I ran into an odd “bug” today when I was running some unit tests in various browsers. I had run the tests in Firefox many times before today, and even IE but apparently not Chrome (v19-dev) yet. When I ran them in Chrome it consistently failed one test because two values I was calculating did not match.
When I really dug into what was happening I realized that the issue was that I was assuming that if I filled an array with 100,000 Math.random() values that they would all be unique (there wouldn’t be any collisions). Turned out that in Chrome that is not true.
In Chrome I was consistently getting at least two pairs of values that matched out of 100,000. Firefox and IE9 never experience a collision. Here is a jsfiddle I wrote just for testing this that creates 1M Math.random() entries in an array: http://jsfiddle.net/pseudosavant/bcduj/
Does anyone know why the Chrome pseudo-random number generator that is used for Math.random is really not that random? It seems like this could have implications for any client-side js encryption routines that ever use Math.random.
Apparently Math.random() in V8 only works with 32 bit values (and didn’t even correctly randomize all of those in the past). And with 32 bits, the probability of a collision reaches 50% around 2^16 = 65k values…