From: Kolja Kaehler <kolja.kaehler@av...>  20100311 11:19:51

Hmm, my post bounced. Let me try again... sorry, if this causes a duplicate! Dan, Ian, Greg, Thanks for your answers. I still have to look into the best way to do my resizable vertex array, but other than that I am probably going to follow Greg's advice and go with the transposed matrices, as well as using homogeneous coordinates throughout... I guess I wasn't too clear on the columnmajor vs. rowmajor topic. To elaborate, if I do (in C++) glLoadIdentity(); glTranslatef(1.0f, 2.0f, 3.0f); float elems[16]; glGetFloatv(GL_MODELVIEW_MATRIX, elems); The returned array reads [1.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 1.0 0.0 1.0 2.0 3.0 1.0]. Following the Red Book, this represents a matrix in columnmajor layout, i.e. 1.0 0.0 0.0 1.0 0.0 1.0 0.0 2.0 0.0 0.0 1.0 3.0 0.0 0.0 0.0 1.0 Thus, vectors are transformed by multiplying from the right: v' = A*v. This would also be how transformations in a GLSL shader are written down. In PyOpenGL, doing the same thing, the resulting matrix is transposed: elems = glGetFloatv(GL_MODELVIEW_MATRIX) results in: [[ 1. 0. 0. 0.] [ 0. 1. 0. 0.] [ 0. 0. 1. 0.] [ 1. 2. 3. 1.]] So, I have to use a row vector and multiply from the left. Likewise for matrixmatrix multiplications. As a PyOpenGL newbie, this is quite surprising to me. Seems like the python code is cleanest when I accept this behavior  just have to be extra careful when exchanging code with other projects using the other convention. Actually, having mentioned shaders  when I pass a matrix to a shader, am I guessing right it will be the transpose of the python matrix? Kolja Original Message From: Dan Helfman [mailto:Dan.Helfman@...] Sent: Tuesday, March 09, 2010 7:14 PM To: Kolja Kaehler Cc: pyopenglusers@... Subject: Re: [PyOpenGLUsers] best way to set up matrices and vertex arrays? Kolja Kaehler wrote: > Does anyone here have some advice for me on these matters? Maybe I am > just missing "the obvious way" to organize these things. Hi Kolja, I haven't had any problems using NumPy's rowmajor arrays with PyOpenGL. For instance if you have a rowmajor NumPy array of dtype [ ( "x", np.float32 ), ( "y", np.float32 ) ], you can pass the array directly to PyOpenGL by putting it in a vertex buffer object and then using glVertexPointer( 2, gl.GL_FLOAT, 0, None ) to tell GL to expect 2 floats for each vertex. The same principle works for 3D as well. And if you'd like to store several pieces of NumPy array data together, then you can just use a recarray, and then pass an individual "member" array to PyOpenGL at any point you want. Perhaps you can give an example of what you're trying to do where the rowmajor/columnmajor distinction gets in the way? There really isn't a good way to dynamically grow a vertex array with NumPy. Some options are: * Use numpy.resize(), which just returns a new, larger array copy. * Use numpy.ndarray.resize(), which can only resize if there are no other references to the array. * Accumulate and grow with a data structure other than a NumPy array, and then convert to a NumPy array when you're done growing. Some candidates are a Python list (slow), or if you're willing to use Cython or C, then accumulate in a C array with realloc() for growing. I haven't really used NumPy matrices, so I'll let someone else comment on that one. Dan 