A workspace is occupied if it is not active and there is at least one window managed by the WM (`_NET_CLIENT_LIST`) that has set `_NET_WM_DESKTOP` to that workspace.
The behavior when `_NET_WM_DESKTOP` is not set is not yet clear but this is unlikely to happen since most WMs will position windows on some desktop.
Closes#874Fixes#1444Fixes#1033
* Set Desktop OCCUPIED if a window moves there
This covers more of an edge-case. I did this first by accident, it might
vanish later on.
* Replace tracking change of WS with currently used WS
* Untrack occupied workspaces
* Track windows and their desktops in pairs
* Match type of occupied_desktops with current_desktop
Because the index needs to be matched later on, type mismatches would be non-ideal.
* Recreate the occupied desktops everytime and remove duplicates
* Readd support for moving windows to other desktops
* Use less characters to empty the vector
* Rename variable storing the desktops
* Recount windows on every occasion
This alone simplifies the management and the lookup for occupation of a
workspace
* Keep track of number of windows in every workspace
* Add debugging output that shall be removed before merging
* Remove obsolete TODO
* m_client_list should always be diff'd, since the desktop may change
Therefore we update the desktop-count tally every time the client_list
changes. It may just be a desktop-change without a change of
clients.size()...
* Add more logging-spam to understand window/desktop lifecycle
* Lock event-handler to serialize handling of events
* Fix occupied workspace counting and change to bool array
Also, performance improvements when diffing new and old client lists
* Fix crash when all clients are removed
* Conform to linter and styleguide
* Shorten conditional as it is standard enough
Since this only guards against 0-divisions, it can be shortened
without risking too much confusion down the road.
* Guard against multiple threads accessing and modifying data
Fixes#1444
Modification of internal data happens through the handle-method, while
the build-method tries to access the data structures for display. Since
some modifications clear e.g. the m_viewports, references may become
invalid between looping over them an accessing them.
The mutex should guard against this simultanuous access.
* Do not 'adopt_lock', because calls come from very different threads
To my understanding, adopt_lock has some dependency on the mutex-ownership. Since
the lock is once called from the inside (in handle) and once from the outside (in
build), there might be a problem. After brief testing, the segfaults happened fewer
times.
See #1444
* Also listen to _NET_WM_DESKTOP
In order to move a window from one desktop to another, it is sufficient
to set the desktop-property of that window. xmonad fires a lot of events
in the case of moving a window, herbstluftwm only updates the
_NET_WM_DESKTOP-atom of the window.
This change reloads the clientlist in order to correctly set the
desktop state "occupied".
* Describe need and use of mutex
It might be possible to relieve the guard in xworkspaces_module::handle,
but I am unsure about this. Since xmonad emits a lot of events on almost
every minor change, I would let the guard keep its post, avoiding
race-conditions in event-handling.
* Give temporary variables better names
* Clarify purpose of loop
About 80% of this comment are taken from
https://github.com/jaagr/polybar/pull/882#discussion_r255317363
* Remove merge-remainder
* Use a simpler method to list occupied desktops.
Co-authored-by: Jérôme Boulmier <jerome.boulmier@outlook.fr>
* Document m_clients field
