byte-codec

Generic byte-level primitives (ByteWriter, ByteReader, CRC-32, deflate/inflate, base64) and PNG/JPEG image encoding/decoding with zero PDF knowledge — the shared utility package for the documents.js family.

Extracted from pdf-codec, where these utilities lived as a directory-isolated subgraph under src/bytes/ + src/image/ with no PDF imports. Both pdf-codec and documents.js consume them from this neutral home rather than one fetching byte utilities from a backend.

byte-codec has no internal dependencies in the documents.js family — its only external dependency is fflate for raw DEFLATE/zlib compression. Both pdf-codec and documents.js depend on it:

graph TD
    bytecodec("byte-codec")
    pdfcodec("pdf-codec")
    documents("documents.js")

    bytecodec --> pdfcodec
    bytecodec --> documents

    click bytecodec "https://github.com/ExaDev/documents.js/tree/main/packages/byte-codec" "byte-codec"
    click pdfcodec "https://github.com/ExaDev/documents.js/tree/main/packages/pdf-codec" "pdf-codec"
    click documents "https://github.com/ExaDev/documents.js" "documents.js"

    style bytecodec fill:#f9a825,stroke:#333,stroke-width:3px

Getting started

Requires Node.js >=20 and pnpm 11.6.0.

pnpm install
pnpm build          # tsdown -> dist/ (ESM + CJS + .d.ts)
pnpm typecheck      # tsc -p tsconfig.json && tsc -p tsconfig.node.json (dual tsconfig)
pnpm lint           # eslint . --fix --cache --max-warnings 0
pnpm test           # vitest run
pnpm test:watch     # vitest
pnpm test:workers   # vitest run --config vitest.workers.config.ts, inside a real Cloudflare Workers (workerd) isolate

To run a single test file, pass its path to vitest directly, e.g. pnpm exec vitest run src/bytes/crc32.test.ts.

What it provides

| Module | Exports | | ------------------ | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | bytes/writer | ByteWriter (chunked growable byte-output builder), concatBytes | | bytes/reader | ByteReader (sequential big/little-endian byte reader), isAsciiWhitespace | | bytes/base64 | bytesToBase64, base64ToBytes (standard-alphabet base64, RFC 4648 section 4, with padding — the one implementation every package in the family shares for image payloads and data: URIs), BASE64_ENCODE_CHUNK_CHARS | | bytes/crc32 | crc32 (IEEE 802.3 / ZIP / PNG polynomial table-driven CRC-32) | | bytes/flate | deflate, inflate, inflateTolerant (fflate-backed DEFLATE compression/decompression with a safety cap) | | image/png-encode | encodePng (raw RGB/RGBA pixels → PNG bytes; a truecolour source image whose pixels reduce to 256 or fewer distinct colours is encoded both as indexed colour, PNG colour type 3 with a PLTE and, where needed, a tRNS chunk, and as plain truecolour, and whichever comes out smaller is returned — indexed colour wins for the large flat-colour images typical of diagrams and screenshots, but the PLTE/tRNS chunk overhead can make it larger for small images, so the encoder measures rather than assumes, at the cost of a full second encode for every eligible image; throws if either width or height is not a positive integer no greater than the PNG spec's own IHDR limit, or if their product exceeds a practical per-call pixel-count ceiling, since PNG's IHDR chunk has no valid encoding for zero, negative, fractional, or non-finite dimensions and this bounds an otherwise-unbounded encode to a fixed worst case) | | image/png-decode | decodePng (PNG bytes → raw pixels, handling every PNG colour type including indexed/palette via its own PLTE/tRNS lookup), RawImage | | image/png-filter | filterScanlines, unfilterScanlines (the five PNG scanline filters) | | image/jpeg-info | readJpegInfo (JPEG header reader: dimensions, components, progressive flag — no sample decoding) | | text/decode | decodeText, tryDecodeText, detectByteOrderMark, isProbablyText, UndecodableTextError, TEXT_DECODE_CHUNK_CODE_UNITS (bytes → text with the character encoding worked out from the bytes; see Text decoding below) |

Text decoding

decodeText turns bytes into text without being told which character encoding they hold. It exists because the plain-text formats in this family used to assume UTF-8 and refuse everything else, which turns away Excel's own "CSV (Comma delimited)" export (the Windows ANSI code page) and anything Notepad saved as "Unicode" or PowerShell 5.1 wrote with a redirection operator (UTF-16 with a byte order mark).

The encoding is settled in a fixed order, and the result says which step settled it:

| Step | source | confidence | What decides it | | ---- | ---------- | -------------------------------- | ----------------------------------------------------------------------------- | | 1 | declared | certain | The caller's own encoding option, which skips detection entirely | | 2 | bom | certain | A byte order mark: UTF-8, UTF-16LE/BE, UTF-32LE/BE | | 3 | detected | low | The NUL interleave a mark-less UTF-16 file holding Latin-script text produces | | 4 | utf8 | certain for ASCII, else high | The bytes satisfy UTF-8's own validity rules | | 5 | detected | low | windows-1252, behind a check that the bytes read as Western text |

Anything guessed (confidence: "low") also carries a warnings entry naming what was guessed and on what evidence, so a guess can be shown as a guess rather than presented as fact. windows-1252 has a character for all but five of its byte values, so trying it and seeing whether it fails proves nothing: the plausibility check, and reporting the result as a guess, are what stand in for a test that cannot exist.

Bytes that are not text at all are refused with an UndecodableTextError before any guess is made, since without that gate windows-1252 would read a PNG as a page of mojibake instead of refusing it. isProbablyText is that judgement on its own, for a caller whose real question is whether bytes are text: a NUL byte, or more than one C0 control byte other than tab, line feed, form feed and carriage return per 64 bytes, says binary.

detectByteOrderMark is step 2 above on its own, for a caller that has to settle the encoding before it can safely read anything past the mark — an XML document's own encoding declaration is the case that occurs: the declaration lives inside the document, in whichever encoding the document turns out to be, so a mark has to be ruled out, or read, before that declaration can even be located. It returns the same encoding and mark length decodeText itself would settle on from a mark, or undefined when the bytes carry none of the five it recognises.

Everything except UTF-8 is decoded by this package rather than through TextDecoder. TextDecoder's legacy single-byte and UTF-16 support comes from the host's ICU build, which a Node binary compiled with small-icu does not carry, and the Encoding Standard has no UTF-32 at all, so delegating would make the same bytes decode differently, or not at all, depending on where the code runs. The pnpm test:workers suite decodes windows-1252, UTF-16 and UTF-32 inside a real workerd isolate for exactly that reason, and one unit test checks the whole windows-1252 table, all 256 bytes, against the platform's own decoder wherever the platform has one.

Three limits are worth knowing rather than discovering:

• UTF-16 holding CJK text has no NUL interleave to see, so without a byte order mark it is not detected. Pass encoding for it.
• A legacy CJK or Cyrillic code page is refused rather than guessed at, since nothing distinguishes those from one another without a statistical model this package deliberately does not carry.
• A truncated UTF-8 sequence is genuinely indistinguishable from windows-1252 text containing the same characters, so it decodes as the latter, with a low confidence and a warning.

Conventions

• Worker-isomorphic (see the family-wide convention): runtime src/ must not import node:*, a bare Node builtin, or use the Buffer global — enforced by a no-restricted-imports/no-restricted-globals ESLint rule and exercised in CI by running the test suite inside an actual workerd isolate (pnpm test:workers). Test files under src/**/*.test.ts are exempt and may use Node APIs for fixtures.
• Only src/index.ts may be named index.* — a custom ESLint rule (local/no-non-barrel-index) rejects any other module using an index basename, since that would be a hidden entry point the exports map in package.json doesn't advertise.
• Releases are fully automated: a push to main runs semantic-release in CI, which determines the version from Conventional Commit messages and publishes to npm via OIDC trusted publishing (no local NPM_TOKEN needed). There is no manual publish step. This does not extend to the npm alias below — see that section.

Install

pnpm add byte-codec
# or
npm install byte-codec

npm alias

This package also published under an alternate name from the pre-monorepo pipeline:

• document-bytes

Republished automatically — the alias's trusted publisher is registered against this repository and workflow (2026-09-10), so every release from the backfill run onward publishes under this name too; the registration evidence is on ExaDev/documents.js#727.

License

MIT