encoding / decoding JS getMesh() format

[stewartoallen]

Is it assumed or required that the mesh is indexed? The example uses ThreeJS indexed geometries. Most of the meshes I deal with come from STLs and are not indexed. It's not hard to index them, but would be nice if it could be avoided.

[stewartoallen]

too soon. reading here seems to answer my question. I'll try to spend more time with the library first. lots of questions about performance relating to marshaling between JS and WASM. for the STL case, it would likely be faster to do the indexing in WASM than JS.

[stewartoallen]

going in and out of ThreeJS BufferAttribute formats presents a significant amount of time spent in JS (for loops + accessors). would you consider adding a helper method that returns the flattened vertex and index arrays from meshes?

[stewartoallen]

works for the WASM build, but mem pointers break the normal build. noted in the PR conversation.

[hrgdavor]

well, direct memory access is more than 110x faster. so that's something. @pca006132 @hrgdavor
...
getMeshDirect() returns flat vertex and index arrays suitable for ThreeJS

I am working on supporting 2 major 3d js engines Babylon and Threejs for jscad (on top of jscad's current support for regl) and there are some nuances and differences between them that also reflect data layout (I think not for vertex data, but for color info).
Looking how resposnive ppl here are, I am sure if needed we could get some options for better supporting Babylon too.

[stewartoallen]

this is just dealing with the vertex data, not colors or normals. and index array if the mesh is indexed. do you have details on how the other engines store vertex data, indexed and un-indexed?

[elalish]

Every renderer I know of uses indexed verts because it requires so much less GPU memory. But more importantly, the verts must be indexed to represent a manifold mesh - this is the problem with STL (an un-indexed file format). Then the topology (exact) must be derived from the geometry (inexact), which is why an STL cannot reliably transmit manifoldness information.

[hrgdavor]

this is just dealing with the vertex data, not colors or normals. and index array if the mesh is indexed. do you have details on how the other engines store vertex data, indexed and un-indexed?

Babylon and Three use same layout for vertices, but they use oposite CCW-CW so in my prototypes i need to invert meshes when switching between them.

color

• babylon only uses 4 channel for color data
• threejs can use RGB and RGBA

I tested this stuff here: jscad/OpenJSCAD.org#944 and last update on that is a bit old now, as I easily get side-tracked. I focused on code execution in worker and handling imports and code editors, which will be added to taht prototype when I catch a bit of free time to code.

[hrgdavor]

I am tracking the discussions, and like very much how far manifold has gone already. Looking fwd when I can find some time to play with wasm. And these extra steps for performance look very interesting.

[hrgdavor]

@elalish I have starting playing with manifold and am able to run it in jscad.app. I maybe asking too soon but I would like to in the future explore using SharedArrayBuffer when getting mesh data from manifold. I have not yet come to concrete need for it, as I am still at early explorative phase, just wondering if someone has found it useful with manifold.

[hrgdavor]

One reason is getting mesh data into SharedArrayBuffer is so I can send it to main thread to render, but also have it ready if I run 3mf export inside the worker. A workaround could be simply using transferable, and returning the ArrayBuffer back from main thread after sending to gpu (if it is possible to give it away and back via transferable).

I am otherwise excited about SharedArrayBuffer and looking for ideas :) how to use them. I have some vague idea that with SharedArrayBuffer I could make a different WASM for producing some procedural shapes and have a performant way to send them to manifold wasm for further operations. Consider it more like brainstorming throw away ideas, I may be missing the target, and likely thinking about optimizations way to soon. I need to tickle my brain with grandiose ideas before going back to making something simple/boring but useful :D :D :D .

[elalish]

Love it! Anyway, I'm no ShareArrayBuffer expert, so let me know what you find and what might make a good addition to the library. I'm also thinking about adding our worker.ts to our npm package so people can easily get Manifold scripts evaluated. Would that be useful to you? It currently hands back GLB and 3MF, but we could talk about a different interface too.

[hrgdavor]

I need to explore a bit more, to get a better feeling of the requirements.

I am ok with the idea of such a sandbox for playing with the library, but I want it to be just that, for initial playing. Instead indefinitely adding features to the script env runner, I would push users to convert to regular JS projects and use the ecosystem.

on this example you can see what it takes to convert jscad V1 code to a normal JS file (imports not pruned, it is compat shim)
https://3d.hrg.hr/jscad/v1-support/#test.csg.js

and there is test branch that supports manifold.
https://3d.hrg.hr/jscad/manifold-test/#test.manifold.js
Manifoldcad env for prototyping is fine as it also handles memory, and that could be all some users need, but any that are more like me will want to get out of the sandbox and shoot them self in the foot while learning :)

I have a worker that handles complex js projects rather well, and I am close to making working with worker as simple as calling async methods let something = await worker.doSomeModeling({...params})

I dream big, but even as slowly as jscadui is progressing, nice things useful things are coming out of it :) ... and personally, I enjoy even more making 3d models for 3d printing using JS (I just prefer JS personally and like to share interactive stuff online).

[hrgdavor]

Also @elalish one huge thing became supported across brosers: https://developer.mozilla.org/en-US/docs/Web/API/HTMLCanvasElement/transferControlToOffscreen

• 2018-09-04 - Chrome
• 2018-09-25 - Opera
• 2020-01-15 - Edge
• 2022-09-20 - Firefox
• 2023-03-27 - Safari

[hrgdavor]

@elalish I am looking to support GLB imort(limited ofc) and export for jscad.