base32k
(You should really use base2048 instead, it's far superior. I had fun with this one, but I didn't know about base2048 at the time.)
base32k is a slightly whimsical binary-to-text encoding, which transforms raw binary data into (possibly obscene combinations of) UTF-8 characters from the CJK and Hangul unicode blocks. Its alphabet consists of 2^15 = 32768 characters, hence the name. In comparison to other encodings like base64 or base122, this does not save space in terms of bytes, but it is smaller than those two in terms of characters. This is only useful when for some reason the medium (* cough * twitter * cough * ) is character-limited rather than byte- limited.
Example
$ echo testing | base32k
整棦릥茻l
$ echo 整棦릥茻l | base32k -d
testing
Installation
Library
go get github.com/grandchild/base32k
Executable Binary
go get github.com/grandchild/base32k/base32k
Encoding Ratio
base32k has an encoding ratio of 15 bits per unicode glyph, which amounts to a ratio of 15/24 (0.625) plus one byte padding in 14 out of 15 cases. This makes it a worse encoding than base64, which has a 3/4 (0.75) and base122, which has 7/8 (0.875).
A twitter message may be 280 characters long, but only 140 CJK glyphs. Still, this encoding slightly outperforms base64 and even base122 in the space available in a single tweet:
| space ratio | char ratio | bytes per tweet | |
| base64 | 0.75 | 6 | 210 |
| base122 | 0.875 | 7 | 245 |
| base32k | 0.625 | 15 | 256 |
| ( more is better for all columns ) | |||
base32k outperforming base122 on twitter results from the fact that twitter counts a CJK or Hangul glyph as two characters, whereas in UTF-8 it's actually 3 characters. This gives us, in effect, an encoding ratio of 15/16 over base122's 7/8, a slight advantage.
So, given good-enough font coverage of the basic multilingual unicode plane, this can be used to transmit data in situations where characters are limited, rather than disk space.
Stability
This implementation will run out of memory when en-/decoding very large chunks of data (several gigabytes). But since this is aimed at character-limited settings this is not likely to be an issue.