Re: Some questions regarding _gdk_window_impl_new() (csw, win32, patch)

[Hans Breuer <hans breuer org>]

At 24.08.2010 10:14, Alexander Larsson wrote:
> On Sat, 2010-08-21 at 14:09 +0200, Hans Breuer wrote:
> > while trying to fix _gdk_window_impl_new() for win32 some more,
> > quite some questions arose around that (apparently undocumented;)) API:
> >
> >    - if the information given to _gdk_window_impl_new() is redundant,
> >      as with GDK_WA_VISUAL set and visual given by parameter, which one
> >      has precedence?
>
> Use the visual parameter, it is based on GDK_WA_VISUAL if set already.
So I could assert on it ...

> From attributes the only things used is: colormap, title, wmclass and
> type_hint.
Will check. Again it might be usefult to assert on nothing else being set?
Or maybe not, because the canconical implementation also uses GDK_WA_NOREDIR ;)

After adding explict checks for remaining atrributes I wonder if 
GDK_WA_CURSOR should explicitly note be used, although passed into 
_gdk_window_impl_new()? It is set by e.g. testgtk::(color selection).

> > - when turning a csw into a native window, shouldn't the new window
> >      position be calculated against the native parent? If so: should
> >      the caller do it and give GDK_WA_(X|Y)?
>
> Ignore the passed in position and size in attributes. The window
> position is already calculated for the csw window, and all you need to
> do is offset it to reach the nearest native window, so the position
> should be:
>    private->x + private->parent->abs_x,
>    private->y + private->parent->abs_y
Thanks, doene and pushed.

> To explain this, private->x/y is the offset from the parent GdkWindow to
> this window, whereas private->abs_x/y is the offset from the "nearest"
> native window to the window. We can't use private->abs_x/y here, because
> it is zero for a native window.
>
> And the size is always private->width x private->height.
>
> None of this is very obvious, nor documented. Just look at how the x
> backend does it, as that is the canonical implementation.
I usually do, but this time I was a bit lost before in all the xattributes 
settings. Thanks for the explaination, it gives much simpler code than I 
had in my patch.

> > - how to get back default OS/window-manager placment of toplevel
> >      windows? (The old implementation used to place toplevel windows
> >      with CW_USEDEFAULT, my new implementation is doing similar)
>
> Not sure what you mean here. Can't you just pass CW_USEDEFAULT for toplevels?
Yes, this is what I meant and did in the patch. I just wanted to have it 
confirmed.

> >    - what's the (visual) depth of a GDK_INPUT_ONLY window?
> >      ( Dia(win32) with gtk>=2.18 is crashing on it's Navigation popup,
> >       which can be fixed by setting the visual depth to something
> >       reasonable, but it used to be 0 before csw. Maybe this is hiding
> >       a bug elsewhere, e.g. returning the wrong window when forced to
> >       be native? )
>
> The visual for input-only windows is 0, but this is set by the common
> code in gdk_window_new, so its not something the backend should have to
> care about.
To rephrase my question: where should I start searching for the 
incompatibility introduced with csw? For some reason Dia tries to create a 
pixmap with the visual depth of the GDK_INPUT_ONLY window.
The fatal attempt to create a pixmap with depth=0 can be overcome by 
setting depth=visual->depth for input-only windows.
It also can be avoided with the attached patch, but both approached look 
quite wrong. Do you have a better idea?

> > Another question is regarding "GDK_NATIVE_WINDOWS". With my current
> > patch at least the drawing seems to work reasonable again, but the
> > (ENTER|LEAVE)_NOTIFY handling is not, so the mouse pointer is almost
> > useless. Is it necessary to ressurect most of the original code (pre-csw)
> > to make it work again, or should there be an easier way?
>
> I'm not sure exactly how enter/leave notification works on win32, but
> yeah, GDK_NATIVE_WINDOWS more or less passes on events as they arrived
> from the windowing system,
Unfortunately not from my understanding. The (ENTER|LEAVE)_NOTIFY handling 
had to be emulated on win32, with TrackMouseEvent. Some initial fix is
available in the gtk-2-22 branch.
But the "automatic grab" needed for menu highlighting is still not working 
again (and I'm unsure if I should continue trying).

> rather than doing the csw rewriting magic.
> GDL_NATIVE_WINDOWS is really more of a hack for apps like eclipse and
> lotus notes that need compatibility with old gdk behaviour when they do
> really weird x stuff.
>
> I don't think supporting it on win32 is very useful. I mean, what is the
> expected behaviour? The natural thing would be to be backwards
> compatible with the exact old pre-csw mouse pointer code, but are any
> apps really depending on that exact behaviour?
Maybe not, but there are at least two bug reports regarding native windows 
on win32. About positioning:
  https://bugzilla.gnome.org/show_bug.cgi?id=625972

Also the drawing seemed to be mostly broken before my patch.
I hope the just pushed changes are also fixing
  https://bugzilla.gnome.org/show_bug.cgi?id=628049

Thanks,
	Hans

-------- Hans "at" Breuer "dot" Org -----------
Tell me what you need, and I'll tell you how to
get along without it.                -- Dilbert

diff --git a/gdk/win32/gdkwindow-win32.c b/gdk/win32/gdkwindow-win32.c
index 53e19d9..61cd9b9 100644
--- a/gdk/win32/gdkwindow-win32.c
+++ b/gdk/win32/gdkwindow-win32.c
@@ -39,6 +39,9 @@
 static GdkColormap* gdk_window_impl_win32_get_colormap (GdkDrawable *drawable);
 static void         gdk_window_impl_win32_set_colormap (GdkDrawable *drawable,
 							GdkColormap *cmap);
+
+static GdkDrawable* gdk_window_impl_win32_get_source_drawable (GdkDrawable  *drawable);
+
 static void gdk_window_impl_win32_init       (GdkWindowImplWin32      *window);
 static void gdk_window_impl_win32_class_init (GdkWindowImplWin32Class *klass);
 static void gdk_window_impl_win32_finalize   (GObject                 *object);
@@ -135,6 +138,9 @@ gdk_window_impl_win32_class_init (GdkWindowImplWin32Class *klass)
 
   drawable_class->set_colormap = gdk_window_impl_win32_set_colormap;
   drawable_class->get_colormap = gdk_window_impl_win32_get_colormap;
+
+  //FIXME: why do we need this and gdk-x11 does not?
+  drawable_class->get_source_drawable = gdk_window_impl_win32_get_source_drawable;
 }
 
 static void
@@ -232,6 +238,12 @@ gdk_window_impl_win32_set_colormap (GdkDrawable *drawable,
     }
 }
 
+static GdkDrawable* 
+gdk_window_impl_win32_get_source_drawable (GdkDrawable  *drawable)
+{
+  return _gdk_root;
+}
+
 void
 _gdk_root_window_size_init (void)
 {