AT-1 WASM decoder

The AT-1 reference decoder compiled to WebAssembly (~196 KB .wasm + ~15 KB JS glue, with xz+zstd linked in). Decodes .at1 files byte-for-byte in the browser, Node, Deno, or at the edge — the same fuzz-hardened C decoder used natively, so malformed input is rejected cleanly (a thrown error), never a crash. Verified against all committed vectors (every backend + codec) plus a malformed-input check: 11/11.

Use

import { decode, version } from '@at1/decoder';

console.log(await version());                 // "AT-1 reference decoder 0.1.0"
const original = await decode(at1Bytes);      // Uint8Array -> Uint8Array (the exact source bytes)

In the browser: open index.html (served over http) and drop a .at1 file — it is reconstructed locally; nothing is uploaded.

Scope

The full build supports all backends (xz, zstd, stored) and all codecs (including qcolumnar) — full parity with the native decoder — by linking a cross-compiled liblzma. The committed at1_decode.wasm is this full build (~196 KB).

If liblzma for wasm isn't present, build_wasm.sh falls back to a -DAT1_NO_XZ build (zstd + stored backends, all codecs except qcolumnar); for that mode, produce wasm-decodable files with at1 compress <domain> --backend zstd ....

Build

Requires Emscripten (emcc on PATH, or EMCC=...) and the zstd source tree at tools/zstd (or ZSTD_SRC).

For full xz support, first cross-compile liblzma to wasm (needs the xz source at tools/xz and the em-tools on PATH):

bash build_liblzma_wasm.sh   # -> tools/xz/src/liblzma/liblzma.a
bash build_wasm.sh           # -> at1_decode.mjs + at1_decode.wasm  (FULL build)

build_wasm.sh auto-detects the liblzma archive; without it you get the zstd/stored fallback build.

Test

python make_zstd_vectors.py  # generate zstd-backend vectors into ./vectors
node demo_node.mjs           # decode each in a JS runtime, assert byte-identity

The committed build is verified against the columnar, log, ssh, and json codecs (byte-identical) plus a malformed-input rejection check (5/5).