Re: [Plib-users] Re: flightgear problem

[Stephen L. <le...@na...>]

Steve Baker,
Thank you for a good answer. Response embedded...
Stephen Lewis

Steve Baker wrote:
> 
> Stephen Lewis wrote:
> >
> > Gentlemen,
> > I have joined this list because I have the "flightgear" problem.
> > I have just read a weeks worth of discussion concerning the
> > "ssgInit called without a valid OpenGL context". I tried to compile
> > FlighGear 0.7.8 on a PowerPC running YellowDogLinux 2.0.
> > Although based on RH 7 there is no MMX and the compiler is 2.95.3
> 
> That's interesting new information.  Which OpenGL version do you have?

I believe GL version 1.2 from Mesa 3.4

> 
> > There has been much speculation and not much science in this discussion.
> 
> Well, there has been science to the extent that I personally can provide
> science.  The facts I have to date are:
> 
>   * glXGetCurrentContext is returning NULL
>   * ...which is *SUPPOSED* to mean that there isn't a
>     valid OpenGL rendering context.
>   * But we demonstrably *DO* have a rendering context at that
>     point because we can comment out the glXGetCurrentContext and
>     the program runs correctly.
>   * The exact same sequence of instructions when run on other OpenGL
>     implementations (eg nVidia's) and/or other windowing systems (eg Windoze)
>     and/or different C++ compilers (eg 2.95.2) *DOES* result in a valid
>     result from glXGetCurrentContext.
>   * This same problem shows up in other OpenGL programs that test
>     glXGetCurrentContext...not just PLIB programs.
>   * Up until today, every instance of the problem had been on
>     systems running Mesa under RH 7.x or Mandrake 8.x.  I've never
>     heard of such a system that worked - or of any other system that
>     didn't work.
> 
> So, what are we to deduce from these facts?
> 
> I'm inclined to believe therefore that this is some kind of an underlying
> bug in the windowing system or OpenGL implementation...probably the latter...
> caused by *something* unique to RH implementations (and other systems that
> share something with it).
> 
> I'm currently working under the presumption that the "known-to-be-dubious"
> C++ compiler (GCC 2.96/2.97) is mis-compiling Mesa in some subtle manner
> that causes glXGetCurrentContext to fail.  That theory seems to fit the
> facts better than anything else that's been suggested - but it's certainly
> not proven.
> 
> The scientific method says that the next thing to do is a controlled experiment
> to test predictions made by this theory.  But since I run SuSE and have only an
> nVidia graphics card, I don't see the problem, I can't easily/cheaply do that.
> 
> Since this is plainly not a problem with anything I've written, it's really
> not my problem to solve.  The issue belongs (IMHO) in the Mesa mailing list...

I agree with this and will forward my findings to them.
The problem seems to me to be in 'glXGetCurrentContext'

> however, I'm subscribed to that list and so far I don't recollect anyone
> complaining there.
> 
> In the realm of OpenSource software, if you don't tell the right person
> the right problem, it won't ever get fixed.  (Although it may just "go away"
> of it's own accord - as I suspect this will).
> 
> So - that was before seeing your email.
> 
> Now we see another RedHat 7 derivitive with the problem - but NOT with
> the 2.96/2.97 compiler.  That's interesting - and probably deflates my
> "broken compiler" theory...although not entirely since most non-RH x86
> users are on 2.95.2 and you are using 2.95.3 - and on a different CPU
> architecture.
> 
> What other things might be common between these broken systems?

I have taken a peek at the source for 'glXGetCurrentContext' and its
full of #ifdefs and on my platform appears to return NULL regardless.
The differences between implementations is probably local #defines
and config files including such things as DIRECT/INDIRECT rendering,
PTHREADS, and complicated deferred execution and mutex code between
X and GL. I'm going to wade through the code but its ugly.

> 
> What Xfree version are you all running?  Is anyone else running that
> version with the same version of Mesa and NOT seeing the problem?
> 

Mine is XFree86 4.0.2-6e built by YDL (terrasoft)

> > I compiled the plib_examples and get the same message from
> > examples/src/ssg/tux/tux_example and examples/src/ssg/viewer/viewer
> > Further investigation indicates that *after* calling 'glutCreateWindow'
> > there is no valid OpenGL context according to 'ssgInit'. How does
> > ssgInit
> > determine this? It calls 'glXGetCurrentContext' which returns NULL even
> > though 'glutCreateWindow' thinks it has created a valid context..
> > Thus the root of the problem is an incompatibility between:
> > 'glutCreateWindow' and 'glXGetCurrentContext'.
> 
> Actually, you can get the same bug using 'freeglut' and in PPE which
> doesn't use GLUT at all.  Hence it's not glutCreateWindow itself
> but whatever underlying X11 call is common to FLTK and GLUT/freeglut...
> XCreateWindow being the most likely candidate.

I'll check that - maybe can get problem without GLUT - just within
GLX - that would narrow it down a bit.

> 
> > 'ssgInit' doesn't
> > have much to do with it except that it generates the error message.
> 
> Indeed.  Which is why this is the wrong mailing list to report the
> problem to.

True - but there was a lot of discussion and (unfortunately) you are
going to get the calls because you generate the message :-)

> 
> > It is easier to debug a small program with these two calls than
> > to begin with FlightGear...
> 
> Yes - something like that would be useful.  Unfortunately, I can't
> write that program because I don't have a system that exhibits the
> bug.

I can. see attached

> 
> > Now my question, to which groups should this information be directed?
> 
> I suggest the Mesa mailing list.  You can find info on how to subscribe
> on www.mesa3d.org.  I'd greatly appreciate it if you would first write
> that 10 line test program so that the problem can be expressed in a
> manner that won't result in it being immediately reflected back onto
> this list as a PLIB problem.

I couldn't get it into 10 lines but its under 60. No PLIB
code either.

> 
> > Platform Macintosh G4 Cube and YellowDogLinux 2.0
> > OpenGL 1.2, Mesa 3.4 and GLUT 3.7
> 
> What 3D hardware do you have?

apparently "ATI Rage 128 Pro PF (AGP)"

> 
> ----------------------------- Steve Baker -------------------------------
> HomeMail : <sjb...@ai...>   WorkMail: <sj...@li...>
> HomePage : http://web2.airmail.net/sjbaker1
> Projects : http://plib.sf.net       http://tuxaqfh.sf.net
>            http://prettypoly.sf.net http://tuxkart.sf.net
>            http://freeglut.sf.net   http://toobular.sf.net

=============================cut
here=======================================
/* Copyright (c) 2001 Stephen Lewis */
/* modified 31-Aug-2001 */
/*

  Illustrates a problem with OpenGL context
  comments please to 'le...@na...'

gcc -O2 -Wall   -c -o main.o main.c
gcc -O2 -Wall  -lglut -lGL -lGLU -lm  -o main main.o
./main

*/

#include <stdlib.h>
#include <stdio.h>
#include <GL/glut.h>
#include <GL/glx.h>

void reshape ( int w, int h )
{
  glViewport ( 0, 0, w, h ) ;
}

void keyboard ( unsigned char key, int x, int y)
{
  exit ( 0 ) ; /* exit for any key */
}

void redraw ()
{
  glClearColor( 1.0, 0.5, 0.0, 0.0 );
  glClear  ( GL_COLOR_BUFFER_BIT | GL_DEPTH_BUFFER_BIT ) ;
  glutSwapBuffers ();
}

int main(int argc, char **argv) {
  int   fake_argc = 1 ;
  char *fake_argv[3] ;
  fake_argv[0] = "SL_Test" ;
  fake_argv[1] = "SL Example Program." ;
  fake_argv[2] = NULL ;

  /*
    Initialise GLUT
  */

  glutInitWindowPosition (10, 10);
  glutInitWindowSize     ( 640, 480 ) ;
  glutInit               ( &fake_argc, fake_argv ) ;
  glutInitDisplayMode    ( GLUT_RGB | GLUT_DOUBLE | GLUT_DEPTH ) ;
  glutCreateWindow       ( fake_argv[1] ) ;
/* without next two callbacks - get fatal error from glutMainLoop */
  glutDisplayFunc        ( redraw   ) ;
  glutReshapeFunc        ( reshape  ) ;

  glutKeyboardFunc       ( keyboard ) ;

  /*
    Check the context
  */

  if (glXGetCurrentContext() == NULL) {
        printf("SL_Test: glXGetCurrentContext returns NULL\n");
        printf("SL_Test: This is incorrect\n");
  } else {
        printf("SL_Test: glXGetCurrentContext returns SOMETHING\n");
        printf("SL_Test: This is (presumably) the correct context\n");
  };

  glutMainLoop ();

  return 0;
}
=============================cut
here=======================================