Re: [matplotlib-devel] Experiments and questions / OpenGL backend

[Nicolas R. <Nic...@in...>]

> 
> Nicolas,
> 
> I would like to add to Mike's comments by making it clear why I am interested in having OpenGL for matplotlib.  It is not for speed.  The last time I checked, a good portion of time is actually spent on the axis tickers and automatic limits.  So, if speed is your goal, then that might be the better place to look.
> 
> No, my main interest in OpenGL is to relax matplotlib's 2D limitation.  At the core of matplotlib is a 2D layering engine that was perfectly suited for 2D graphing.  mplot3d then takes advantage of that layering feature to hack-on 3D functionality.  However, as has been noted many, many times, this does not always produce correct results.  I would love for the 2D layering core to have some sort of logic where if the OpenGL backend is in use, and if 3D art objects are detected, then bypass the dimension reduction that is done in the layering engine and send the full 3D data over to OpenGL and let it figure out the correct 2D projection.
> 
> However, I am not an OpenGL programmer and I wouldn't even know where to start.  That is why I am very happy to see someone give this another try.  Also, as a side note, in discussions with Ryan May (who did the previous attempt at using pyglet), the main reason he discontinued development at that time was because the text rendering in pyglet was very poor and not anti-aliased.  Do you have any indication if that has improved or not?
> 
> Ben Root

As I just posted, I will try to complete the GL backend (using standard GL and GLUT). For the font rendering, I think freetype-py can be used to offer decent rendering with antialiasing (see http://code.google.com/p/freetype-py/, all examples rendered without LCD filtering). But I do not know if it can compares with agg text rendering.
If not, I could cache glyphs produced by the agg renderer and just use them.


Nicolas