Fixes desktop window losing the keep-below flag

Just saying: this doesn't make sense. Keep-below doesn't matter for a desktop window. I just checked KWin's source code and the layer of the desktop window is only determined by the window type. Keep above/below has no influence on desktop windows. (For those interested kwin.git file abstract_client.cpp method belongsToLayer())

Overall the proposed change worries me. I totally believe you that it fixes the problem and also think that the current code is wrong. What worries me is that we don't understand what's going on here in a very central part of our desktop application. We don't know when we set the window type. That's urgh.

I would say the ensurePlatformWindow needs to be called from the QPlatformSurfaceEvent::SurfaceCreated event. But we should test that whole thing.

What's really important to know is that KWin does not support changes of window type once a window is mapped. This makes the existing code pointless - it cannot update. But it also means that somewhere there is a race that the window type is not properly set.

snip from xprop -spy on desktop window when clicking on desktop window and other windows:

_NET_WM_USER_TIME(CARDINAL) = 894873
_NET_WM_USER_TIME(CARDINAL) = 894853
_NET_WM_WINDOW_TYPE(ATOM) = _NET_WM_WINDOW_TYPE_DESKTOP
_MOTIF_WM_HINTS(_MOTIF_WM_HINTS) = 0x2, 0x1, 0x0, 0x0, 0x0
WM_HINTS(WM_HINTS):
                Client accepts input or input focus: True
                Initial state is Normal State.
_NET_WM_WINDOW_TYPE(ATOM) = _NET_WM_WINDOW_TYPE_DESKTOP
_NET_WM_STATE(ATOM) = 
_NET_WM_ALLOWED_ACTIONS(ATOM) = _NET_WM_ACTION_CHANGE_DESKTOP
_NET_WM_USER_TIME(CARDINAL) = 896471
_NET_WM_USER_TIME(CARDINAL) = 896471
_NET_WM_WINDOW_TYPE(ATOM) = _NET_WM_WINDOW_TYPE_DESKTOP
_MOTIF_WM_HINTS(_MOTIF_WM_HINTS) = 0x2, 0x1, 0x0, 0x0, 0x0
WM_HINTS(WM_HINTS):
                Client accepts input or input focus: True
                Initial state is Normal State.
_NET_WM_WINDOW_TYPE(ATOM) = _NET_WM_WINDOW_TYPE_DESKTOP
_NET_WM_ALLOWED_ACTIONS(ATOM) = _NET_WM_ACTION_MOVE, _NET_WM_ACTION_MINIMIZE, _NET_WM_ACTION_FULLSCREEN, _NET_WM_ACTION_CHANGE_DESKTOP, _NET_WM_ACTION_CLOSE
_NET_WM_STATE(ATOM) = _NET_WM_STATE_BELOW
_NET_WM_WINDOW_TYPE(ATOM) = _NET_WM_WINDOW_TYPE_DESKTOP
_MOTIF_WM_HINTS(_MOTIF_WM_HINTS) = 0x2, 0x1, 0x0, 0x0, 0x0
WM_HINTS(WM_HINTS):
                Client accepts input or input focus: True
                Initial state is Normal State.
_NET_WM_WINDOW_TYPE(ATOM) = _NET_WM_WINDOW_TYPE_DESKTOP
_NET_WM_STATE(ATOM) = 
_NET_WM_USER_TIME(CARDINAL) = 898190

it's probably due from the xcb platformintegration setting window type only after the window has been shown?

In D2121#39399, @drosca wrote:

snip from xprop -spy on desktop window when clicking on desktop window and other windows:

can you try that with starting plasmashell? My gut feeling is that it's first a different type and changed afterwards.

hmm xprop won't do that (lalala), maybe we can see it in xev, though

Patch for setting through surface event:

diff --git a/shell/desktopview.cpp b/shell/desktopview.cpp
index 14a25af..38c6621 100644
--- a/shell/desktopview.cpp
+++ b/shell/desktopview.cpp
@@ -226,6 +226,7 @@ bool DesktopView::event(QEvent *e)
             switch (pe->surfaceEventType()) {
             case QPlatformSurfaceEvent::SurfaceCreated:
                 setupWaylandIntegration();
+                ensureWindowType();
                 break;
             case QPlatformSurfaceEvent::SurfaceAboutToBeDestroyed:
                 delete m_shellSurface;

In D2121#45267, @graesslin wrote:

Patch for setting through surface event:

No, that doesn't help, nor does calling ensureWindowType in QEvent::Expose.

As it is now, calling ensureWindowType in QEvent::FocusIn is pointless (and causes this bug), as proper window type is already set *after* QEvent::Show (ensureWindowType is called in showEvent) - that is after native window was created and exposed.