This is good to think about! Often when programming games that need certain properties, I'll write my own toy PRNG, but I'll admit I'd love to learn more about the craft.
Would you have any suggested reading material on the topic?
Would you have any suggested reading material on the topic?
It's advocating for a particular new kind of PRNG, but it also includes a lot of great explanations and citations for more reading.
Yes, since the original post wanted seedability:
> > For games, the most important features (other than the features of an RNG that all applications desire, such as good performance, high period, etc.) are being able to seed the state, and save/restore the state (required for save game support if you need seeding and want that to work after they save and restore the game).
Given that restriction, what's wrong with the generator I suggest?
No, it's a pseudorandom generator. It's a key attribute of a cryptographically secure function like SHA256 that its output is indistinguishable from random bits.
> You are just returning a different representation of the seed you already have.
I'm returning a function of the seed, yes. That's the whole point of a seeded PRNG: that seed(x); rng(); rng(); rng() will always return the same sequence of results. This is used for example in procedural generation, to ensure that a dungeon level or map chunk is always generated the same way (a cool hack, then, is to just store the seed instead of the level itself, and rerun the level-generation algorithm with the seed when one wishes to regenerate the level). In such an application one wants the appearance of randomness, but with repeatability.
Which was what the original post asked for.
What I suggested is not secure for generating keys or for other cryptographic purposes.
If it does, doesn't that mean it's flawed as a cryptographic hash?
Performant enough strongly depends on the game. However looking at the implementation of the algorithm on wikipedia[0], that looks extremely slow by RNG standards. Most RNGs finish in a handful of instructions, but building the 'message schedule array' takes 64 steps, and the 'main compression loop' also takes 64 steps. Consider that to reduce these to a range you'll have to throw out iterations (you need to do this with any RNG to avoid bias), you're looking at a lot of wasted time.
Maybe you can afford to waste that time, but given that there are much simpler and more efficient RNG algorithms, I'm not sure why you wouldn't use one of those.
Of course you need to make sure the initial seed is sufficiently random otherwise you'll risk generating the same sequence each time, which could make this a chicken->egg->chicken situation!
The Art of Computer Programming, Vol. 2, (c) 1969, by Donald Knuth, has quite an extensive discussion of random numbers, their properties, algorithms, and tests. It's about 160 pages in my edition. It's very readable, and even if you skip all the exercises you will still get a very good grounding.
I don't know how much has changed in his more recent editions. And I don't know how if there are better books. But, in general, you can't go wrong by starting with Knuth. He's one of the all time greats!