Support seeded random number generators

[julasamer]

Hi!

I've added support for seeded RNG by allowing the passing of any RandomNumberGenerator to Faker.

Caveats:

• I ignored Swift <= 4.2. It might not compile with the older swift versions; if it does, it won't used the RNG that was passed in. Not sure if dropping support for 4.2 is an option at this point.
• I didn't write a unit test (I'm not familiar with QuickSpec and don't have more time to spend on this right now).

That being said, you can use this like this:

let faker = Faker(randomNumberGenerator: SomeAdvancedRandomNumberGenerator())

Unfortunately, seeded RNG is somewhat hidden in iOS; my first idea was to use the RNG provided by GameKit:

 import GameKit

 extension GKMersenneTwisterRandomSource: RandomNumberGenerator {}
 let faker = Faker(randomNumberGenerator: GKMersenneTwisterRandomSource(seed: 42))

We need to make GKMersenneTwisterRandomSource conform to RandomNumberGenerator, which is fortunately quite easy. We can then use it as the seeded RNG.

However, this requires that GameKit is available. In the end, I didn't want introduce a GameKit dependency, and the whole thing is overkill anyway for my testing purposes, so I implemented a custom "RNG" which works fine for my case:

public class AdvancedRandomNumberGenerator: RandomNumberGenerator {
    var currentIndex = 0
    
    // chosen by fair dice roll
    // guaranteed to be random
    lazy var randomValues: [UInt64] = [4151371853236615391, 7134936793715064765, 8637388537612094686, ...]
    
    public func next() -> UInt64 {
        currentIndex += 1
        return randomValues[currentIndex % randomValues.count]
    }
}

[mrylmz]

Isn't the Swift random API seedable, like UInt64.random(in:)

[julasamer]

@mrylmz: That's the API I used. You have to pass something to the in parameter though. That's where the GKMersenneTwisterRandomSource comes in: a seedable RNG. The Swift random API itself doesn't have a seedable RNG afaik, you can just specify some RNG to use.

[crayment]

Would love to have this! Would be nice to have predictably random mock data for use in tests.

[Tyler-Keith-Thompson]

What ever happened here? Determinism in these tests would be amazing. I'd be happy to take another crack at this, if the only thing preventing moving forward is that it's out of date.