove_borders` which should always be used when we
want to remove multiple borders or filter the current existing
borders. It takes in a `condition` function that takes a ref to the
border's container id and a ref to the border and should return a
bool. This function is then applied to each existing border and if it
evaluates to true it will call `remove_border` on it.
- Apply these new functions on all the code that was previously manually
removing borders.
- When a container is a stack, we now check it's unfocused windows, in
case they had borders attached to them we remove them, otherwise these
borders would persist and be drawn below other borders.
- We now check if a container's border was previously tracking a different
window, if it was we destroy that border and remove it's hwnd from the
`FOCUS_STATE` and then we check if its `tracking_hwnd` still points to
the same border and remove it if it does (if it doesn't that means
that a new border was already attached to that window so we don't
remove it).
We don't call `remove_border` here since we don't want to actually
remove the border but instead we replace it with a new one tracking
the correct window. (I've tried updating the `tracking_hwnd` instead
of destroying the border and creating a new one but that didn't work
since it still kept tracking the previous window...).
This commit makes it so when you toggle between workspace layers it
moves all windows of that layer to the top of the z-order.
So if you move to `Floating` layer, then all floating windows are moved
to the top, and if you go back to the `Tiling` layer, all tiled
containers are moved to the top.
This commit stops the `FocusChange` event from focusing a floating
window when it is the one emitting said `FocusChange` event, since it's
not needed and is the cause of some flicker bugs reported!
If that floating window was the one emitting the `FocusChange` event
then it means it is already the foreground window, and there is no
reason for us to focus the window again (since that will create an
infinite loop of events).
When the window emitting this event is not floating we don't try to
focus the window, we simply set the focus index for the container of
that window and the focused index for the window of that container.
Except in one case, which is if the workspace has a monocle container,
then it does focus a window, it focuses the monocle window to make sure
the monocle keeps showing in front of everything and doesn't let
anything come in front of it.
Previously the reaper at startup would lock it's own `HWNDS_CACHE` and
then try to lock the WM to get its `known_hwnds`.
However if there was an event in the meantime, process_event would lock
the WM first and then it would try to lock the reaper's `HWNDS_CACHE` to
update it. This would deadlock since that would be locked by the reaper
waiting for the WM lock to be released.
This commit now makes it so we pass the `known_hwnds` to the reaper as
an argument at startup and it also rearranges the order of loading the
listeners.
Now komorebi first loads all the manager-type listeners, and only
afterwards does it load the commands and events listeners.
After some testing this seems to be the best order that doesn't cause
any issues at all!
There were some other issues that I've noticed before when starting
komorebi while having other 3rd parties trying to subscribe to it (like
komorebi-bar and YASB), which would make those subscribers lock the
`process_command` thread. This doesn't seem to be happening on my tests
anymore with this new order.
This commit fixes an issue where if you started komorebi without a
monitor connected, and then connected it later, it wasn't properly
loading the data returned from `win32-display-data`.
This commit adds the ability to set container and workspace padding per
monitor.
To do so (and to simplify any future need of changing some value per
monitor), and have it pass through to each workspace a new field was added
to `Workspace` called `globals` which has a new struct called
`WorkspaceGlobals`.
`WorkspaceGlobals` includes any global values that might be needed by
the workspace.
This field is updated by the monitor for all its workspaces whenever the
config is loaded or reloaded. It is also updated on `RetileAll` and on
the function `update_focused_workspace`.
This should make sure that every time a workspace needs to use it's
`update` function, it has all the `globals` up to date!
This also means that now the `update` function from workspaces doesn't
take any argument at all, reducing all the need to get all the
`work_area`, `work_area_offset`, `window_based_work_area_offset` or
`window_based_work_area_offset_limit` simplifying the callers of this
function quite a bit.
Lastly this commit has also (sort of accidentaly) fixed an existing bug
with the `move_workspace_to_monitor` function.
This was previous removing the workspace from a monitor, but wasn't
changing it's `focused_workspace_idx`, meaning that komorebi would get
all messed up after that command. For example, the `border_manager`
would get stuck and the komorebi-bar would crash.
Now, the `remove_focused_workspace` function also focuses the previous
workspace (which in turn will create a new workspace in case the removed
workspace was the last workspace).
Previously if a workspace had any floating windows it would always focus
the first one when restoring. Now it only focus the floating window if
the workspace layer is `Floating`.
This commit introduces an implementation of workspace layers to
komorebi.
Workspace layers change the kinds of windows that certain commands
operate on. This implementation features two variants,
WorkspaceLayer::Tiling and WorkspaceLayer::Floating.
The default behaviour until now has been WorkspaceLayer::Tiling.
When the user sets WorkspaceLayer::Floating, either through the
'toggle-workspace-layer' command or the new bar widget, the 'move',
'focus', 'cycle-focus' and 'resize-axis' commands will operate on
floating windows, if the currently focused window is a floating window.
As I don't have 'cycle-focus' bound to anything, 'focus up' and 'focus
down' double as incrementing and decrementing cycle focus commands,
iterating focus through the floating windows assigned to a workspace.
Floating windows in komorebi belong to specific workspaces, therefore
commands such as 'move' and 'resize-axis' will restrict movement and
resizing to the bounds of their workspace's work area (or more
accurately, the work area of the monitor that the workspace belongs to,
as floating windows are never constrained by workspace-specific work
area restrictions).
This commit changes the way the reaper works.
First this commit changed the `known_hwnds` held by the `WindowManager`
to be a HashMap of window handles (isize) to a pair of monitor_idx,
workspace_idx (usize, usize).
This commit then changes the reaper to have a cache of hwnds which is
updated by the `WindowManager` when they change. The reaper has a thread
that is continuously checking this cache to see if there is any window
handle that no longer exists. When it finds them, the thread sends a
notification to a channel which is then received by the reaper on
another thread that actually does the work on the `WindowManager` by
removing said windows.
This means that the reaper no longer tries to access and lock the
`WindowManager` every second like it used to, but instead it only does
it when it actually needs, when a window actually needs to be reaped.
This means that we can make the thread that checks for orphan windows
run much more frequently since it won't influence the rest of komorebi.
Since now the `known_hwnds` have the monitor/workspace index pair of the
window, we can simply get that info from the map and immediately access
that monitor/workspace or use that index info.
The `WorkspaceConfig` stored on `Workspace` was changed to not be
serialized, however it needs to be serialized and deserialized when
caching a monitor (after a disconnect), so that when it reconnects it is
able to read all the workspace rules, which include:
- initial_workspace_rules
- workspace_rules
- window_container_behaviour_rules
- layout_rules
- custom_layout_rules
This commit changes the serde skip to only skip if is is `None`.
This means that the `komorebic state` command will have all this
information as well and it will send it when notifying subscribers too,
which isn't good at all, so we need to find another way of excluding it
from the state.
This commits adds a few more events that can trigger a
`DisplayConnectionChange` event.
Some of these events are redundant and after a display is
disconnected/reconnected it emits multiple `DisplayConnectionChange`
events.
However, when trying to remove them to have just one it stopped behaving
as it should, as if it was missing the update, while having them
duplicated it works properly.
Therefore it appears to be better to keep them for now, since the
duplicated events will exit early as soon as they see that the monitor
counts match (on the first event the counts don't match so it
adds/removes the monitor and the following events see that the counts
match).
When there is a monitor disconnect/reconnect, usually it produces
multiple monitor events along with it, like the monitor resolution
change and the work area change. With the bound set to 1 it would
sometimes result in missed events.
This commit increases the bound to 20 to prevent this from happening.
This commit makes sure that if the state on file isn't up to date with
the expected `State` struct (maybe after an update) it doesn't panic
komorebi entirely, instead it ignores the state and continues with a
clean state.
Jerry Kingsbury