Fast SipHash for Javascript

A fast impure Javascript implementation of SipHash.
SipHash is a fast pseudo-random function optimized for short inputs published by Jean-Philippe Aumasson and Daniel J. Bernstein.

Usage

var SipHash = require("./siphash.min.js");
var hash = SipHash(2,4)("somerandomkeystr")("a message");
var lsw = hash.l; // low word of the hash
var msw = hash.h; // high word of the hash
var hex = String(hash); // hexadecimal string
var uint = +hash; // 53-bit unsigned integer

The SipHash function takes two arguments corresponding to SipHash parameters c and d as described in the original paper. They are optional and, if omitted, default of SipHash-2-4 will be used. It returns curried SipHash-c-d function, i.e. a function that takes a key and returns a function that maps messages to their hashes This has obvious advantages in case you want to compute hashes for multiple messages with the same key.

Key should be a binary string of length 16. If a number is given, it will be treated as 53-bit integer zero-padded to form 128-bit key. In all other cases, a random key will be generated so you don't have to provide it at all when it's irrelevant. Random key is generated using crypto.randomBytes when running on nodejs, crypto.getRandomValues when running on WebKit and finally plain old Math.random when nothing better is found.

The finally returned pre-keyed function takes a binary string as its sole argument and computes its hash. It returns an object which has two properties: l and h which are (signed) 32-bit integers corresponding to a low and high words (respectively) of a 64-bit result. When casted to string, it will yield the value of hash in hexadecimal. Primitive value of the returned object is an unsigned integer equal to lower 53 bits of the result (truncated hash).

Examples

Compute test value from appendix A of SipHash paper:

var siphash24 = SipHash(2,4);
var k = "\x00\x01\x02\x03\x04\x05\x06\x07\x08\x09\x0a\x0b\x0c\x0d\x0e\x0f";
var m = "\x00\x01\x02\x03\x04\x05\x06\x07\x08\x09\x0a\x0b\x0c\x0d\x0e";
String(siphash24(k)(m)); // returns 'a129ca6149be45e5'

Hash table protected from hash flooding regardless of goodness of Javascript engine's hash (not the most efficient implementation):

var hash = (function() {
  var h = SipHash()();
  var d = {};
  return {
    set: function(k,v) {
      return d[k+h(k)] = v;
    },
    get: function(k) {
      return d[k+h(k)];
    }
  };
})();

Impurity

I sacrified all the good practices for sake of performance because I thought it would be fun thing to do. And it was!

In order to make things fast, the core of SipRound doesn't make new objects, doesn't call any functions, and doesn't touch anything from outside the nearest scope. These restrictions mean that I deliberately refused to use the usual object-oriented mechanisms provided by the language to solve the problem of lack of 64-bit integers in Javascript. This might seem like bad decision, but it's very beneficial to performance on most platforms.

Of course code like this is hard to write and debug, and it could easily get out of hand. So I opted for a solution typical for C or assembly programmers, i.e. I used a preprocessor. Javascript, like most modern languages, doesn't have it's own preprocessor and relies on native metaprogramming capabilities instead. However, it's quite common today to compile Javascript with tools like Google's Closure Compiler or minify before actually using it, so why couldn't I just put some additional preprocessing in that pipeline.

I've chosen to use good old m4 macro processor. It's not the most suitable processor for Javascript, but, unlike others, it is readily available just about everywhere (it's in POSIX) and powerful enough to actually do the job easily. The code should be compillable to plain Javascript on any Unix-like system which provides m4. I tested it under m4 from the Heirloom Project and under GNU.

Reinventing the wheel

There exist an earlier Javascript implementation of SipHash-2-4 which is linked from SipHash author's page. It is more readable and more adhering to programming standards than this one (and thus, by necessity, much slower). It has different API and it implements only SipHash with c=2 and d=4 (which isn't a real limitation, but there is no real reason for it either; unrolling loops doesn't give that much of an advantage in Javascript). Another difference is that I didn't bother to make this one installable via npm.