[Bf-committers] Proposal for unifying nodes

[Matt Ebb]

On 15/06/2009, at 7:38 PM, Aurel W. wrote:

> 2009/6/15 Matt Ebb <matt at mke3.net>:
>> No they're not. Textures need to be completely independent:
>> coordinate in -> values out.
> The problem is, with every node operating on more, than just the
> current coordinate, as a filter node does, you would need to evaluate
> the incoming texture on every coordinate, which is needed for the
> calculation.

Well, it's quite simple, you can't have things like filter nodes as  
part of a shading pipeline after the texture has been evaluated. The  
idea of 'the surrounding pixels' doesn't really even make that much  
sense, especially when you're talking about 3d procedural textures.

What could be possible, is make an image texture reference a node tree  
that generates an image buffer independently as a render pre-process,  
then that image is baked at the start of the render process,  
generating mipmaps etc. and used like a normal image texture. However  
everything leading up to that image has to be stand-alone (i.e.  
independent inputs only) - of course you couldn't access any shading  
variables like surface normal for example, it would have to be  
separated in some way, like compositing nodes are now.

Matt