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  PARITY_SHARDS,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}9394/**95 * Split `data` into fixed-size shards and compute `PARITY_SHARDS` parity96 * shards. Pads the trailing data shard with zeros — the original byte length97 * is recovered via the `CodexPayload.vaultBlob` (no length prefix needed on98 * the wire because the outer JSON already bounds the data).99 */100export async function encodeWithErasure(data: Uint8Array): Promise<EncodeResult> {101  const rs = await freshReedSolomon();102103  const dataShards = Math.max(1, Math.ceil(data.length / CHUNK_SIZE));104  const totalShards = dataShards + PARITY_SHARDS;105  const buffer = new Uint8Array(totalShards * CHUNK_SIZE);106  buffer.set(data, 0); // Trailing bytes after `data.length` stay zero — the pad.107108  const code = rs.encode(buffer, dataShards, PARITY_SHARDS);109  if (code !== ReedSolomonErasure.RESULT_OK) {110    throw new CodexError(111      "ERASURE_CODING_VERIFICATION_FAILED",112      `Reed-Solomon encode failed: ${ReedSolomonErasure.getResultMessage(code)}`,113    );114  }115116  // Slice() copies so each shard owns its memory and callers cannot mutate117  // the working buffer through a returned view.118  const dataChunks: Uint8Array[] = [];119  for (let i = 0; i < dataShards; i++) {120    dataChunks.push(buffer.slice(i * CHUNK_SIZE, (i + 1) * CHUNK_SIZE));121  }122  const parityChunks: Uint8Array[] = [];123  for (let i = 0; i < PARITY_SHARDS; i++) {124    const base = (dataShards + i) * CHUNK_SIZE;125    parityChunks.push(buffer.slice(base, base + CHUNK_SIZE));126  }127128  return { dataChunks, parityChunks, totalChunks: totalShards, dataShards };129}130131/**132 * Reassemble the data-shard region from a partial set of chunks. Missing133 * shards are reconstructed from parity. The returned buffer is exactly134 * `dataShards * CHUNK_SIZE` bytes long, including any trailing pad bytes135 * — callers strip the pad using their out-of-band length information.136 */137export async function decodeWithErasure(138  chunks: Map<number, Uint8Array>,139  dataShards: number,140  totalShards: number,141): Promise<Uint8Array> {142  const parityShards = totalShards - dataShards;143  if (parityShards < 0) {144    throw new CodexError(145      "ERASURE_CODING_VERIFICATION_FAILED",146      `Invalid shard layout: dataShards=${dataShards} totalShards=${totalShards}`,147    );148  }149150  if (chunks.size < dataShards) {151    throw new CodexError(152      "INSUFFICIENT_CHUNKS",153      `Need at least ${dataShards} chunks, have ${chunks.size}`,154    );155  }156157  const rs = await freshReedSolomon();158  const buffer = new Uint8Array(totalShards * CHUNK_SIZE);159  const available: boolean[] = new Array(totalShards).fill(false);160161  for (const [index, chunk] of chunks) {162    if (index < 0 || index >= totalShards) {163      throw new CodexError(164        "ERASURE_CODING_VERIFICATION_FAILED",165        `Chunk index ${index} out of range [0, ${totalShards})`,166      );167    }168    if (chunk.length !== CHUNK_SIZE) {169      throw new CodexError(170        "ERASURE_CODING_VERIFICATION_FAILED",171        `Chunk ${index} has length ${chunk.length}, expected ${CHUNK_SIZE}`,172      );173    }174    buffer.set(chunk, index * CHUNK_SIZE);175    available[index] = true;176  }177178  const code = rs.reconstruct(buffer, dataShards, parityShards, available);179  if (code !== ReedSolomonErasure.RESULT_OK) {180    // The library distinguishes "not enough shards" with its own code; map it181    // back to our INSUFFICIENT_CHUNKS so handlers upstream can surface a182    // clean "scan more codes" prompt instead of a generic library error.183    if (code === ReedSolomonErasure.RESULT_ERROR_TOO_FEW_SHARDS_PRESENT) {184      throw new CodexError(185        "INSUFFICIENT_CHUNKS",186        `Need at least ${dataShards} valid chunks for reconstruction`,187      );188    }189    throw new CodexError(190      "ERASURE_CODING_VERIFICATION_FAILED",191      `Reed-Solomon reconstruct failed: ${ReedSolomonErasure.getResultMessage(code)}`,192    );193  }194195  return buffer.slice(0, dataShards * CHUNK_SIZE);196}