Return to Answer

or a variant of that which parses a String. Or both! If you have the string parsing and you have that createRNG, connecting the two together should be simple enough, making something like

Note that the DistributionType never makes it into the MyRNGs themselves - they don't need an external enum to tell them what their type is, because they already know their own types.

or a variant of that which parses a String. Or both! After all, if you have the string parsing and you have that createRNG, connecting the two together should be simple:

Note that the DistributionType never makes it into the MyRNGs themselves - they don't need an external enum to tell them what their type is, because information about their type is contained in, well, their type.

It seems natural to me to more-or-less scrap the enum, and make MyRNG into an interface. That way, the MyRNG type doesn't know anything about generating random numbers, but it can still guarantee that any object belonging to that type does know how to generate random numbers - whatever subtype they belong to will have to define it for them. Then you can easily add various distributions like

And thanks to subtyping, you can just go MyRNG rng = new GaussianRNG(5.66, 1.2);, even though MyRNG has no idea GaussianRNG exists! Need another distribution? You can just make one!

Now, you'll notice I kind of dropped the ability to pick a distribution type at runtime. But there's nothing saying you can't have that too! You're definitely allowed to have something like

public static MyRNG createRNG(DistributionType distribution, double... parameters) {
 switch (distribution) {
 case UNIFORM_DOUBLE:
 if (parameters.length != 2) throw new IllegalArgumentException();
 return new UniformDoubleRNG(parameters[0], parameters[1]);
 case GAUSSIAN:
 if (parameters.length != 2) throw new IllegalArgumentException();
 return new GaussianRNG(parameters[0], parameters[1]);
 case CHOICE:
 return new ChoiceRNG(parameters);
 }
}

It seems natural to me to start without the enum, instead beginning with MyRNG as an interface. The MyRNG type doesn't know anything about generating random numbers, but it can still demand that any object belonging to that type does know how to generate random numbers. Then you can easily add various distributions like

And thanks to subtyping, you can just go MyRNG rng = new GaussianRNG(5.66, 1.2);, even though MyRNG has no idea GaussianRNG exists! And when you want another distribution, you just add another subtype, and nobody necessarily needs updating.

Now, you'll notice I kind of dropped the ability to pick a distribution type at runtime. But there's nothing saying you can't have that too! You're definitely allowed to have factory methods à la

public static MyRNG createRNG(DistributionType distribution, double[] parameters) {
 switch (distribution) {
 case UNIFORM_DOUBLE:
 if (parameters.length != 2) throw new IllegalArgumentException();
 return new UniformDoubleRNG(parameters[0], parameters[1]);
 case GAUSSIAN:
 if (parameters.length != 2) throw new IllegalArgumentException();
 return new GaussianRNG(parameters[0], parameters[1]);
 case CHOICE:
 return new ChoiceRNG(parameters);
 }
}

or a variant of that which parses a String. Or both! If you have the string parsing and you have that createRNG, connecting the two together should be simple enough, making something like

public static MyRNG createRNG(String description) {
 DistributionType distributionType = typeFromDescription(description);
 double[] parameters = parametersFromDescription(description);
 return createRNG(distributionType, parameters);
}

Note that the DistributionType never makes it into the MyRNGs themselves - they don't need an external enum to tell them what their type is, because they already know their own types.