tabatkins
commented
5 years ago Hahaha, I literally just changed away from a generator-based design.
I think that the current design makes more sense overall.
- The API concordance between
Math.random()andprng.random()is very nice. - Afaict, it's much more likely that random numbers are used either one-off, or inside of a loop, not directly pulled as the loop variable. Both of those cases are less convenient with iterators; you have to say
prng.next().value. (Basically, a sequence of random numbers is typically treated differently from other infinite iterators; it's pulled as needed, not until the value hits some condition.) - Making an iterator of the PRNG is trivial:
function* rand(...args) { const prng = Math.seededPRNG(...args); while(1) yield prng.random(); }. Doing the reverse, while not hard, is less trivial.
I encourage you to actually go thru some code examples that currently use Math.random() and see how they look with prng.random() vs an iterator.
As an aside, I consider the .next() method of generators to be, most of the time, an implementation detail, not an affirmatively-exposed part of your API. You rarely call next() directly in Python, for example. The {next,done} return value is inconvenient and weird to work with directly.
It would seem to be nice for
Math.seededPRNG()to return an iterator, so instead of.random()the user would instead call the more standard.next(). It would also be easier to reason aboutMath.seededPRNG()as a generator function.