Single thread npm

[MatiasManevi]

This PR modifies the build pipeline to generate and package two distinct versions of the MuJoCo WASM bindings: Single-Threaded and Multi-Threaded.

Previously, users had to choose one build type at compile-time. Now, both are shipped in a single @mujoco/mujoco npm package, allowing developers to opt-in to multi-threading by changing their import path.

Fixes #3144

---

Key Changes

1. CMake Configuration

• Introduced the MUJOCO_WASM_THREADS option (defaults to ON).
• Conditioned Emscripten linker flags (-pthread, PTHREAD_POOL_SIZE) on this toggle.

2. Build Pipeline (build_steps.sh)

• Updated the WASM build function to execute two sequential passes:
  • Pass 1: Single-thread build (outputs to package root).
  • Pass 2: Multi-thread build (outputs to package ./mt).

3. Package Exports (package.json)

• Implemented the exports field to support subpath exports.

[okmatija]

While we're here lets set this PTHREAD_POOL_SIZE=0

[MatiasManevi]

I tested it and seems that both variants are now behaving in the "same way" => multi-threading won't require COOP/COEP headers anymore. Is that expected?

[MatiasManevi]

Is there a way I can quickly test this (maybe a function we can call) to ensure it won't break anything?

[MatiasManevi]

Also, if 0 is the value we want, we shouldn't even set it. because its default value is zero already, wdyt?

[MatiasManevi]

Running this test

import { MainModule, MjVFS, MjSpec, MjModel } from "../../dist/mt/mujoco"
import loadMujoco from "../../dist/mt/mujoco.js"

declare function loadMujoco(): Promise<MainModule>;

async function main() {
  const mujoco: MainModule = await loadMujoco();

  const assetsCount = 10;
  let xml = `<mujoco>
      <asset>`;
  const cube1 = `
    v -1 -1  1
    v  1 -1  1
    v -1  1  1
    v  1  1  1
    v -1  1 -1
    v  1  1 -1
    v -1 -1 -1
    v  1 -1 -1`;
  const dummyTex = 'dummy texture data';

  for (let i = 0; i < assetsCount; i++) {
    xml += `<mesh name="cube${i}" file="cube${i}.obj"/>`;
    xml += `<texture name="tex${i}" type="2d" builtin="checker" width="16" height="16"/>`;
  }
  xml += `</asset><worldbody>`;
  for (let i = 0; i < assetsCount; i++) {
    xml += `<geom type="mesh" mesh="cube${i}"/>`;
  }
  xml += `</worldbody></mujoco>`;

  let spec: MjSpec | null = null;
  let model: MjModel | null = null;
  let vfs: MjVFS | null = null;
  try {
    spec = mujoco.parseXMLString(xml);

    vfs = new mujoco.MjVFS();
    for (let i = 0; i < assetsCount; i++) {
      vfs.addBuffer(`cube${i}.obj`, new TextEncoder().encode(cube1));
      vfs.addBuffer(`tex${i}.txt`, new TextEncoder().encode(dummyTex));
    }

    model = mujoco.mj_compile(spec, vfs);
    console.log(model.nmesh, assetsCount);
    console.log(model.ntex, assetsCount);

    for (let i = 0; i < assetsCount; i++) {
      const meshId = mujoco.mj_name2id(
        model, mujoco.mjtObj.mjOBJ_MESH.value, `cube${i}`);
      console.log(meshId, i);
      const tex = model.tex(`tex${i}`);
      console.log(tex!.name, `tex${i}`);
    }
  } finally {
    spec?.delete();
    vfs?.delete();
    model?.delete();
  }
}

main()

With PTHREAD_POOL_SIZE=0 throws an error.

[MatiasManevi]

Adding spec.compiler.usethread = 0; to the script would make it run, but if we are importing mujoco from /mt is that really what we want?