open-standards/wai/wai-web/reach/tools/wasm.mjs

// A small reader for the parts of a WebAssembly binary the reach build needs: the
// function signatures of every import and export. It reads the type, import,
// function and export sections and skips the rest, so the build does not depend on
// any tool's text format.
//
//   import { readModule } from './wasm.mjs';
//   const m = readModule(bytes);
//   m.imports  [{ module, name, kind, params?, results? }]
//   m.exports  [{ name, kind, params?, results? }]   kind: func | table | memory | global | tag

const VALTYPE = { 0x7f: 'i32', 0x7e: 'i64', 0x7d: 'f32', 0x7c: 'f64', 0x7b: 'v128', 0x70: 'funcref', 0x6f: 'externref' };
const KIND = ['func', 'table', 'memory', 'global', 'tag'];

export function readModule(bytes) {
  const b = bytes instanceof Uint8Array ? bytes : new Uint8Array(bytes);
  let p = 0;
  const fail = (msg) => { throw new Error(`wasm: ${msg} at byte ${p}`); };
  const u8 = () => (p < b.length ? b[p++] : fail('unexpected end'));
  const leb = () => {
    let result = 0, shift = 0, byte;
    do {
      byte = u8();
      result += (byte & 0x7f) * 2 ** shift;
      shift += 7;
      if (shift > 35) fail('LEB128 too long');
    } while (byte & 0x80);
    return result;
  };
  const name = () => {
    const n = leb();
    if (p + n > b.length) fail('name runs past the end');
    const s = new TextDecoder('utf-8', { fatal: true }).decode(b.subarray(p, p + n));
    p += n;
    return s;
  };
  const valtype = () => {
    const t = u8();
    return VALTYPE[t] ?? fail(`value type 0x${t.toString(16)} not read here`);
  };
  const limits = () => {
    const flag = u8();
    if (flag > 7) fail(`limits flag 0x${flag.toString(16)}`);
    leb();
    if (flag & 1) leb();
  };

  if (b.length < 8 || b[0] !== 0 || b[1] !== 0x61 || b[2] !== 0x73 || b[3] !== 0x6d) fail('not a WebAssembly binary');
  if (b[4] !== 1 || b[5] !== 0 || b[6] !== 0 || b[7] !== 0) fail('not version 1');
  p = 8;

  const types = [];
  const imports = [];
  const funcTypes = []; // type index of every function, imported ones first
  const exports = [];
  while (p < b.length) {
    const id = u8();
    const size = leb();
    const end = p + size;
    if (end > b.length) fail(`section ${id} runs past the end`);
    if (id === 1) {
      const n = leb();
      for (let i = 0; i < n; i++) {
        const form = u8();
        if (form !== 0x60) fail(`type form 0x${form.toString(16)} not read here`);
        const params = Array.from({ length: leb() }, valtype);
        const results = Array.from({ length: leb() }, valtype);
        types.push({ params, results });
      }
    } else if (id === 2) {
      const n = leb();
      for (let i = 0; i < n; i++) {
        const module = name(), field = name(), kind = u8();
        if (kind === 0) {
          const t = leb();
          funcTypes.push(t);
          imports.push({ module, name: field, kind: 'func', ...types[t] });
        } else if (kind === 1) { u8(); limits(); imports.push({ module, name: field, kind: 'table' }); }
        else if (kind === 2) { limits(); imports.push({ module, name: field, kind: 'memory' }); }
        else if (kind === 3) { valtype(); u8(); imports.push({ module, name: field, kind: 'global' }); }
        else if (kind === 4) { u8(); leb(); imports.push({ module, name: field, kind: 'tag' }); }
        else fail(`import kind ${kind}`);
      }
    } else if (id === 3) {
      const n = leb();
      for (let i = 0; i < n; i++) funcTypes.push(leb());
    } else if (id === 7) {
      const n = leb();
      for (let i = 0; i < n; i++) {
        const field = name(), kind = u8(), index = leb();
        const e = { name: field, kind: KIND[kind] ?? fail(`export kind ${kind}`), index };
        if (kind === 0) Object.assign(e, types[funcTypes[index]] ?? fail(`export ${field}: no type for function ${index}`));
        exports.push(e);
      }
    }
    p = end;
  }
  return { types, imports, exports };
}

// The i64 boundary of a module: every function import and export with an i64
// parameter or result. The reach runtime splits and joins these values at the JS
// boundary (an i64 crosses as a (low, high) pair of i32 values, and a result's high
// half travels through env.setTempRet0). A multi-value result is refused: neither
// legalisation nor the runtime handles one.
export function i64Boundary(m) {
  const sig = (f) => {
    if (f.results.length > 1) throw new Error(`${f.name}: a multi-value result is not supported`);
    return { params: f.params, result: f.results[0] ?? null };
  };
  const has64 = (f) => f.params.includes('i64') || f.results.includes('i64');
  const exports = {}, imports = {};
  for (const e of m.exports) if (e.kind === 'func' && has64(e)) exports[e.name] = sig(e);
  for (const i of m.imports) {
    if (i.kind !== 'func' || !has64(i)) continue;
    if (imports[i.name]) throw new Error(`import ${i.name} appears under two modules`);
    imports[i.name] = { module: i.module, ...sig(i) };
  }
  return { exports, imports };
}