win32: hang after main window close
Brought to you by:
mcfletch
Menu
▾
▴
Logged In: YES
user_id=34901
Weird thing is that I can't see *how* PyOpenGL 2.x fixed it
when I spelunk through the code. It seems to have "just
worked". I've added code to "fix" it, basically I start a
.5s timer in glutMainLoop() that checks for a NULL context
and kills the process if it's discovered.
Anyway, give it a try with latest CVS and let me know if
it's suitable for you (or if you have a better approach in
mind).
Logged In: YES
user_id=1455846
Yeah, this works, i tried a very similar fix (although i did
it in my app), but it seems very much like a hack.
It seems the question is not how 2.x fixed it (apparently it
didn't address this altogether), but what does 3.x wrong to
stall the process. A normal GLUT program doesn't need to do
this, so why should a python wrapper need this check? Looks
to me like somewhere something should be cleaned up, or
weakref'd or something similar to allow the process to die,
but i couldn't find anything yet...
Logged In: YES
user_id=11105
The reason for the process staying alive is that on Windows
glut.h contains special code in the header file that must be
reimplemented in ctypes. The "glutCreateWindow", for
example, is redefined like this:
#if defined(_WIN32) && !defined(GLUT_DISABLE_ATEXIT_HACK)
GLUTAPI int APIENTRY __glutCreateWindowWithExit(const char
*title, void (__cdecl *exitfunc)(int));
#ifndef GLUT_BUILDING_LIB
static int APIENTRY glutCreateWindow_ATEXIT_HACK(const char
*title) { return __glutCreateWindowWithExit(title, exit); }
#define glutCreateWindow glutCreateWindow_ATEXIT_HACK
#endif
In other words: Instead of calling the glutCreateWindow
function you must call __glutCreateWindowWithExit, and pass
a pointer to the standard 'exit' function as additional last
parameter.
I have attached a patch which demonstrates how this could be
fixed for the glutCreateWindow() function, glutInit() and
glutCreateMenu() must probably be changed as well. The
patch is not really correct, thought, since it hardcodes the
C library dll name as 'msvcr71', which may not be correct,
depending on how Python itself was built.
PS: Seems I'm not allowed to attach a patch. Here it is (to
be applied to OpenGL\raw\GLUT\__init__.py):
--- __init__.py~ Mon Nov 13 10:54:38 2006
+++ __init__.py Mon Nov 13 11:03:35 2006
@@ -159,15 +159,29 @@
argNames=('window', 'x', 'y', 'width', 'height'),
)
+import os
-# /usr/include/GL/freeglut_std.h 398
-glutCreateWindow = platform.createBaseFunction(
- 'glutCreateWindow', dll=platform.GLUT, resultType=c_int,
- argTypes=[STRING],
- doc='glutCreateWindow( STRING(title) ) -> c_int',
- argNames=('title',),
-)
+if os.name == "nt":
+ __glutCreateWindowWithExit = platform.createBaseFunction(
+ '__glutCreateWindowWithExit',
dll=platform.GLUT, resultType=c_int,
+ argTypes=[STRING, c_void_p],
+ doc='glutCreateWindow( STRING(title) ) -> c_int',
+ argNames=('title',),
+ )
+ libc = CDLL("msvcr71")
+
+ def glutCreateWindow(title):
+ __glutCreateWindowWithExit(title, libc.exit)
+
+else:
+ # /usr/include/GL/freeglut_std.h 398
+ glutCreateWindow = platform.createBaseFunction(
+ 'glutCreateWindow', dll=platform.GLUT,
resultType=c_int,
+ argTypes=[STRING],
+ doc='glutCreateWindow( STRING(title) ) -> c_int',
+ argNames=('title',),
+ )
# /usr/include/GL/freeglut_std.h 442
glutDestroyMenu = platform.createBaseFunction(
fix to special.py
Logged In: YES
user_id=1455846
Very nice, thanks for the input! I attached a patch that
does this in GLUT/special.py (as the comment at the top of
raw/GLUT/__init__.py says "do not edit"). My patch also
doesn't hardcode msvcr71 but uses sys.exit with a ctypes
callback instead. btw, is it preferred to check on os.name,
instead of sys.platform? i usually use the latter...
Logged In: YES
user_id=1455846
Second version of the patch does this for all
__glut*WithExit functions.
2nd version
Logged In: YES
user_id=1455846
Arr, too quick with the posting, glutCreateMenu is probably
a bit more complicated, because its already overwritten in
special.py, this needs some more attention.