> some of the universal, underlying data flow architecture that the likes of Maya and Houdini had from the start

Could you elaborate on that? I'm not sure I understand what you mean by "data flow" in the context of a CGI package.

Maya has its dependency graph architecture, while Houdini has its many layers of networks. They both make explicit their pipelines of data transformation, using a common building block, and that allows the system and the artist to reason about what gets done and when. You might create some geometry (a grid), then apply some noise to it to create a landscape, then use it as part of a physics simulation as a static object, then also use it as part of the rendering of the final image. You expect to be able to change the parameters of each of those stages and reason about the result.

You'll generally see these pipelines depicted as editable, directed graphs (DAGs) in the application. Blender uses DAGs for a number of things - eg an image compositor, and for shading networks. It also provides modifier layers (like 3DS MAX) for geometry transformation. But it doesn't go down to the core of the application which is why you can get some problems when it comes to certain tasks in animation for instance. But they are working on this.

It's a bit like all these various architectures for the web - Flux, RxJS etc. They're making explicit the transformations and dependencies that happen, by means of a data flow graph (even if it is not visualised). This usually means less surprises and more opportunity for optimization.

They're working on it, though: http://code.blender.org/index.php/2015/02/blender-dependency...

Oh yes, I see now. Actually, that was one of the few things I missed in Blender coming from Softimage. Not too much, though, but I do appreciate the power of a unified editable operation stack/pipeline for bigger projects.