I think we can simply remove benchmarks for StepRng.

I think we can simply remove benchmarks for StepRng.

Yes...

The more questionable case are the tests in src/rng.rs and src/rngs/reseeding.rs. Some of those appear to want a ConstRng.

Then there is a usage in uniform_float.rs which actually uses stepped output (lowering_max_rng).

Maybe we should keep StepRng (but make it private to the crate)?

Note: I updated the motivation above (last two paragraphs). I also note one usage in rand_distr (testing the Triangular distribution), but we don't need to care much about that (we could copy the StepRng code to that crate).

The easiest solution would be to just use #[allow(deprecated)] on the affected test modules. In future it may be worth to add const initialization methods to some of small PRNGs and rewrite tests accordingly

add const initialization methods to some of small PRNGs

This is not why those tests use StepRng. They use it in order to feed specific inputs to the random algorithms being tested. This use-case is a valid use of StepRng, but one that's only applicable when writing random algorithms (as in rand and rand_distr, though the latter only has one use-case).

They use it in order to feed specific inputs to the random algorithms being tested.

For this we can use an internal mock RNG which just reads values from a static array. Something like this:

pub MockRng32 {
    values: &'static [u32],
    pos: usize,
}

Or indeed keep StepRng and make it private (which is equivalent to removal for external users).

Tests still generate warnings like this:

warning: use of deprecated constant `rngs::mock::tests::test_bool`
   --> src/rngs/mock.rs:99:5
    |
99  | /     fn test_bool() {
100 | |         use crate::{distr::StandardUniform, Rng};
...   |
105 | |         assert_eq!(&result, &[false, true, false, true, false, true]);
106 | |     }
    | |_____^
    |
    = note: `#[warn(deprecated)]` on by default

There's already #![allow(deprecated)] at the top of the file; additional annotations don't help. I'm guessing this may be a side effect of the #[test] annotation.

Shall I just delete the tests?

I think, yes. It does not look like such tests add anything useful (I assume stability of distributions using a proper PRNG is tested in relevant modules).

Hello,

Is there any suggestion for downstream projects that use this now deprecated feature in tests? It seems like there is some admission that it's useful for testing yet the only solution seems to be to flag with allow deprecated which doesn't seem like a good long-term solution.

Users should use a proper PRNG, e.g. rand::rngs::SmallRng or rand_xoshiro directly if you want reproducibility across targets.

Thanks, using SmallRng::seed_from_u64(1) works for determinism. I'm not sure how I came to use StepRng in the first place.

Thanks, using SmallRng::seed_from_u64(1) works for determinism. I'm not sure how I came to use StepRng in the first place.