Hi Jon, Thanks,  I appreciate your comments. I've replied below. Best wishes, Ray On 14/05/2026 13:50, Jon Bondy wrote: > Ray: > > What an interesting effort!  I wish you luck.  What are you trying to > achieve, with your new CAD software, that is not available, or > convenient, in other contexts? > I like the parametric design concept in openscad, but not the sort of functional language structure. I do not like the Python syntax, either, since  using spaces or tabs as part of the code, to me, is a bit daft. Floating point numbers give problems in cad, or anywhere that precision is needed, so I started looking at the Manifold library. Using AI, I was able to write a couple of cad test programs in Python, relatively quickly, since there is a lot of Python code for it to find. However, I also discovered, when I was using it last year, that Manifold did not handle booleans in Python. So, I decided to test out AI in writing cpp. It has been a mixture of fun, and hours of frustration, but it is getting better. It has got me thinking fundamentally about various aspects of the 3d world, and modelling thereof. For important calculations, i.e. money, then fp numbers are not used, Integers or fixed dp are used. This is what manifold uses. It is default limited to 6dp, but that can be changed. This means that there is no ambiguity that is visible in openscad fast render/etc. But it brings in a load of other problems, e.g.  .net, ensuring the shape calculations are all manifold. Also, in my real world, 2d shapes do not exist, so that removes a lot of conflicts, which is shown by the way in which colour is handled, for example. Also, I can selectively chamfer edges, since my viewer allows me to select vertices visually with the mouse. So, In effect I can create, with AI exactly the results I want, since I do not write cpp, but eventually I will, I guess. I wrote a load of graphic software about ten years ago, using c#, but that used but now i do not want  .net, so I decided on cpp. As it is, I can create in my cad, virtually anything I want, and if not, then I can get AI to write the cpp function and include that. There are a few downsides, however, but nonetheless, I now use it to design objects in preference to using any other software. > 1.8 GB of code sounds at least 3 orders of magnitude larger than > necessary.  I found this figure to be astonishingly large. It is large, but that is the development file, all the raw cpp, test cad files, etc. The .exe is about 18KB, plus a few dlls. However, the problem with the AI that I have used, is that it wants to patch, instead of using existing functions, even functions it created ten minutes earlier. It all needs a good going-through to refactor, but I think there may be one or two other functions to be developed > > What is the "x/y problem" please? > > :) Once you've seen this, you will see it everywhere! Best not to answer forum posts by replying 'x/y' ... A good explanation  and examples at https://meta.stackexchange.com/questions/66377/what-is-the-xy-problem > > Jon > > > On 5/14/2026 8:18 AM, Raymond West via Discuss wrote: >> HI, >> >> Last November, I started developing my own cad program, written in >> cpp, using AI (co-pilot in visual studio code). It has now got to the >> stage that I can not remember all the function details, and I have >> used copilot to generate an html helper sheet. It also has a short >> list of pseudonyms, since if you don't know the exact word, you will >> not normally find its functionality.  So, for example, if you want to >> 'add' a couple of shapes, and do not know that it is a boolean union >> operation,  then a search on 'add' will bring up the union operation >> description. If more detail is required, then in the program editor I >> can enter '? union full', to get a full description and example of >> 'union'. If, in the editor, I enter '?', then on the console I get a >> full list of functions, or I can select select just the parameters, >> examples or the algorithm. >> >> Furthermore, there is a verbose flag that can be set in the code, >> that gives the details of each calculation within any selected >> function, as the program runs, which is useful for debugging. >> >> I can also ask copilot to generate code in my  cad language. I've not >> tested if it can translate openscad to my cad, or vice versa. One >> problem is, not to fall into the x/y problem situation, which I often >> do, and continuously see on fora/elsewhere. >> >> I think, since I make many typos, that I may need it to create a >> pre-parser, but already it is 1.8 GB of code, which needs >> considerable re-factoring. >> >> I've attached the html helper below, if anybody wants to play with it. >> >> Best wishers, >> >> Ray >> >> >> _______________________________________________ >> OpenSCAD mailing list >> To unsubscribe send an email to discuss-leave@lists.openscad.org >

On 5/14/26 11:21 AM, Raymond West via Discuss wrote: > Integers or fixed dp are used. This is what manifold uses https://github.com/elalish/manifold/blob/73c687839a8d3b2c53f48309eaf28f31289ebf97/src/impl.h#L56 https://github.com/elalish/manifold/blob/73c687839a8d3b2c53f48309eaf28f31289ebf97/include/manifold/common.h#L40 https://github.com/elalish/manifold/blob/73c687839a8d3b2c53f48309eaf28f31289ebf97/include/manifold/manifold.h#L110-L111 What leads you to believe Manifold is using fixed point?

On 5/14/2026 11:21 AM, Raymond West via Discuss wrote: > For important calculations, i.e. money, then fp numbers are not used, > Integers or fixed dp are used. Decimal is no better.  People think it's better because its failings are more natural to them. I just filled out a beneficiary form for my three children.  It wanted to know what percentage to apply to each of them. One third cannot be represented precisely in decimal; the only option was to list them as 33.3%, 33.3%, and 33.4%. Similarly, integers "feel" better because their failings are more "natural".  Indeed, it's poor when you add 0.1 ten times and you end up with 0.9999, but it's also pretty bad when you add 1/10 ten times and end up with zero.

Hi Cory, sorry for the sloppy statement wrt manifold and fixed dp, The more precise answer is, (from the manifold header)  manifold calculates in either single or double precision internally. I use the 64 bit library. from an ai enquiry on my code. * Manifold uses*double-precision floats*internally (64-bit) * Tolerance is*mesh-specific and user-adjustable*via|SetTolerance()| * The 6 decimal places you're seeing is hardcoded in YOUR export functions for OBJ/PLY files * There is*no fixed 6dp setting in Manifold itself* I think what I remember was something from the python versions from earlier, where there was probably a default setting somewhere for 6dp, maybe in trimesh or numpy. No idea, been asleep since then. Best wishes, Ray On 14/05/2026 20:42, Cory Cross via Discuss wrote: > On 5/14/26 11:21 AM, Raymond West via Discuss wrote: >> Integers or fixed dp are used. This is what manifold uses > > https://github.com/elalish/manifold/blob/73c687839a8d3b2c53f48309eaf28f31289ebf97/src/impl.h#L56 > > > https://github.com/elalish/manifold/blob/73c687839a8d3b2c53f48309eaf28f31289ebf97/include/manifold/common.h#L40 > > > https://github.com/elalish/manifold/blob/73c687839a8d3b2c53f48309eaf28f31289ebf97/include/manifold/manifold.h#L110-L111 > > > What leads you to believe Manifold is using fixed point? > _______________________________________________ > OpenSCAD mailing list > To unsubscribe send an email to discuss-leave@lists.openscad.org