PyOpenGL

V. Armando Sole wrote:
...
> The following code was working fine on PyOpenGL 2.0.1.09:
>
>                  GL.glVertexPointerf(self.vertices)
>                  GL.glColorPointerf(self.vertexColors)
>                  GL.glEnableClientState(GL.GL_VERTEX_ARRAY)
>                  GL.glEnableClientState(GL.GL_COLOR_ARRAY)
>                  GL.glDrawArrays(GL.GL_POINTS, 0, xsize*ysize)
>   
...
> Just changing glColorPointerf(self.vertexColors) by 
> glColorPointerd(self.vertexColors) solved my problem.
>   
Okay, that's weird.  I've just written a test like so (against current CVS):

    def test_numpyConversion( self ):
        """Test that we can run a numpy conversion from double to float 
for glColorArray"""
        import numpy
        a = numpy.arange( 0,1.2, .1, 'd' ).reshape( (-1,3 ))
        glEnableClientState(GL_VERTEX_ARRAY)
        glColorPointerf( a )
        glColorPointerd( a )

and it produced no errors.  Can you produce a minimal test that shows 
the error occurring?  The system *should* be able to convert from double 
to float (as shown with this test).  Could be that it's some bug I don't 
recall fixing, but that doesn't seem likely.

Good luck,
Mike

-- 
________________________________________________
  Mike C. Fletcher
  Designer, VR Plumber, Coder
  http://www.vrplumber.com
  http://blog.vrplumber.com

Hello,

Quoting "Mike C. Fletcher" <mcfletch@...>:

>  ... Can you produce a minimal test that shows 
> the error occurring?  The system *should* be able to convert from double 
> to float (as shown with this test).  Could be that it's some bug I don't 
> recall fixing, but that doesn't seem likely.
> 

I send you a file showing the problem.

python OpenGLTest.py 0 uses float32
python OpenGLTest.py 1 uses float64 and crashes
python OpenGLTest.py 2 uses float64 and glColorPointerd and does not crash.

My platform is windows XP. The problem is present on Python 2.4 and Python 2.5

BTW, glReadPixelsub(x, y, 1, 1, GL_RGBA) returns 'int8' instead of 'uint8' and I
am forced to implement things like:

color = GL.glReadPixelsub(x, y, 1, 1, GL.GL_RGBA)
#workaround a PyOpenGL bug?
if color.dtype == 'int8':
    color = color.astype(numpy.uint8)

Best regards,

Armando

-------------------------------------------------
This mail sent through IMP: http://horde.org/imp/

sole@... wrote:
> Hello,
>
> Quoting "Mike C. Fletcher" <mcfletch@...>:
>
>   
>>  ... Can you produce a minimal test that shows 
>> the error occurring?  The system *should* be able to convert from double 
>> to float (as shown with this test).  Could be that it's some bug I don't 
>> recall fixing, but that doesn't seem likely.
>>
>>     
>
> I send you a file showing the problem.
>
> python OpenGLTest.py 0 uses float32
> python OpenGLTest.py 1 uses float64 and crashes
> python OpenGLTest.py 2 uses float64 and glColorPointerd and does not crash.
>
> My platform is windows XP. The problem is present on Python 2.4 and Python 2.5
>   
The problem was actually in the caching of the temporary array.  Under 
the covers, PyOpenGL has to retain a copy of the copied array in order 
to prevent it going out of scope and causing crashes.  There was a bug 
in the code that stores the copy, so that it was recording based on the 
array data-type (e.g. float) rather than the array role (e.g. vertex 
array).  So if you used two arrays that both had to be converted and 
both were of the same final type, you'd overwrite the one array with the 
next and wind up with a crash when you tried to access the data in the 
first array.

This patch seems to fix the issue:

Index: OpenGL/GL/pointers.py
===================================================================
RCS file: /cvsroot/pyopengl/OpenGL-ctypes/OpenGL/GL/pointers.py,v
retrieving revision 1.22
diff -C3 -r1.22 pointers.py
*** OpenGL/GL/pointers.py       8 Feb 2007 02:24:40 -0000       1.22
--- OpenGL/GL/pointers.py       9 Oct 2007 02:13:50 -0000
***************
*** 105,120 ****
        function= wrapper.wrapper( baseFunction )
        assert not getattr( function, 'pyConverters', None ), """Reusing 
wrappers?"""
        if arrayType:
!               arrayType = arrays.GL_CONSTANT_TO_ARRAY_TYPE[ glType ]
!               function.setPyConverter( 'pointer', 
arrays.asArrayType(arrayType) )
!               function.setCResolver( 'pointer', 
arrayType.voidDataPointer )
        else:
                function.setPyConverter( 'pointer', 
arrays.AsArrayOfType('pointer','type') )
                function.setCResolver( 'pointer', 
arrays.ArrayDatatype.voidDataPointer )
        function.setCConverter( 'pointer', converters.getPyArgsName( 
'pointer' ) )
        if 'size' in function.argNames:
                function.setPyConverter( 'size' )
!               function.setCConverter( 'size', 
arrays.arraySizeOfFirstType(arrayType,defaultSize) )
        if 'type' in function.argNames:
                function.setPyConverter( 'type' )
                function.setCConverter( 'type', glType )
--- 105,120 ----
        function= wrapper.wrapper( baseFunction )
        assert not getattr( function, 'pyConverters', None ), """Reusing 
wrappers?"""
        if arrayType:
!               arrayModuleType = arrays.GL_CONSTANT_TO_ARRAY_TYPE[ glType ]
!               function.setPyConverter( 'pointer', 
arrays.asArrayType(arrayModuleType) )
!               function.setCResolver( 'pointer', 
arrayModuleType.voidDataPointer )
        else:
                function.setPyConverter( 'pointer', 
arrays.AsArrayOfType('pointer','type') )
                function.setCResolver( 'pointer', 
arrays.ArrayDatatype.voidDataPointer )
        function.setCConverter( 'pointer', converters.getPyArgsName( 
'pointer' ) )
        if 'size' in function.argNames:
                function.setPyConverter( 'size' )
!               function.setCConverter( 'size', 
arrays.arraySizeOfFirstType(arrayModuleType,defaultSize) )
        if 'type' in function.argNames:
                function.setPyConverter( 'type' )
                function.setCConverter( 'type', glType )

> BTW, glReadPixelsub(x, y, 1, 1, GL_RGBA) returns 'int8' instead of 'uint8' and I
> am forced to implement things like:
>
> color = GL.glReadPixelsub(x, y, 1, 1, GL.GL_RGBA)
> #workaround a PyOpenGL bug?
> if color.dtype == 'int8':
>     color = color.astype(numpy.uint8)
I'm using glReadPixelsub in the readpixelleak test in OpenGLContext and 
the result is coming back as a uint8 array there.  Sorry to be a pain, 
but have you got a bit of code where I could see the failure (it's 
pretty easy to fire up pdb and trace through the wrappers to see where 
something goes wrong with the 3.x version)?

HTH, and thanks for the bug report,
Mike

-- 
________________________________________________
  Mike C. Fletcher
  Designer, VR Plumber, Coder
  http://www.vrplumber.com
  http://blog.vrplumber.com

Hello,

Thank you for your continued PyOpenGL development and support. It is really 
great.

At 22:36 08/10/2007 -0400, Mike C. Fletcher wrote:
>I'm using glReadPixelsub in the readpixelleak test in OpenGLContext and 
>the result is coming back as a uint8 array there.  Sorry to be a pain, but 
>have you got a bit of code where I could see the failure


Just adding the code below to the previously submitted file illustrates the 
problem. I send the file as attachment but it is just to avoid problems 
because it is the same file as before. (Just click on the displayed widget)

     def mousePressEvent(self, event):
         self.lastPos = qt.QPoint(event.pos())
         #get the color
         x = self.lastPos.x()
         y = self.lastPos.y()
         y = self.height()- y
         self.makeCurrent()
         color = GL.glReadPixelsub(x, y, 1, 1, GL.GL_RGBA)
         #workaround a PyOpenGL bug
         if color.dtype == 'int8':
             print "WARNING: int8 received"
             color = color.astype(numpy.uint8)
         else:
             print "received %s" % color.dtype
         print "Color = ", color

(Well, actually is not needed to use that PyQt specific approach,  just 
adding the lines below to the end of the paintGL method is enough):

         color = GL.glReadPixelsub(0, 0, 1, 1, GL.GL_RGBA)
         if color.dtype == 'int8':
             print "WARNING: int8 received"


>(it's pretty easy to fire up pdb and trace through the wrappers to see 
>where something goes wrong with the 3.x version)

After all your effort to convert PyOpenGL to ctypes I guess I will have to 
learn to use pdb  ... :)

Best regards,

Armando

V. Armando Sole wrote:
...
> Just adding the code below to the previously submitted file 
> illustrates the problem. I send the file as attachment but it is just 
> to avoid problems because it is the same file as before. (Just click 
> on the displayed widget)
With the latest CVS version, on clicking I get:

mcfletch@...$ python 
glreadpixelsubformat.py 1
received uint8
Color =  [[[128 128 128 255]]]
...
>         if color.dtype == 'int8':
>             print "WARNING: int8 received"
>             color = color.astype(numpy.uint8)
>         else:
>             print "received %s" % color.dtype
>         print "Color = ", color
which suggests that the problem has been fixed between the last release 
and now.  I'm planning to do a beta-1 release this week, incidentally.

Good luck,
Mike

-- 
________________________________________________
  Mike C. Fletcher
  Designer, VR Plumber, Coder
  http://www.vrplumber.com
  http://blog.vrplumber.com