Re: Probable bug in WGL implementation (AIGLX) of GLX calls in XWin -wgl

2012-10-16 Thread Jon TURNEY
On 01/10/2012 22:34, Tim Edwards wrote:
I have a tool I maintain called magic, a VLSI layout editor.  One
 of its nicer features is a graphics mode based on OpenGL.  Occasionally
 I generate Cygwin versions of it, and was delighted to discover on my
 last update of Cygwin that there is a support for hardware-accelerated
 OpenGL using some translation between GLX and WGL calls at the level
 of the X server.  I tried using this with my Cygwin version of magic,
 and for the most part it works.  But it does have the strange effect
 of overwriting the OpenGL window with contents of other windows.
 
My setup is very non-standard but works under Linux and OS-X.  The
 application is built as an extension of Tcl/Tk.  Because the application
 makes all the OpenGL calls from C routines, it generates a generic
 window using a call to Tk_CreateWindow(), and maps it using Tk_MapWindow().
 The returned window is then passed to glXMakeCurrent().  All of this
 works fine.
 
The window that is used for the OpenGL rendering is framed by
 scrollbars on the side and bottom that are canvas windows in Tk.
 What I am seeing is that any time the scrollbars are redrawn, the
 OpenGL window is over-drawn, looks like with the default Tk background
 gray color.  A similar thing happens if I pop a window on top of the
 OpenGL window;  when I pop it down, the image of the window remains
 in the OpenGL window.  I presume that in the way GLX is supposed to
 work, X11 has reserved pixmap space somewhere for the window, but
 once the call to glXMakeCurrent() has been made, the contents of this
 pixmap should not show up on the screen.  Yet that is what I am seeing.
 Any clue as to what might be going on? 

Yes, unfortunately.

The current implementation of GLX using WGL takes a few shortcuts, basically
anything that is drawn with OpenGL isn't composed into the screen, it's just
drawn on top of it.

This works well enough when the GLX window is a top-level window, or is
non-top-level and has no occluding relatives and is drawn after anything it
occludes, but mis-renders in more complex scenarios.

This is discussed a bit more in [1]

As mentioned there I have done a bit of work fix the mis-rendering in some
cases.  I can't quite tell from you description exactly what's going wrong, so
I am not sure if those changes are going to help in this particular case.

I have built a test release including the changes discussed there, available
at [2], if you would like to test if it makes things better/worse/no difference.

The proper solution is probably something like rendering the OpenGL to an
offscreen buffer, and then composing it into the un-occluded area of the
window, but that is considerably more complex to implement.

One I tarball up this version of magic, I can send a pointer to
 where it can be obtained if anybody wants to download it and test
 for the bug.

Thanks.  This would be useful as a test case for any future work to improve 
this.

[1] http://sourceware.org/bugzilla/show_bug.cgi?id=10472
[2] ftp://cygwin.com/pub/cygwinx/XWin.20121012-git-3807fe48a7282459.exe.bz2

-- 
Jon TURNEY
Volunteer Cygwin/X X Server maintainer

--
Unsubscribe info:  http://cygwin.com/ml/#unsubscribe-simple
Problem reports:   http://cygwin.com/problems.html
Documentation: http://x.cygwin.com/docs/
FAQ:   http://x.cygwin.com/docs/faq/



Re: Probable bug in WGL implementation (AIGLX) of GLX calls in XWin -wgl

2012-10-16 Thread Tim Edwards

Hello Jon,

Thanks for the detailed response.


The current implementation of GLX using WGL takes a few shortcuts, basically
anything that is drawn with OpenGL isn't composed into the screen, it's just
drawn on top of it.


I wouldn't want to sound too peevish, as I was quite happy to find that
OpenGL hardware acceleration was even possible, as it was not on the
previous version of Cygwin that I had downloaded.  That it fails in
obscure applications is sort of to be expected.


This works well enough when the GLX window is a top-level window, or is
non-top-level and has no occluding relatives and is drawn after anything it
occludes, but mis-renders in more complex scenarios.

This is discussed a bit more in [1]

As mentioned there I have done a bit of work fix the mis-rendering in some
cases.  I can't quite tell from you description exactly what's going wrong, so
I am not sure if those changes are going to help in this particular case.

I have built a test release including the changes discussed there, available
at [2], if you would like to test if it makes things better/worse/no difference.


I downloaded the link, ran the X server, ran my application, and get no
difference in the behavior.


The proper solution is probably something like rendering the OpenGL to an
offscreen buffer, and then composing it into the un-occluded area of the
window, but that is considerably more complex to implement.


The two problems with this approach are that (1) I have found more
buggy implementations in OpenGL servers by doing offscreen rendering;
and (2) this particular tool is a VLSI layout editor and must be
rendered directly on the front buffer.  The general approach is to
render everything as fast as possible and always be willing to break
on key interrupt to start over.  And I have misappropriated the back
buffer for backing store purposes. . .

The issue here is that I have the whole tool running as an extension
of Tcl/Tk.  I make a low-level call to Tk to give me an X11 window,
which is a simple frame window in Tk but also part of a grid of
sub-windows that includes a menu on top and scrollbars on side and
bottom.  I make a GLX call to render into the window I get from Tk.

There are two occasions when the window gets rudely overdrawn,
apparently by Tk (that is, the X server appears to know when not
to draw into the OpenGL window except when the drawing requests
come from the same process):  One is that if I raise the Tk console
window to obscure part of the OpenGL window, and then lower it behind
the OpenGL window, the interior contents (not the frame, therefore,
only those drawing requests sent to the server by Tk) continue to be
drawn onto the OpenGL window.  The second case is a little bit strange
to me, in that Tk apparently wants to draw the background of the left-
side scrollbar as a gray rectangle that covers not just the scrollbar
window but most of the OpenGL window, too.  It may be a Tk error that
it draws outside the bounds of its own sub-window, but in a correctly
working X server, it does not take precedence over the OpenGL window
contents.  I have found that if I disable the scrollbar, the effect
disappears.  So I can manage to work around everything (if necessary)
except for the obscuring window problem.


One I tarball up this version of magic, I can send a pointer to
where it can be obtained if anybody wants to download it and test
for the bug.


Thanks.  This would be useful as a test case for any future work
to improve this.


The program (Cygwin version) is a two-part install that includes an
X11-based version of Tcl/Tk for Cygwin:

http://opencircuitdesign.com/cygwin

Where tcltk_x11_w7.tgz is the 64-bit version that I was using when
I found the problem, and the VLSI layout tool is

http://opencircuitdesign.com/cygwin/magic.html

where the 64-bit Windows 7 version is magic-8.0.116w7.tgz.  The
direct download URLs are

http://opencircuitdesign.com/cygwin/archive/tcltk_x11_w7.tgz
http://opencircuitdesign.com/cygwin/archive/magic-8.0.116w7.tgz

The latter one installs a shell script /usr/local/bin/magic that
launches the layout tool.  Use magic -d OGL from a Cygwin xterm to
get the OpenGL-based version.  The error can be seen by alternately
raising the Tk console window and the layout window, with the contents
of the console window continuing to be drawn after the window is pushed
under the OpenGL window.
Regards,
Tim

++-+
| Dr. R. Timothy Edwards (Tim)   | email: t...@opencircuitdesign.com|
| Open Circuit Design, Inc.  | web:   http://opencircuitdesign.com |
| 22815 Timber Creek Lane| phone: (301) 528-5030   |
| Clarksburg, MD 20871-4001  | cell:  (240) 401-0616   |
++-+

--