"Don't do that then" (http://www.catb.org/~esr/jargon/html/D/Don-t-do-that-then-
.html)

GTK doesn't handle switching screen resolution on the fly on X11, either. 
(Yeah, I know that in traditional X servers this isn't even possible. But isn't 
there nowadays some extension that makes it possible?) More importantly, 
neither does any GTK application.

Is it too harsh to resolve as NOTABUG? Or maybe there already is a more generic 
enhancement request that this could be duped on.

GTK+ does handle them since 2.2 at least in sofar as there is a
GdkScreen::size-changed signal, which gets emitted when the screen resolution
changes. Does the win32 backend not emit that signal ?

Oops, hadn't noticed. No, gdk/win32 does not emit that. Will have to implement 
it. But would that help the original poster's problem, does gtk use that signal 
for anything? Or is it enough if GDK's idea of root window size is changed?

I haven't played around with RANDR on X, really. Thus I can't say if emitting
::size-changed will make the menus behave better, but I think getting that
signal would be the prerequisite to fixing the menu behaviour, if necessary.

A further complication on Windows is that it's possible to switch screen depth, 
too, not just resolution. That's not possible with RANDR. But it probably is 
safe to just ignore that issue, a gtk/win32 application continues to work even 
if the screen depth changes.

Should there also be a monitors_changed (or whatever) signal, to indicate that 
the number and/or positioning of the monitors that the screen consists of has 
changed? Or does size_changed also cover that?

I think XRANDR (the spec) also allows to change depth, but that part of the spec
hasn't been implemented yet. Is it possible to change the number of monitors on
the fly on win32 ? Standard X doesn't provide for changing the number of screens
and Xinerama doesn't provide for changing the number of monitors, AFAIK. If this
is possible on win32, then a GdkScreen::monitors-changed signal would make a lot
of sense, I'd say. (Though handling that in gtk would mean catering for 
0.2 - 0.5 % of the GTK user base, I guess)

> Is it possible to change the number of monitors on the fly on win32 ?

Yes, and their relative position.

Looking further at the X11 backend code, it does call init_xinerama_support() 
in _gdk_x11_screen_size_changed(), so I guess this means that size_changed is 
intended to cover also monitor positioning (or even number) changes.

Another thing: The X11 backend doesn't seem to set the width and height fields 
of it's per-screen root_window's GdkWindowImplX11 when it handles a screen size 
change? Should it?

Hmm, init_xinerama_support() is called from there to update the size information
for the monitors, I don't think it implies that ::size-changed was intended to
cover monitor geometry changes. But it certainly makes sense to reinterpret the
signal in this way.

You are right about the root_window size, though it may be that we get a
ConfigureNotify for the root window, and pick up the new size from that. I'll
experiment.

Yea, the root window seems to pick up the new size via a configure event.
I'm attaching a small test program.

Created attachment 31201 [details]
test GdkScreen::size-changed

Created attachment 31202 [details] [review]
Suggested patch, untested yet

The test program needs to actually create a window to receive any WM_* messages 
on Windows, just creating and showing an empty toplevel window is OK. There was 
a minor bug in the patch (forgot to use the "impl" field of GdkWindowObject), 
but otherwise it seems to work. Applied to HEAD and gtk-2-4.

Thanks a lot. The response on my bug report has been incredible. Cheers to you all!