Hacker News

Could you please elaborate on how this is different than the other python based modeling tools - build123d[0] and CadQuery[1].

I recently also got annoyed with OpenSCAD and its limitations and therefore started experimenting with Build123d. I'm very much a beginner in the CAD space and would love to understand what inspired you to build sdf.

My basic understanding is that STL files are essentially like Bitmap images and store a list of triangles and their positions, whereas STEP files are more like Vector art where there is a list of instructions on how to build the model. Most CAD GUI programs also operate on a similar model to vector art where they record a list of operations one on top of another. It's why STEP files are a standardized format and can be imported / exported from most GUI based CAD builders. I think.

Given that SDF also seems like it builds only STL files (I could be wrong), wouldn't learning build123d or CadQuery work better if one cares about compatibility with existing GUI based CAD modeling software?

Additionally, atleast build123d offers a similar conceptual model to using Fusion360 and FreeCad - I have limited experience here - but essentially you sketch something in 2D on a particular plane, and then apply some operations to convert it to 3d in a particular manner - the simplest being extruding. This means the mental modeling of how to construct something is very similar across both GUI based CAD programs and Build123d, and that makes it easier for me to jump between GUI based and code based CAD modelling.

I'd love to understand your point of view, and learn more.

[0] - https://github.com/gumyr/build123d

[1] - https://github.com/CadQuery/cadquery

> You can even load an existing 3D mesh and operate on it as an SDF. Great for hollowing, chopping, eroding/dilating, etc. existing models.

This has my instant interest. Multiple times I have wanted to take an existing .STL file and cut a hole on it or add another object to it and have never had success.

I've tried things like Meshlab, but while the interface has what appears to be a hundred different functions, attempting to use anything returns some error code that requires a PhD to understand and none of the "repair" functions seem to help.

I mean seriously: Mesh inputs must induce a piecewise constant winding number field.

How the hell am I supposed to accomplish that on a STL file?

Bless you for your service, sir! I have used `sdf` to create a whole bunch of stuff (buttons for my mother, tealight holders, etc.) and `gg` gets used in a bunch of places (including a couple of bots).

> I should probably do more with this project.

I, for one, would be glad.

I've used your library and really like it!

See also my site fncad https://fncad.github.io ! It's basically intended as "SDFs in your browser with realtime preview with openscad-like syntax". I mostly use it for 3d models for printing.

What do you use for sdf meshing? I never really got the perf where I wanted it.

SDFs are very neat up until the point where you need to build parts that have very precise specifications.

Something like two precisely interlocking gears with a tooth geometry with a profile that's the developed curve of the opposite tooth is a nightmare to build with SDFs

Or precise fillets.

Or hard intersections and differences.

Very useful for doing soft, squishy shapes, less so for hard CAD.

Also, a suggestion: in your project, please consider using Wavefront OBJ as an output format, it is a much, much better choice than STL (STL can't represent the actual topology of the object, it has to be reconstructed).

Commenting off of you since I wrote all of this and then realized it's basically exactly what you're saying. But to +1 everything you just said in my own words:

I love OpenSCAD. I've been 3D printing for a while, but I never really got to a place where I could design interesting parts until I started to get the precision of doing models in code. Sometimes it is slower, for sure.

Every time I've used as a CAD GUI program I would get to this point where I would need to alter a single dimension by 0.25mm and realize that _all_ of my fastener holes, cutouts, etc have to be nudged with the keyboard or mouse to accommodate it. The input devices lack precision for that kind of task, and having to repeat the operation dozens of times (or bulk select) gave me a terrible sinking feeling, and I'd often just step away and give up on the design at that point out of frustration. I try to approach everything in OpensSCAD in a way that means I never have to experience that feeling again.

I will also say that doing everything from scratch in OpenSCAD would be it's own special kind of hell. Libraries like [BOSL2](https://github.com/BelfrySCAD/BOSL2) provide a good set of core ideas and preferences that help set you on a good path. A good example: BOSL2 shapes tend to have a "center origin" by default, which is different than the OpenSCAD default, but makes doing transforms later way easier.

Anyway, happy to see OpenSCAD getting some attention here :)

> I would really recommend using the git master than the latest release though.

This. The master version is so much further ahead of the last tagged version. The render time can be orders of magnitude faster for more complex models.

Agree but you quickly run into its limitations. Like if you 3d print something, you need to eliminate when possible sharp edges. That's not fun to do with OpenSCAD.

BOSL2 is SO nice and powerful, can also highly recommend it! :)

OpenSCAD isn't a CAD tool, IMO. It does nothing at all to aid your design. It barely even helps describe it in any abstract way.

It is a useful tool for programmatically describing either very simple or heavily geometric objects. For everything else it's the wrong tool.

Not just that but it also positions everything in absolute coordinates and does not have the ability to reason about solids, just surfaces. Basically if you want to model something like a bolt you need to create a cylinder for the shaft, a separate head of the bolt, and then a thread profile you can rotate around the cylinder. You must ensure there is enough overlap between these three separate parts so the resultant object is a single surface and not three separate ones.

You can use modules to create a semblance of relative measurements but you still cannot do things like “attach this surface of object A to that surface of object B)”. In practice this means that if you want to create something like a spacer or a bracket you can do that easily enough. But if you want to make a part that matches some real world design you are stuck doing a lot of caliper measurements and math to try to create a part that lines up correctly. The you 3D print it and find that you positioned some hole based on its edge and not center and so nothing quite fits.

OpenSCAD is easy to start but difficult to scale because of these limitations and because once you hard-code any measurement you are stuck with it. The “proper” way to do this is to give everything a variable but honestly that makes reasoning about how to line things up even more difficult. “Does base_width include the width of the vertical walls? What about the margin to make the parts fit together?”

I have never been able to understand how things like FreeCAD lay out their UI. TinkerCAD is relatively simple but clearly a lot less powerful. I did try cadquery which solved a lot of OpenSCAD’s issues by having all offsets be relative by default but also introduces a few issues of its own.

One tip I will give about OS: grab a copy of the latest beta/dev release. The renderer is several orders of magnitude faster.

> The great thing about OpenSCAD is that it makes it easy to 3D model things which may be described using spheres, cylinders, and cubes which are stretch, and/or rotated, and arranged in 3D space.

It also has hulls and minkowski sums, which are powerful once you understand them.

There's also OpenJSCAD. Which, being JavaScript ofc, can run in the browser

https://openjscad.xyz/

https://github.com/jscad/OpenJSCAD.org

You can also make more complex shapes if you're willing to define points, even on a 2D plane that you can then extrude. Downside is that it's basically a MENSA test to define all of the points in the correct order. I've done it a few times, but it's never fun.

I think people should also give Bowler Studio a shot. It has a pretty decent Java-based CAD thing, and it even has built in Clojure support, with physics simulation support and everything. I really like it.

Also you can use the new backend by writing C++: https://github.com/elalish/manifold

Which avoids using the OpenSCAD language, but also means you can't use BOSL2. Might as well use FreeCAD.

Another interesting option is FreeCAD, which is scriptable in Python but its primary interface is a GUI. So you can use a script to create things programmatically, edit graphically, or both.

I've had really good luck with https://github.com/SolidCode/SolidPython

I ended up adding the https://github.com/BelfrySCAD/BOSL2 library to OpenSCAD and it had some reasonable options for some gear and rack-and-pinion modeling that I needed to do.

(3D printing a sacrificial gear for a seat position adjustment mechanism)

Or perhaps take a look at cadquery? It’s also pretty neat - http://cadquery.readthedocs.io/

> https://pythonscad.org/

Thanks for this, didn't know about it.

How does that compare to solid python 2?

https://github.com/jeff-dh/SolidPython

> The awful thing about OpenSCAD is that what one can model in 3D is limited by one's ability to mathematically stretch, rotate, and/or arrange spheres, cylinders, and cubes in 3D.

Not at all. You can build any polyhedron you want using the polyhedron command.

Which would be an enormous pain to write out by hand every time, but I wrote a function once upon a time to generate a polyhedron from a stack of layers, each of which is a list of points, and haven't had to mess with old cubes and spheres since.

One annoying thing is that the default way of writing programs in OpenSCAD uses 'modules', which are a bit limited compared to functions (you can't store them as values to to functions or other modules). I worked around that by writing a module that interprets arrays (think S-expressions) that representing the shape, and then just build up that S-expression-like thing with functions and whatever.

Once you've built your own programming language inside OpenSCAD it's perfectly usable. :-)

This might sound like sarcasm but I actually do prefer this to dealing with Python's mutable state jungle / package management nightmare.

I was once a big OpenSCAD user myself but I'm really skeptical that there are many use cases where it's actually more intuitive than a traditional CAD program, even if you're a programmer. It's true CAD programs have a huge amount of features but the basic sketch, extrude, revolve, and loft tools aren't conceptually difficult and are basically the same between Onshape, Fusion, Solidworks, etc. Those tools are sufficient to make 99.99% of OpenSCAD models I'm seeing.

I also have the opposite experience about understanding previous scripts. Unless it's dead simple I'm usually thinking why the hell did I multiply this thingy by sqrt(3)/2 plus this other thing. Maybe a documentation problem, but it's inescapable that sometimes you need a lot of math for what are trivial constraints in an interactive sketch. A real CAD program will let you roll back to any feature to figure out how it's constructed step by step so there's really nothing to decipher.

I find that I'm often making one-offs. I take a part I designed and I need a slight modification for it for some reason.

Fusion is great for that as long as there's not too many parts. But sometimes I'll want a new variant or a series of new variants.

And reaching for python makes that easy.

OpenSCAD is the only CAD tool I use. I can'd figure out how to operate a graphical 3D software with a 2D GUI so it's just easier to describe things mathematically.

I just wish it had operations for subtractively chamfering, rounding, etc. as doing minkowski() with cones and spheres to achieve that result can be unwieldy.

I have played with this but been underwhelmed. However I do think probably on the right track.

I know the ecosystem not-at-all (sum total knowledge of the CAD ecosystem is that my kids got a Bambu printer for Hanukkah) but it feels to me that current LLMs should be able to generate specs for something like https://partcad.readthedocs.io/en/latest/, which can then be sliced etc.

Curious to know what others think? I come at this from the position of zero interest in developing the fine design skills needed to master but wanting to be able to build and tweak basic functional designs.

I want to like Zoo but the rendering engine is so buggy currently that it’s not really usable for more than simple shapes. The text-to-CAD feature they highlight is slow and error-prone, so much so that they explicitly use a “prebuilt” version in the tutorial, and each time I tried it, it gave tool errors or took so long I just did it manually.

Ooh, an LLM that could take requests for custom Gridfinity bins would be a good step towards AGI indeed.

I was just about to say the same: OpenSCAD is CAD for programmers. It's very different from what's generally considered "CAD".

That's not necessarily a bad thing; there's a clientele for it, especially here on HN. But as a mechanical engineer who's used quite a few industry standard CAD systems, I'm sorry to say I wouldn't touch it with a ten foot pole.

Funnily, just a few days ago I tried FreeCAD, and found it to be better than I had expected.

Fusion 360 is free for hobbyists and it's not too hard to learn for basic stuff

+1 for CADQuery - it lets you easily specify surfaces as the basis for manipulation, and results in much less fragile designs than OpenSCAD.

The notable limitations for OpenSCAD are:

- functional programming model --- some folks find not having traditionally mutable variables limiting

- output is as an STL, or DXF using polylines

- native objects are spheres, cylinders, cubes, with functions for hull and Minkowski, so filleting and other traditional CAD operations can be difficult

> The power of parametric cad is such that I wouldn't be a 10th as productive using an interactive cad system.

This sounds like parametric and interactive CAD are polar opposites. Normal CAD software is generally parametric too.

The other points about the advantage of text files still applies though.

It does have caching, but you are otherwise correct. I would also add that it can't do arcs; any curve is discretized immediately upon creation.

Now they say newest version of OpenSCAD has this functionality builtin. It took only a year.

It's switched to a new backend https://github.com/elalish/manifold

Which you can also use directly instead by writing C++. Trivial operations are two orders of magnitude faster and complex ones every faster.

The openscad-nightly is lightning-fast, but makes occasional occlusion errors.

Assuming you make all the necessary adjustements in preferences.

There is at least one blind designer who has been using this approach to surprisingly good effect:

https://makerworld.com/en/models/2040939-accessible-christma...

I came here to ask how good LLMs are at working with this. I wonder if a person could take it a step further with some MCP tools that the agent could use to verify and work with the design.

Yes. The "official release" is just so old as to be useless at this point. They should either update it or take it down and point people at github or something, IMO.

I use the latest version all the time. The newer renderer ("manifold", IIRC) is much faster, and there are newer facilities that make it possible to build 3MF files containing multiple objects for multi-color printing, though that takes a bit of thought to do correctly.

Apparently the nightlies are the one to use. At least, they build it for Apple Silicon in those.

The last release was 2021.01 but the GitHub repo seems to be recently updated. So I'd say... Maybe?

That said, there are often times software gets so stable that not having a new release for years is fine. Maybe this is one of them?

(I'm very new to OpenSCAD so I haven't run into bugs yet... But maybe it's pretty solid?)

Yes, but the main downloads on the site are very old for some reason. Just get the nightly version instead, and then in Preferences -> Advanced -> Backend change it to “Manifold”. It will make your models “render” 10x faster (or more!).

Yes. https://github.com/openscad/openscad

If you're inclined to use Python there is:

https://pythonscad.org/

It actually renders things incredibly fast if you get the nightly version and set the backend to Manifold. It is probably 100x faster (!!). In fact it renders so fast that I put a render() command at the top of my hierarchy so that everything just renders all the time, it’s faster and more performant. I make incredibly complex models with it too, with hundreds of holes, complex svg files with text in them, etc.

I'm a programmer. I once had an idea stuck in my head for a 3d model that I just needed to get down somewhere. I tried learning the basics of AutoCAD but after 2 days of tutorials I still felt overwhelmed.

I looked into alternatives and learned about OpenSCAD. The immediate visual feedback makes picking up the language a breeze. Within an hour of downloading I familiarized myself with the language and had manifested my idea into a 3d model

I think that's a perfect example of a use-case where OpenSCAD shines. It's extremely easy to pick up if you have programming experience and it might even be a good thing to learn before moving onto more professional CAD software. From a teaching perspective, being able to have almost immediately-useful output is priceless

The thing is, I've crashed-and-burned every time I've tried to do traditional 3D CAD --- the closest I've come to success was making it all the way through the tutorial for Dune 3D:

https://news.ycombinator.com/item?id=37979758

https://news.ycombinator.com/item?id=40228068

OpenSCAD "just works", even on quite limited hardware, and if one has trouble modeling something, well, arriving at a solution is just a matter of learning the appropriate mathematics.

excplicitely, you mean?

I wonder what kind of things OpenSCAD can't ever offer... It's pretty automatable.