Base85 Encoder and Decoder

What is Base85?

Base85 is a family of encodings that turn every 4 bytes into 5 characters. Five base-85 digits can hold 85 × 85 × 85 × 85 × 85 = 4,437,053,125 values, just enough for any 32-bit number, so the output is only 25% larger than the input, compared with 33% for Base64.

Several Base85 alphabets exist and they are not interchangeable: Ascii85 (Adobe, PDF), Z85 (ZeroMQ) and the one on this page. It uses 0-9, A-Z, a-z and 23 symbols: !#$%&()*+-;<=>?@^_`{|}~. The quote characters, comma, period, slash, colon, backslash and square brackets are left out.

RFC 1924, Git and Python

The character set comes from RFC 1924, "A Compact Representation of IPv6 Addresses", published on 1 April 1996 as an April Fools' joke. It proposed writing a 128-bit IPv6 address as one base-85 number of exactly 20 characters. Nobody writes IPv6 addresses that way, but the alphabet was later reused for general binary data:

• Git stores binary files in patches (git diff --binary, git format-patch) as zlib-compressed data encoded with this alphabet, with a length character at the start of each line.
• Python 3.4 and later provides base64.b85encode() and b85decode(), whose output matches this tool.
• Mercurial uses the same encoding for binary files in Git-style diffs.

Note that RFC 1924 converts the whole 128-bit address at once, while Git, Python and this tool work in 4-byte groups, so a 16-byte value gives a different result from the RFC method.

Short final groups

When the input length is not a multiple of 4, the last group is filled with zero bytes, encoded, and only the first n + 1 characters are kept: 1 byte becomes 2 characters, 2 bytes become 3 and 3 bytes become 4. The decoder reverses this, so no padding characters are needed. This variant has no shortcut for zero bytes, unlike the z of Ascii85.

# Python 3.4+
import base64
base64.b85encode(b'hello')      # b'Xk~0{Zv'
base64.b85decode(b'Xk~0{Zv')    # b'hello'