Switch to portable RNG in examples

[TimLeach635]

Objective

Fixes issue #12613 - the RNG used in examples is deterministic, but its implementation is not portable across platforms. We want to switch to using a portable RNG that does not vary across platforms, to ensure certain examples play out the same way every time.

Solution

Replace all occurences of rand::rngs::StdRng with rand_chacha::ChaCha8Rng, as recommended in issue #12613

---

Changelog

• Add rand_chacha as a new dependency (controversial?)
• Replace all occurences of rand::rngs::StdRng with rand_chacha::ChaCha8Rng

[github-actions]

Welcome, new contributor!

Please make sure you've read our contributing guide and we look forward to reviewing your pull request shortly ✨

[BD103]

Looks good! Thank you :)

[Bluefinger]

ChaCha8Rng implements Clone I think, so if we want, we could just clone a single instance over and over if they are all using the same seed though. Buuuut, it seems like a bit of a code smell to reuse the same seed for multiple RNG instances, even if an example. If we want things to be deterministic with generated seeds, we could make use of a single ChaCha8Rng::seed_from_u64(42), and then everything else gets ChaCha8Rng::from_rng(&mut rng).unwrap().

[TimLeach635]

The documentation says the use of unwrap is discouraged, do you think it's fine here or shall I try to use a match or something? (I'm very new to Rust, so apologies, I'll have more questions than usual!)

[BD103]

ChaCha8Rng implements Clone I think, so if we want, we could just clone a single instance over and over if they are all using the same seed though. Buuuut, it seems like a bit of a code smell to reuse the same seed for multiple RNG instances, even if an example. If we want things to be deterministic with generated seeds, we could make use of a single ChaCha8Rng::seed_from_u64(42), and then everything else gets ChaCha8Rng::from_rng(&mut rng).unwrap().

I disagree with cloning in this case. I think it will make reading the code more difficult, since the variable will need to be created beforehand. As for changing the seeds, it may be good practice in an actual game but changing the seeds now will mess up the screenshot tests. I don't think it's worth the change.

[Bluefinger]

The documentation says the use of unwrap is discouraged, do you think it's fine here or shall I try to use a match or something? (I'm very new to Rust, so apologies, I'll have more questions than usual!)

For OS/Hardware sources which can fail, then yes, you shouldn't unwrap(). But for deterministic pseudorandom sources where you are in full control of the seeding (such as with ChaCha8Rng), then usage of the Rng source will never fail. Not all entropy sources are fallible, only when they interface hardware/OS APIs can failures happen. If a PRNG algorithm can fail, then it is not deterministic in output.

[Bluefinger]

I disagree with cloning in this case. I think it will make reading the code more difficult, since the variable will need to be created beforehand. As for changing the seeds, it may be good practice in an actual game but changing the seeds now will mess up the screenshot tests. I don't think it's worth the change.

Then perhaps we need some comments here to explain that a bit more clearly. But yeah, if we have already a bunch of screenshot test suites in place, we shouldn't change that unless absolutely necessary.

[bushrat011899]

Looks good to me! This is an important issue for any multiplayer games, so it's definitely worth calling out in the examples.