(I'm sure people who've actually used CAD software know this:)
@omar It sounds less magic if you realize your actions are creating new nodes in an operator DAG that is essentially memoized. So when you edit a parameter of an early operator node then it has to reevaluate the graph up to that point.
@omar I mean in theory it will. Or it will fail horribly.
@avi yeah it feels like there's an implicit skill of interacting with the CAD software in such a way that you construct a useful parametric history for you to propagate through later
@omar what I’m interested in right now is: can you construct parametric histories that are robust enough that you can use some physical fitness function on the output to optimize the early parametric choices.
@disconcision @omar very much agreed, and I think “how do you better capture design intent in CAD” is a hugely impactful and open area of UX research.
@omar @avi I enjoyed this article and some of the links on issues relating to this: https://mattferraro.dev/posts/cadmium
@omar
I have mechanical design friends who know this but I'm not sure they quite understand how much it really is programming
I'd love to see 3d parametric CAD tool that exposes parallel GUI/text representations, so you can build something up like in F360 but then refactor in text for robustness. E.g. change a manual face selection by a snippet that expresses something like "uppermost face added by operation XYZ"
@s_ol yeah -- there are deep advantages to both direct and text representations -- reminds me of the Sketch-n-Sketch work (although that's more like 2D SVG medium) if you haven't seen that https://ravichugh.github.io/sketch-n-sketch/
@omar@social.omar.website only tangentially related, but nanDeck (software for generating components for board games) has a very similar system where there's a visual editor and a code representation of whatever thing you're making in the visual editor that you can freely modify in both directions (https://www.nandeck.com/examples)