use cgltf or tinygltf as a no-assimp compile mode for meshIO?

[starseeker]

Was thinking a little about meshIO - it's very handy to have when one needs to write out example "bad" inputs for debugging, but needing assimp complicates life a bit. On the other hand, using it does provide support for a lot of formats.

What would the devs say to using something really small and embed-able like cgltf or tinygltf to support building meshIO when assimp isn't available, and using assimp instead when it is? Essentially that would allow for an "always on" aspect of the meshIO api where it is always handy to write out very basic glb meshes (intended primarily to support debugging cases), but would preserve the current features of assimp support when it is available? Looking over the meshIO code the comments indicate the assimp support is "primarily intended as a demonstration" which makes me think maybe one of the minimalist gltf readers and writers could provide enough to be useful?

I'd consider giving it a go and submitting a patch, but wanted to check how such a thing might be received before diving in.

[elalish]

I think that's a good idea. I'd be especially supportive if you felt like adding our EXT_mesh_manifold glTF extension so that we could losslessly roundtrip the data (I already did this in TS) - with that I'd be willing to drop Assimp entirely; the only formats that have any value for manifoldness are glTF and 3MF. Mainly I want to avoid bloating our various binaries, increasing our dependency count, or being on the hook for file format maintenance. To have proper glTF support, we'd need to support textures, but that also means having to keep up with the glTF spec.

I'm open to ideas, but somehow we have to set expectations, both for our level of maintenance and for what features work. With my TS, all I need to do is update our dependency on gltf-transform to pull in all the latest, and no one is sad about an npm dependency. Not so with C++!

[starseeker]

Do I understand correctly that the assimp I/O currently in place does adequately support the EXT_mesh_manifold extension? If so then I'd imagine it'll be able to figure it out - I'll look into it.

[elalish]

No, Assimp has none of that. There's TS code for it here though: https://github.com/elalish/manifold/blob/master/bindings/wasm/examples/manifold-gltf.ts

[starseeker]

OK. I'll see what I can figure out and submit a PR once I have something worth discussing.

[starseeker]

This is the correct/current extension documentation? https://github.com/KhronosGroup/glTF/blob/main/extensions/2.0/Vendor/EXT_mesh_manifold/README.md

[elalish]

Yes, that's right.