Last active
August 22, 2017 22:30
-
-
Save davidkellis/2c5a1226cc67924daf2c18e2583b612c to your computer and use it in GitHub Desktop.
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
package hashids | |
MinAlphabetLength = 16 | |
SepDiv = 3.5 | |
GuardDiv = 12 | |
DefaultAlphabet = "abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ1234567890" | |
DefaultSeps = "cfhistuCFHISTU" | |
struct Encoder { | |
salt: String | |
minHashLength: Int | |
alphabet: String | |
seps: String | |
guards: String | |
} | |
fn NewEncoder(salt: String, minHashLength: Int) -> Encoder = NewEncoder(salt, minHashLength, DefaultAlphabet, DefaultSeps) | |
fn NewEncoder(salt: String, minHashLength: Int, candidateAlphabet: String, candidateSeps: String) -> Encoder = { | |
validate(minHashLength, candidateAlphabet) | |
(alphabet, seps, guards) = setup(salt, minHashLength, candidateAlphabet, candidateSeps) | |
Encoder(salt, minHashLength, alphabet, seps, guards) | |
} | |
fn validate(minHashLength: Int, alphabet: String) { | |
raise "The alphabet may not include spaces." if alphabet.include?(" ") | |
raise "The alphabet must include at least $MinAlphabetLength unique characters." if alphabet.chars.uniq.count < MinAlphabetLength | |
raise "The min length must be greater than or equal to 0." if minHashLength < 0 | |
} | |
// set up alphabet, seps, and guards | |
fn setup(salt: String, minHashLength: Int, alphabet: String, seps: String) -> (String, String, String) { | |
alphaChars = alphabet.chars.to[LinkedHashSet Char] // http://docs.oracle.com/javase/8/docs/api/java/util/LinkedHashSet.html | |
sepsChars = seps.chars.to[LinkedHashSet Char] | |
commonChars = sepsChars.intersection(alphaChars) // could also use sepsChars & alphaChars, but the order of the operands | |
// would be significant (because we want the order of the elements | |
// in sepsChars to remain intact), which might be unintuitive | |
alphabet = (alphaChars - commonChars).join | |
seps = commonChars.join // the order here is correct, because `sepsChars & alphaChars` maintains | |
// the relative order of the elements as they existed in sepsChars | |
seps = shuffle(seps, salt) | |
if seps.empty? || (alphabet.length / seps.length) > SepDiv { | |
sepsLength = (alphabet.length / SepDiv).ceil | |
sepsLength = 2 if sepsLength == 1 | |
if sepsLength > seps.length { | |
diff = sepsLength - seps.length | |
seps += alphabet.take(diff) | |
alphabet = alphabet.drop(diff) | |
} else { | |
seps = seps.take(sepsLength) | |
} | |
} | |
alphabet = shuffle(alphabet, salt) | |
// set up guards | |
gc = (alphabet.length / GuardDiv).ceil | |
if alphabet.length < 3 { | |
guards = seps.take(gc) | |
seps = seps.drop(gc) | |
} else { | |
guards = alphabet.take(gc) | |
alphabet = alphabet.drop(gc) | |
} | |
(alphabet, seps, guards) | |
} | |
fn shuffle(alphabet: String, salt: String) -> String { | |
return alphabet if salt.empty? | |
v = 0 | |
p = 0 | |
chars = alphabet.chars.toArray | |
slen = salt.length | |
(alphabet.length-1).to(1).each { i => | |
v = v % slen | |
p += n = salt[v].codePoint | |
j = (n + v + p) % i | |
chars[i], chars[j] = chars[j], chars[i] | |
v += 1 | |
} | |
chars.join | |
} | |
fn encode(e: Encoder, numbers: *Int) -> String { | |
return "" if numbers.empty? || numbers.any?(_ < 0) | |
alphabet = e.alphabet | |
length = numbers.length | |
hashInt = numbers.iterWithIndex.reduce(0) { sum, (n, i) => sum + n % (i + 100) } | |
lottery = ret = alphabet[hashInt % alphabet.length] | |
length.times { i => | |
num = numbers[i] | |
buffer = lottery + e.salt + alphabet | |
alphabet = shuffle(alphabet, buffer.take(alphabet.length)) | |
last = hash(num, alphabet) | |
ret += last | |
if i + 1 < length { | |
num %= last.codePoint + i | |
ret += seps[num % seps.length] | |
} | |
} | |
if ret.length < e.MinHashLength { | |
ret = guards[(hashInt + ret[0].codePoint) % guards.length] + ret | |
if ret.length < e.MinHashLength { | |
ret += guards[(hashInt + ret[2].codePoint) % guards.length] | |
} | |
} | |
halfLength = alphabet.length \ 2 // integer division | |
while ret.length < e.MinHashLength { | |
alphabet = shuffle(alphabet, alphabet) | |
firstHalf, lastHalf = alphabet.takeDrop(halfLength) | |
ret = lastHalf + ret + firstHalf | |
excess = ret.length - e.MinHashLength | |
ret = ret.substring(excess \ 2, e.MinHashLength) if excess > 0 | |
} | |
ret | |
} | |
fn decode(e: Encoder, hash: String) -> Array Int { | |
return array.empty if hash.empty? | |
ret = array.empty[Int] | |
breakdown = hash.translate(e.guards, " ") | |
array = breakdown.split(" ") | |
i = array.length.in?(2..3) ? 1 : 0 | |
breakdown = array[i] | |
if !breakdown.empty? { | |
lottery = breakdown.first | |
breakdown = breakdown.drop(1).translate(@seps, " ") | |
array = breakdown.split(" ") | |
array.length.times { j => | |
subHash = array[j] | |
buffer = lottery + e.salt + alphabet | |
alphabet = shuffle(alphabet, buffer.take(alphabet.length)) | |
ret << unhash(subHash, alphabet) | |
} | |
ret = array.empty[Int] if e.encode(ret) != hash | |
} | |
ret | |
} | |
fn hash(input: Int, alphabet: String) -> String { | |
len = alphabet.length | |
hash = "" | |
loop { | |
input, rem = input.divmod(len) | |
hash << alphabet[rem] | |
break if input == 0 | |
} | |
hash.reverse | |
} | |
fn unhash(input: String, alphabet: String) -> Int { | |
num = 0 | |
input.length.times { i => | |
pos = alphabet.index(input[i]) | |
raise InvalidInput("unable to unhash") unless pos | |
num += pos * alphabet.length ** (input.length - i - 1) | |
} | |
num | |
} | |
fn main() { | |
e = NewEncoder("foo", 8) | |
encodedId = e.encode(1) | |
puts(encodedId) | |
decodedId = e.decode(encodedId) | |
puts(decodedId) | |
} |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment