bug#48809: 28.0.50; Missing Lucid/GTK2 graphical elements on tiling WM

[martin rudalics]

> Further to the tangent in bug#48229, the menu, tool, and scroll bars are
> sometimes not drawn when creating new Lucid/GTK2 frames under the tiling
> WM Xmonad.  The missing elements reappear after the affected frames are
> resized.
>
> I cannot reproduce any of these issues with GTK3/PGTK, but the Lucid
> behaviour is present since at least Emacs 24.5.1 (I don't have old GTK2
> builds at hand).  Affected Emacs 28 configurations follow my signature.

Thanks.  The only thing that is evident from your examples is that only
toolkit elements are affected.  Elements Emacs draws itself are not
missing in your examples.  I suppose the behavior you see happens
because Xmonad rejects a first size proposal of Emacs for these elements
and does not draw them, notifies Emacs that it wants different sizes for
these elements and that notification gets lost - somehow and at least
sometimes.  I have no good explanation why things work with GTK3 and not
with GTK2 though.  Are all behaviors you reported reliably reproducible?

One thing I'd like to exclude is that this is yet another instance of
Tom Gillespie's observation that Emacs does not play well with window
managers that do not set _NET_WM_STATE.  x_get_current_wm_state
apparently fails on such systems and we probably never (or at least not
always) get a not_hidden frame in the MapNotify case and have to wait
for a ConfigureNotify instead (which your remark about the reappearance
of missing elements after resizing would confirm).

To check this please do (setq frame-size-history '(100)) and proceed
with some of your broken behaviors.  I'd be interested if when you then
do (frame--size-history) and look into the *frame-size-history* buffer
you can see

"MapNotify, not hidden & not iconified"

entries.  Alternatively, you could set a watchpoint at xterm.c's

          bool not_hidden = x_get_current_wm_state (f, event->xmap.window, &value, 
&sticky);

and see whether a value of true is ever returned.

One thing you could try is to change

      if (v == 1 && f->visible != 1)
        redisplay_other_windows ();

to

      if (v == 1)
        redisplay_other_windows ();

in frame.h's SET_FRAME_VISIBLE so maybe Emacs would send the modified
elements (modified because the WM changed their size to something Emacs
didn't ask for) again.  Honestly, I have no hope that this would help.

martin