+1 from me. I have tried to use Mathematica to generate stl files for use in both FEA (Comsol) and optics (Zemax). Without much success.

I suggest that the issue be expanded to ask for support for CAD files in general. I frequently see claims in WRI literature that Mathematica supports the major CAD formats, such as:

"The Wolfram Language supports import and export of 3D geometry from all standard formats - with its symbolic representation of 3D objects allowing immediate faithful interchange."

But it doesn't. The list of supported 3D CAD formats is really a list of obsolete or very specialized forms. Any such support would include STEP and IGES, and perhaps Parasolid. These are "standard" 3D formats in solids modeling.

With respect, Vitaliy, I ran your code on my I5 laptop, just up to exp = Show . . .

It took 20 minutes, and this just to tessellate a simple solid which was only a composition of rectangular solids -- a trivial job. None of my real applications are so limited. For any meaningful STL export, Mathematica needs a fully functional tessellation capability for solids with curved surfaces. A first step is to be able to handle all the solids which can be generated by Mathematica 3D primitives. But what is really needed is to extend that to all solids defined by closed parametrically defined surfaces.

This does not cover export of solids to modern CAD standard formats such as STEP, IGES, and Parasolid. For this there may well be a restriction requiring the solid to be a composition of accepted primitives. But this needs to be undertaken before WRI claims to be capable of interchange with current standards,

Kind regards,

David

3D support appears to have gone in the opposite direction. I have code that ran perfectly well five years ago to output .ply files. Now the same code has odd diagonal artifacts that are not in the data. Instead, I'm using .stl format for export. Code that five and ten years ago ran in reasonable times, now takes: an stl file that has 1.5 million faces took twenty hours to export. One that has six million faces is still running, and I expect it to take 80 to 100 hours. Ten years ago the same program took about eight hours.