erasure.ts

1/**2 * erasure.ts — Reed-Solomon erasure coding around the WASM library.3 *4 * The library wants a single contiguous buffer of length5 *   (dataShards + parityShards) * shardSize6 * with data shards laid out first, followed by space for parity. `encode()`7 * fills the parity region in-place. `reconstruct()` rewrites missing shards8 * in-place using a `shardsAvailable: boolean[]` mask.9 *10 * We expose a chunk-oriented API (`encodeWithErasure` / `decodeWithErasure`)11 * because the rest of the codex pipeline thinks in terms of individual12 * envelopes. Both adapters allocate a fresh buffer per call — the library is13 * not re-entrant on a shared buffer.14 */1516import { ReedSolomonErasure } from "@digitaldefiance/reed-solomon-erasure.wasm";17import {18  CHUNK_SIZE,19  CodexError,20  computeParityShards,21} from "../../shared/codex-types";22import { getReedSolomon } from "./rs-loader";2324/**25 * The shared rs-loader caches a single `ReedSolomonErasure` instance, which26 * is fine for `encode()` but trips a known library bug on the SECOND call to27 * `reconstruct()` against the same instance: the internal __wbindgen state28 * survives between calls and the second reconstruction reports29 * RESULT_ERROR_TOO_FEW_SHARDS_PRESENT even when enough shards are present.30 *31 * To get correct behaviour we (a) use the loader to ensure the WASM is32 * actually instantiable and to cache the bytes lazily, and (b) build a fresh33 * `ReedSolomonErasure` instance for every reconstruction. The cost is a few34 * milliseconds per call — negligible compared to compressing the vault and35 * rendering Aztec codes.36 */37let cachedWasmBytes: Uint8Array | null = null;3839async function freshReedSolomon(): Promise<ReedSolomonErasure> {40  if (cachedWasmBytes) {41    return ReedSolomonErasure.fromBytes(cachedWasmBytes);42  }43  // Capture bytes by intercepting the loader's first instantiation. The44  // loader caches its own promise, so calling it twice is cheap on the second45  // pass. We pull bytes from the WASM memory image after the fact — see46  // `loadWasmBytes` below for the lazy initialization.47  await getReedSolomon();48  if (!cachedWasmBytes) {49    cachedWasmBytes = await loadWasmBytes();50  }51  return ReedSolomonErasure.fromBytes(cachedWasmBytes);52}5354/**55 * Load the WASM bytes through the same probing logic rs-loader uses, so this56 * file does not assume any specific filesystem layout. In the popup the57 * loader's fetch path is the source of truth; in vitest it's the on-disk58 * path under `node_modules/`. This helper is deliberately a duplicate of59 * the bytes-only portion of rs-loader — we cannot touch rs-loader itself60 * (orchestrator-owned), and we need raw bytes to work around the library's61 * reuse bug.62 */63async function loadWasmBytes(): Promise<Uint8Array> {64  const inPopup =65    typeof chrome !== "undefined" &&66    typeof chrome.runtime !== "undefined" &&67    typeof chrome.runtime.getURL === "function";68  if (inPopup) {69    const url = chrome.runtime.getURL("codex/reed_solomon_erasure_bg.wasm");70    const response = await fetch(url);71    if (!response.ok) {72      throw new CodexError(73        "ERASURE_CODING_VERIFICATION_FAILED",74        `Failed to fetch WASM: ${response.status}`,75      );76    }77    return new Uint8Array(await response.arrayBuffer());78  }79  const fs = await import("node:fs");80  const path = await import("node:path");81  const wasmPath = path.resolve(82    "node_modules/@digitaldefiance/reed-solomon-erasure.wasm/dist/reed_solomon_erasure_bg.wasm",83  );84  return new Uint8Array(fs.readFileSync(wasmPath));85}8687export interface EncodeResult {88  dataChunks: Uint8Array[];89  parityChunks: Uint8Array[];90  totalChunks: number;91  dataShards: number;92  parityShards: number;93}9495/**96 * Split `data` into fixed-size shards plus `computeParityShards(dataShards)`97 * parity shards. Pads the trailing data shard with zeros — the original byte98 * length is recovered via the `CodexPayload.vaultBlob` (no length prefix99 * needed on the wire because the outer JSON already bounds the data).100 */101export async function encodeWithErasure(data: Uint8Array): Promise<EncodeResult> {102  const rs = await freshReedSolomon();103104  const dataShards = Math.max(1, Math.ceil(data.length / CHUNK_SIZE));105  const parityShards = computeParityShards(dataShards);106  const totalShards = dataShards + parityShards;107  const buffer = new Uint8Array(totalShards * CHUNK_SIZE);108  buffer.set(data, 0); // Trailing bytes after `data.length` stay zero — the pad.109110  const code = rs.encode(buffer, dataShards, parityShards);111  if (code !== ReedSolomonErasure.RESULT_OK) {112    throw new CodexError(113      "ERASURE_CODING_VERIFICATION_FAILED",114      `Reed-Solomon encode failed: ${ReedSolomonErasure.getResultMessage(code)}`,115    );116  }117118  // Slice() copies so each shard owns its memory and callers cannot mutate119  // the working buffer through a returned view.120  const dataChunks: Uint8Array[] = [];121  for (let i = 0; i < dataShards; i++) {122    dataChunks.push(buffer.slice(i * CHUNK_SIZE, (i + 1) * CHUNK_SIZE));123  }124  const parityChunks: Uint8Array[] = [];125  for (let i = 0; i < parityShards; i++) {126    const base = (dataShards + i) * CHUNK_SIZE;127    parityChunks.push(buffer.slice(base, base + CHUNK_SIZE));128  }129130  return {131    dataChunks,132    parityChunks,133    totalChunks: totalShards,134    dataShards,135    parityShards,136  };137}138139/**140 * Reassemble the data-shard region from a partial set of chunks. Missing141 * shards are reconstructed from parity. The returned buffer is exactly142 * `dataShards * shardSize` bytes long, including any trailing pad bytes143 * — callers strip the pad using their out-of-band length information.144 *145 * Shard size is derived from the actual chunks rather than the current146 * `CHUNK_SIZE` constant so that Codexes printed under an older constant147 * (e.g. 1024-byte shards) still restore on a build whose constant has moved.148 * Reed-Solomon requires all shards to share one length; the first chunk149 * establishes it and every other chunk must match.150 */151export async function decodeWithErasure(152  chunks: Map<number, Uint8Array>,153  dataShards: number,154  totalShards: number,155): Promise<Uint8Array> {156  const parityShards = totalShards - dataShards;157  if (parityShards < 0) {158    throw new CodexError(159      "ERASURE_CODING_VERIFICATION_FAILED",160      `Invalid shard layout: dataShards=${dataShards} totalShards=${totalShards}`,161    );162  }163164  if (chunks.size < dataShards) {165    throw new CodexError(166      "INSUFFICIENT_CHUNKS",167      `Need at least ${dataShards} chunks, have ${chunks.size}`,168    );169  }170171  // First scanned chunk fixes the shard size for the whole reconstruction.172  const firstChunk = chunks.values().next().value;173  if (!firstChunk || firstChunk.length === 0) {174    throw new CodexError(175      "ERASURE_CODING_VERIFICATION_FAILED",176      "First chunk is empty; cannot determine shard size",177    );178  }179  const shardSize = firstChunk.length;180181  const rs = await freshReedSolomon();182  const buffer = new Uint8Array(totalShards * shardSize);183  const available: boolean[] = new Array(totalShards).fill(false);184185  for (const [index, chunk] of chunks) {186    if (index < 0 || index >= totalShards) {187      throw new CodexError(188        "ERASURE_CODING_VERIFICATION_FAILED",189        `Chunk index ${index} out of range [0, ${totalShards})`,190      );191    }192    if (chunk.length !== shardSize) {193      throw new CodexError(194        "ERASURE_CODING_VERIFICATION_FAILED",195        `Chunk ${index} has length ${chunk.length}, expected ${shardSize}`,196      );197    }198    buffer.set(chunk, index * shardSize);199    available[index] = true;200  }201202  const code = rs.reconstruct(buffer, dataShards, parityShards, available);203  if (code !== ReedSolomonErasure.RESULT_OK) {204    // The library distinguishes "not enough shards" with its own code; map it205    // back to our INSUFFICIENT_CHUNKS so handlers upstream can surface a206    // clean "scan more codes" prompt instead of a generic library error.207    if (code === ReedSolomonErasure.RESULT_ERROR_TOO_FEW_SHARDS_PRESENT) {208      throw new CodexError(209        "INSUFFICIENT_CHUNKS",210        `Need at least ${dataShards} valid chunks for reconstruction`,211      );212    }213    throw new CodexError(214      "ERASURE_CODING_VERIFICATION_FAILED",215      `Reed-Solomon reconstruct failed: ${ReedSolomonErasure.getResultMessage(code)}`,216    );217  }218219  return buffer.slice(0, dataShards * shardSize);220}