Limits and partial reads

A limit on the way out

A megabyte of zeros gzips to about a thousand bytes. That ratio is the whole point of compression, and also the whole point of a zip bomb. So decompress stops at 256 MiB by default, and you can set it lower:

decompress("deflate", bomb, { container: "gzip", limit: 100_000 });
// LimitError: deflate: output passes the limit of 100000 bytes

The check happens before a byte lands, not after the buffer has grown. A size the stream claims up front only sizes the first buffer, and never past the limit or 64 MiB. A liar gets a small buffer and the same error.

identify and peel take the same limit, applied to every attempt.

The tools are stricter

Long output comes in windows. The header says showing 0 to 16384; pass offset for the rest, and the model asks for the next window with offset.

Partial reads

A stream cut in half still holds half the data. Ask for it:

decompress("deflate", cut, { container: "gzip", partial: true });
// { bytes: …everything before the cut…,
//   details: { error: "deflate: gzip member ends before its CRC-32 and size" } }

Without partial, the same bytes ride on the error:

try {
  decompress("deflate", cut, { container: "gzip" });
} catch (error) {
  if (error instanceof DecompressError) error.partial;  // the same bytes
}

Every format writes as it goes, so what comes out before the damage is real output, not a guess. A wrong checksum works the same way: the bytes are all there, the check just failed. LZMA2 decodes a cut chunk as far as it gets. Inflate never writes a block end it didn't actually read.

Errors

All of them extend CompressionError, so one catch takes them all.