pdfFlate

FlateDecode ISO 32000-2 §7.4.4 — wrapper over @awacloud/fw zlib (RFC 1950), with predictors.

Module pdfFlate | Source packages/front/office/pdf/src/syntax/filters/flate.js | Deps pdfErrors, zlib (@awacloud/fw/io/compress/zlib) | Worker-safe yes

PDF FlateDecode is full zlib framing (CMF/FLG header + deflate payload + Adler32 trailer). The raw compression is delegated to @awacloud/fw's zlib module, which produces and consumes exactly that format.

/DecodeParms /Predictor is handled here, in both directions:

Columns, Colors and BitsPerComponent are read alongside Predictor (defaults 1, 1, 8) and accepted in both the PDF spelling (Predictor, Columns, …) and the lowercase spelling (predictor, columns, …).

Resolve

const flate = runtime.resolve('pdfFlate');
// Returns: { decode, encode }

API

params is optional: omit it, or leave Predictor at 1, and the byte stream is passed through the raw zlib codec untouched.

Truncated streams

Some producers write FlateDecode streams whose deflate data ends before the final block — the input runs out mid-stream, though /Length is exact. Viewers display what decodes, and so does decode: when the strict inflate fails only because the input ended before the final block, the payload is decoded again through fw's streaming decoder (UnzlibStream, without the end-of-input signal) and the bytes decoded so far are returned.

• The returned Uint8Array carries a non-enumerable own property truncated: true. It is invisible to Object.keys, spread and deep equality, so callers that ignore it see an ordinary byte array; callers that care test out.truncated === true. A complete stream never carries it.
• With a Predictor, only the whole predicted rows of the prefix are decoded (a trailing partial row is dropped), so a truncated image or xref stream does not fail with pdf/flate/png-row-mismatch. A complete stream keeps the strict row check.
• The streaming decoder holds back the last 4 input bytes as a presumed Adler-32 trailer, so the recovered prefix can stop a few bytes short of everything the input encodes.

Everything else still throws pdf/flate/inflate-failed: a bad zlib header, an invalid block type, length or distance, a stream truncated inside its header, and a truncated stream from which nothing could be decoded. The error's cause is the original strict-inflate error.

const out = flate.decode(streamObj.raw);
if (out.truncated) {
    // a decoded prefix — the stream ended before its final deflate block
}

Examples

Decode a Flate stream

const flate = runtime.resolve('pdfFlate');
const decoded = flate.decode(streamObj.raw);

Decode with a PNG predictor

const decoded = flate.decode(streamObj.raw, {
    Predictor: 12, Columns: 5, Colors: 1, BitsPerComponent: 8
});

Round trip

const enc = new TextEncoder().encode('hello, pdf');
const compressed = flate.encode(enc);
const back = flate.decode(compressed);
new TextDecoder().decode(back);   // 'hello, pdf'

Through the dispatch

const dispatch = runtime.resolve('pdfFilterDispatch');
const out = dispatch.decode(streamObj);  // resolves /Filter and /DecodeParms

Errors

See also

• pdfFilterDispatch — filter orchestrator.
• Filters index
• pdfObjStream — typical consumer (ObjStm streams are almost always Flate-encoded).