Interest in upstreaming re_renderer changes from a downstream fork (browser WGSL fixes, atomics-wasm, WebGL2 picking, optional crates)?

[virtualritz]

Hello — asking before writing PRs, so I don't send you work you don't want.

We build a node-based 3D modeller (Akatela) on re_renderer, and have carried a fork for a while (~23 commits on top of main, plus merges). Some of it is clearly ours and belongs nowhere near upstream; some looks like it fixes things you'd hit too. I'd rather ask which, if any, you want than open a pile of PRs.

Probably useful to you -- portability and build-shape fixes, no behaviour change:

1. Browser WGSL strictness (3 small shader fixes): add return after discard in the mesh picking layer; replace a trailing if/else-return pattern; use var + trailing return in camera_ray_to_world_pos. Some browsers' WGSL front-ends reject the current forms.
2. Multi-threaded (atomics) wasm support -- plus making re_web::fs optional so atomics-wasm builds at all, and silencing dead-code warnings the atomics config produces in the error tracker.
3. WebGL2-compatible mesh selection via the Limited device tier, for machines where WebGPU is unavailable or falls back to a software adapter.
4. re_video/re_byte_size/re_tuid behind feature flags, for consumers that only want the renderer.
5. Relaxing the wasm-bindgen =0.2.108 pin to 0.2.
6. Slope-scaled depth bias for overlay lines and points.

Ours, and I'd expect you to say no -- listing for completeness: matcap shading (including an use_matcap == 0 albedo-at-corner-UV path), external depth-texture sharing with premultiplied compositing and depth-only/picking-only mesh modes, per-face element_id picking with a selection SSBO and hover/selection tinting, a wireframe selection cue, and a glam 0.33 bump (you're on 0.30, so that one is a fork artefact, not a proposal).

Questions

• Is any of group 1–6 worth a PR? I'd send them separately and small, each rebased on current main.
• For the WGSL fixes specifically: would you want them as-is, or do you have a preferred idiom for shaders that must pass every browser front-end?
• If the answer is "none of it", that's a genuinely fine answer and saves us both time -- I'll keep them downstream and stop wondering.

Happy to provide the diffs or a branch for anything you want to look at first.

[grtlr]

Thank you for checking back with us before opening PRs, that's a nice touch!

I'd be interested in taking a look at the diffs for:

2. Multi-threaded (atomics) wasm support — plus making re_web::fs optional so atomics-wasm builds at all, and silencing dead-code warnings the atomics config produces in the error tracker.

and

5. Relaxing the wasm-bindgen =0.2.108 pin to 0.2.

Note the WASM target is a bit brittle over all and we need to make sure not to break the viewer in unexpected places, but maybe there's something we can do.

I'm sure @Wumpf will be interested in some of the other fixes too.

[Wumpf]

1. Browser WGSL strictness (3 small shader fixes): add return after discard in the mesh picking layer; replace a trailing if/else-return pattern; use var + trailing return in camera_ray_to_world_pos. Some browsers' WGSL front-ends reject the current forms.

Sounds like straight up bugs that need fixing. Which browsers reject those forms? Today we default enable WebGPU only on Chrome but do test Firefox and Safari every now and then

3. WebGL2-compatible mesh selection via the Limited device tier, for machines where WebGPU is unavailable or falls back to a software adapter.

WebGL2 has fully working mesh selection, no? What is not working for you?

4. re_video/re_byte_size/re_tuid behind feature flags, for consumers that only want the

Byte size may get annoying because it goes into a lot of stuff. Video makes sense, could just have all the video things be feature flagged. Fairly sure re_tuid is also only in there because of video

5. Slope-scaled depth bias for overlay lines and points.

this is for 2d layered things in 3d? I think it would make more sense to have these things rendered similar to transparent objects. We need a better way to force "2d mode" on some of this stuff and then render without z-buffering and software ordering only