• You went past making spaghetti and straight into circuit boards.

    How much of the render time is just the CPU compiling your shader?

    • It’s almost instanteneous. Most of the nodes are math/vector math and read/store attributes.

  • YES! readable tree is overrated! jk

    I try to reuse same input node as many places as I can but this is too much.

    No sim, no closure, big spaghetti. What the heck is this thing doing though?

    • believe it or not, it does reuse many nodes. sim/repeat won’t be of much help here, there’re no repetitions. I’m testing an addon I’ve been creating which converts regular Python logic into geometry nodes. this is like ~500 lines of Python code converted into nodes. It’s making a procedural full-quad wall with crenellations from an edge. unsurprisingly, most of the nodes are to correctly treat edge cases/self intersections, removing stray edges/faces etc.

        • looks nice! i wanted something more code friendly, so mine actually parses math expressions, if statements, lambda functions, loops etc and decides internally what nodes it needs to put together (hence the horrible looking node sausage). so you literally write a python logic, which it then transpiles and compiles into a node tree. in other words, you never actually have to think in nodes.

          • My goal was to think about nodes as functions and operations that could be described with simple Pythonic syntax. Every node is a function with positional arguments as socket inputs and a return value as a socket output. The mathematical operation nodes could be described as simple operators like + - // etc. The links between nodes are described as simple variable assignments and passing variables into functions and so on.

            Basically, the idea it to think about Geometry Nodes setup as a script and reduce the Blender API Geometry Nodes boilerplate, while keeping it possible to address a specific node when needed.

            I also put a bunch of work into the Extension API, which makes it possible to extend the functionality of the DSL and use it like a third-party library in a familiar way, with its own defined methods and logic.

            I’d be happy if you tried it out! 🌽

            • yep, that’s the same idea! i’m happy to share the code once i clean it up a bit, and you should feel free to transfer any of the functionality. there are few features i added which i really like, such as a support for LSP and type hints in an external IDE, plus an ability to write the code outside, while the add-on auto-regenerates the node-group if a file is changed. another cool thing is you can inline functions, i.e.:

              # these two are automatically inlined (i.e., nodes added to the active group)
              @inline
              def foo(x: float, y: bool):
                return x**2 if y else x**3
              
              foo = lambda x, y: x**2 if y else x**3
              
              # while this becomes a reusable node group
              def foo(x: float, y: bool):
                return x**2 if y else x**3
              
              • Interesting, I’m looking forward to seeing your project! I’ll note the features you’ve mentioned, the LSP should be a really handy one : )
                The declaring of inline functions i.e. creating nodes in the root node group is something I feel is lacking in my project right now. Would you make a post about the addon somewhere?

                • yep ill just post it on the same lemmy instance. can also dm you the link