invlex

invlex — Inverse Lexicographic Encoding

For encoding numbers so that higher numbers appear first, sorted lexicographically (i.e. in a cloud storage bucket that only has ascending LIST operations), in as few bytes as possible.

Usage

npm install --save-dev invlex
pnpm add -D invlex

import invlex from 'invlex'

invlex.encode(0) // "zf"
invlex.encode(1) // "ze"
invlex.encode(15) // "z0"
invlex.encode(16) // "yef"
invlex.encode(1959) // "x858"
invlex.encode(49584741) // "td0b659a"
invlex.encode(350371465044) // "qae6c3bb8ab"
invlex.encode(84296151321634) // "ob35545d833dd"

import { base36 as invlex } from 'invlex'

invlex.encode(0) // "zz"
invlex.encode(1) // "zy"
invlex.encode(15) // "zk"
invlex.encode(16) // "zj"
invlex.encode(1959) // "xyhk"
invlex.encode(49584741) // "v6h862"
invlex.encode(350371465044) // "svj1hvjcb"
invlex.encode(84296151321634) // "r64avryqxp"

import { base62 as invlex } from 'invlex'

invlex.encode(0) // "zz"
invlex.encode(1) // "zy"
invlex.encode(15) // "zk"
invlex.encode(16) // "zj"
invlex.encode(1959) // "yUO"
invlex.encode(49584741) // "vwdwk6"
invlex.encode(350371465044) // "ttpYL2Pz"
invlex.encode(84296151321634) // "sc3uyZfif"

Also defined is base32, and each export has a corresponding .decode function.

Format

Numbers are converted to a hex string (or base-32 or base-64 string, see below), then inverted (so "001" becomes "ffe"), then prefixed with a single letter that encodes the length, starting at 'z' and moving up the alphabet, meaning that longer (larger) numbers appear first when sorted lexicographically.

This uses a lot less bytes than a naive approach (Number.MAX_SAFE_INTEGER - n, for example), and by converting to hex/base32/64, makes the IDs look opaque enough to not look obviously "wrong".

Development

pnpm install
pnpm test

Benchmarks

Absolutely no effort has been put into optimising this, but I was curious. There are two code paths within invlex, base16 and base32 use Number.toString(radix) and parseInt(radix), whereas base36 and base64 both use a lookup into a preset array.

Running "Baseline" suite...
Progress: 100%

  Just randoms (baseline):
    72 705 140 ops/s, ±1.62%   | fastest

Finished 1 case!
Running "Converting random numbers to inverse lexicographic encoding" suite...
Progress: 100%

  base16.encode():
    1 832 049 ops/s, ±1.13%   | slowest, 38.26% slower

  base32.encode:
    2 007 291 ops/s, ±3.42%   | 32.35% slower

  base36.encode:
    2 733 974 ops/s, ±2.19%   | 7.86% slower

  base62.encode:
    2 967 347 ops/s, ±0.96%   | fastest

Finished 4 cases!
  Fastest: base62.encode
  Slowest: base16.encode()
Running "Converting inverse lexicographic encoding back to numbers" suite...
Progress: 100%

  base16.decode():
    2 326 712 ops/s, ±2.45%   | 15.23% slower

  base32.decode():
    2 744 656 ops/s, ±1.15%   | fastest

  base36.decode():
    1 947 971 ops/s, ±2.50%   | slowest, 29.03% slower

  base62.decode():
    1 962 106 ops/s, ±4.35%   | 28.51% slower

Finished 4 cases!
  Fastest: base32.decode()
  Slowest: base36.decode()
Running "Converting to and from inverse lexicographic encoding" suite...
Progress: 100%

  base16.encode():
    1 143 259 ops/s, ±6.10%   | slowest, 19.05% slower

  base32.encode:
    1 346 793 ops/s, ±1.76%   | 4.63% slower

  base36.encode:
    1 359 610 ops/s, ±1.59%   | 3.73% slower

  base62.encode:
    1 412 220 ops/s, ±3.46%   | fastest

Finished 4 cases!
  Fastest: base62.encode
  Slowest: base16.encode()

Acknowledgements

This was @kentonv's idea, I just ported it to JS and tried out non-base16 implementations.