nupp.codec.base64

Standard base64.

const base64 = require("nupp.codec.base64")

const text = base64.encode("Nupp")

print(text)                          -- TnVwcA==
print(assert(base64.decode(text)))   -- Nupp

Both entries are @aot, and both work a word at a time rather than a byte at a time: twelve input bytes are three aligned words and four twenty-four bit groups, and four output bytes are one word. The alphabet and its inverse are const strings, so a compiled entry reads them out of static arrays whose bounds the code generator holds rather than out of a rooted argument whose length it has to load. With aot = "off" this is the same source running on LuaJIT.

bench/base64 measures what that is worth, and measures it against the vectorized C it could have been.

function base64.decode(text: string): string?, string?

assert(nupp.codec.base64.decode("TnVwcA==") == "Nupp")
local bytes, reason = nupp.codec.base64.decode("TnVwcA")
assert(bytes == nil and reason ~= nil)

function base64.encode(value: string): string

assert(nupp.codec.base64.encode("Nupp") == "TnVwcA==")