Cannot use Rng methods on `Gen` when implementing `Arbitrary`

[nicoonoclaste]

I noticed when trying to update a project to quickcheck 1.x, that Gen is now an opaque struct and users implementing Arbitrary can only use Gen::choose to get randomness. This isn't sufficient in some usecases, which were previously served by using Gen's Rng methods: for instance, in uutils' factor I need to draw integers from large ranges known at runtime, to generate integers of known factorization:

impl quickcheck::Arbitrary for Factors {
    fn arbitrary(gen: &mut quickcheck::Gen) -> Self {
        use rand::Rng;
        let mut f = Factors::one();
        let mut g = 1u64;
        let mut n = u64::MAX;

        // Adam Kalai's algorithm for generating uniformly-distributed
        // integers and their factorization.
        //
        // See Generating Random Factored Numbers, Easily, J. Cryptology (2003)
        'attempt: loop {
            while n > 1 {
                n = gen.gen_range(1, n);
                if miller_rabin::is_prime(n) {
                    if let Some(h) = g.checked_mul(n) {
                        f.push(n);
                        g = h;
                    } else {
                        // We are overflowing u64, retry
                        continue 'attempt;
                    }
                }
            }

            return f;
        }
    }
}

(Technically, I could repeatedly use gen.choose(&[0, 1]) to draw individual, random bits, but this would be beyond silly)

[nicoonoclaste]

Seems related to #267

[BurntSushi]

Yes. The plan is to make a seed available, and then you should be able to materialize any RNG from that.

[audunska]

This really hurts the ability to write good Arbitrary impls for complicated data. I'm sure there are lots of examples, but my use case can be reduced to just a vector of vectors, Vec<Vec<u32>>. To make sure I sample the data properly I want to generate both tall skinny arrays and short wide ones. This means that the size parameter must be divided randomly into two factors in the arbitrary method, which requires a bounded random float.

If the issue with gen_range is that it exposes rand types, could I propose simply adding a public method Gen::gen_uniform(&mut self) -> f32 which generates a uniformly distributed float in the unit interval? That should satisfy most needs.

[divagant-martian]

Is there a known plan to make this work that would be followed by a first time contributor?

[bergmark]

Perhaps a bit hacky, but the workaround I ended up using was:

pub fn range(gen: &mut Gen, range: Range<u64>) -> u64 {
    u64::arbitrary(gen) % (range.end - range.start) + range.start
}

[ozongzi]

also I could not use Fake::fake_with_rng when I use quick check 1.x