Needed due to weird hang (sometimes crash) when creating (probably only when
done rapidly?) same relocatable setting more than once.
(noticed when implementing workspace settings cycling)
Clutter's transition framework will delay painting the ease until some sort of
painting is requested or whatnat, even if duration == 0.
In particular on reload this means clones are invisible for quite a while.
Simply force positions on startup.
There's a few differences in how onComplete is run:
- We need to run moveDone on every clone transition, as the transitions aren't
done when cloneContainer's transition finishes
- We cannot run layout until startup is done. For some reason onComplete runs
before monitor setup is done properly.
squash! Use clutter animations
- Can't use addTween as a timeout, but transitions are now finished when
onComplete is run, so there's no need for this anymore.
Floating windows tend not to trigger size-changed at all. Make sure we position
the clone correctly at least once.
Might of course be better to just run allocateClone when the actor is allocated
still, but I'm not entirely sure if the frame and actor is necessarily synced at
that point.
Instead of listening on the actor's allocation signal set the clones size when
the windew is resized. This works around 3.34 not allocating hidden actors.
In general: NPE in stackoverlay when target window is non-`TabList.NORMAL`
crashes `insertWindow` midway. (eg. size-changed signal never connected).
A `TabList.NORMAL` window is focusable and doesn't `mw.skip_taskbar`.
Example: The main window of pycharm on startup is (sometimes) non-normal(!).
Focusing the window fail so it's likely to become the stackoverlay target.
(java applications is a WM's worst nightmare..)
Seem the ordering of "focus", "window-created", "window-added" and "first-frame"
and the corresponding update of eg. global.display.focus_window and the mru list
is quite complicated and non-deterministic..
Some sampling reveals the following possible combinations on 3.28 X11:
fw = global.display.focus_window, mw = metaWindow in insertWindow (new window)
fw = mw mw = mru0 !fw fw = mru0 fw = mru1
false false false false false
false false false true false
false true false false true
false true true false false
true false false false true
true true false true false
So it's quite messy, but we can (hopefully) always rely on the currently
focused ("adult") window to be in the mru.
Note: the reason we doesn't simply use `maximize(Meta.MaximizeFlags.HORIZONTALLY`
for toggle-maximize-width is that maximized windows are unmovable [1] and (more
debatable) that maximized width should honor minimumMargin.
[1] doesn't play to well with paperwm, though we have some hacks in place for
regular maximized windwos
Ref: #157
Windows sometimes was registered (`registerWindow`) twice on startup (observed
on X11 3.28) causing all sorts of problems.
Seems like the extension sometimes is enabled/disabled twice on startup causing
the 'startup-complete' signal to be connected twice. Use the `Signals` class to
manage the signal to guard against this scenario.
I added a one-shot function to `Signals` since it's an error to disconnect an
already disconnected signal and that would happen on disable if the
'startup-complete' signal disconnected the signal itself.
assoc: d8069078f2
Ole Jørgen Brønner