Alle 15:43, luned� 28 luglio 2003, hai scritto:
> Quoting Monge Maurizio ([EMAIL PROTECTED]):
> > Hello.
> >
> > Working on Qt/DirectFB code suggested me a couple of things about
> > DirectFB ( the port is beginning to work pretty well! give a look
> > at the latest screenshot at qt-directfb.sf.net ).
>
> Great!
>
> What about moving newer screenshots to the top? ;)

I have never been a great site designer...

>
> > First, it could be nice if an app could find a window at (x,y), or could
> > use some window-manager like policies about size, etc ...
> > Some time ago someone was speaking about a window manager, isn't it?
>
> You may want to read the todo section on the web site.
>
> > Second, in DirectFB 0.9.19 there is a bug in the generic fbdev driver
> > that makes unusable the multi-app core.The bug does not affect the sdl
> > backend either the matrox driver. I will give a look as soon as i can...
>
> Are you talking about the software renderer?

Yes. But sdl (that i guess to be using software rendering) is working well.

>
> > Third, it could be nice that a window with alpha channel could decide of
> > not receiving mouse event where alpha = 0, with this shaped window can be
> > implemented in Qt and receive selected events. Al that can be realized
> > with the attached patch, that add the "DWOP_SHAPED" flag (to be used with
> > DWOP_ALPHACHANNEL)
>
> I was thinking of that some time ago. Unfortunately the locking to read one
> pixel for each mouse movement may be too much overhead and a faster
> solution should be thought about.

Anyway for each mouse movent there is "rebuild_stack" for the mouse window, 
that i guess to be much more exepensive, isn't it?
If the surface is in system mem reading pixels is not expensive.
Since reading a pixel is so fast, for every surface there could be lock flag 
(just a var in mem, no locking object) that if set lock will lock the surface 
manager. If a surface has to be moved and the flag is set the surface manager
will shed_yield until the var is unset again (that should happan immediately).
This solution makes expensive waiting for lock, but makes very fast 
lock/unlock, and this could be perfect for reading 1 pixel.

>
> I will apply the patch now. However, DWOP_SHAPED has to be implemented for
> color keyed windows, too. You should think about using the primary
> pixelformat (pixel format of the layer, i.e. no DWCAPS_ALPHACHANNEL). You
> can implement shaped windows via color keying then which is way faster and
> consumes less video memory. XDirectFB uses color keying for shaped windows.

True.

Maurizio Monge



--
Info:  To unsubscribe send a mail to [EMAIL PROTECTED] with
"unsubscribe directfb-dev" as subject.

Reply via email to