Hawkynt

ZPAQ (Context Mixing)

The context-mixing compressor at the heart of ZPAQ: four direct context models over hashed orders 1 to 4 predict each bit of the message, their predictions are averaged, and a carry-propagating binary range coder turns confident predictions into fractions of a bit. The order hashes are rebuilt after every byte with ZPAQ’s HASH step, h = (h + b + 512) * 773 modulo 2 to the 32. Covers the modelling and coding stages only - the journaling archive container, the general ZPAQL virtual machine with its configurable COMP/HCOMP sections, deduplication and versioning are not implemented.

Properties

Property Value
Category Compression Algorithms
Sub-category Context Mixing
Security status πŸŽ“ Educational Only
Complexity Expert
Inventor Matt Mahoney
Year 2009
Origin πŸ‡ΊπŸ‡Έ United States
Source algorithms/compression/zpaq.js

Security

Status: πŸŽ“ Educational Only

No vulnerabilities are recorded for this implementation.

Documentation

References

Test vectors

7 vectors ship with this algorithm and run in the test suite. Byte values are hexadecimal.

Vector 1 β€” Empty input - length header only

Field Value
input (empty)
expected 00000000

Vector 2 β€” Single zero byte - eight bits coded at one half

Field Value
input 00
expected 0100000000feff8000

Vector 3 β€” Single byte 0x41 - pins the subrange convention

Field Value
input 41
expected 0100000000bdff8000

Vector 4 β€” Natural text round-trip

Source: Regression test for model desync

Field Value
input 74686520717569636b2062726f776e20 666f78206a756d7073206f7665722074 6865206c617a7920646f672e20
expected (empty)

Vector 5 β€” All 256 byte values round-trip

Source: Regression test for model desync

Field Value
input 000102030405060708090a0b0c0d0e0f 101112131415161718191a1b1c1d1e1f 202122232425262728292a2b2c2d2e2f 303132333435363738393a3b3c3d3e3f 404142434445464748494a4b4c4d4e4f 505152535455565758595a5b5c5d5e5f 606162636465666768696a6b6c6d6e6f 707172737475767778797a7b7c7d7e7f 808182838485868788898a8b8c8d8e8f 909192939495969798999a9b9c9d9e9f a0a1a2a3a4a5a6a7a8a9aaabacadaeaf b0b1b2b3b4b5b6b7b8b9babbbcbdbebf c0c1c2c3c4c5c6c7c8c9cacbcccdcecf d0d1d2d3d4d5d6d7d8d9dadbdcdddedf e0e1e2e3e4e5e6e7e8e9eaebecedeeef f0f1f2f3f4f5f6f7f8f9fafbfcfdfeff
expected (empty)

Vector 6 β€” Long run round-trip

Source: Regression test for high-confidence predictions

Field Value
input 61616161616161616161616161616161 61616161616161616161616161616161 61616161616161616161616161616161 61616161616161616161616161616161 … (1024 bytes; the full value is in the source)
expected (empty)

Vector 7 β€” Alternating pattern round-trip

Source: Regression test for renormalization

Field Value
input 61626162616261626162616261626162 61626162616261626162616261626162 61626162616261626162616261626162 61626162616261626162616261626162 61626162616261626162616261626162 61626162616261626162616261626162 61626162616261626162616261626162 61626162616261626162616261626162
expected (empty)

← All algorithms