Move OsRng to rand_core

[dhardy]

Closes #854 which discuses this idea.

We don't need to rush this decision, so this PR will stay open a while for discussion.

[dhardy]

Note: this implies one breaking change when next publishing rand: OsRng::new is now removed. However this was deprecated since 0.7.0 so I think this is acceptable.

[vks]

What about the rand changelog?

[dhardy]

There's not any change (unless removal of OsRng::new counts).

[newpavlov]

I think it also worth to release rand_os v0.3.0 with a description stating that users should use rand_core::OsRng instead.

[dhardy]

You mean we should count this as a breaking release for rand_os?

[newpavlov]

No, I mean, if we deprecate rand_os, I think we should notify users that it should not be used anymore. Releasing an empty rand_os v0.3 is a good way of doing so in my opinion.

[dhardy]

You don't think the deprecation warning and comment I added to the README already are enough?

[newpavlov]

Ah, haven't noticed that.

[burdges]

I do kinda think you're better off with a free fn instead of the SeedableRng::from_entropy, but not my reason is that rand_core 0.5 does not have from_entropy, so this avoids one change that people like me may mistake for breaking changes. ;)

#[cfg(feature="getrandom")]
fn make_rng<R: SeedableRng, S: Default+RngCore = OsRng>() -> R {
    R::from_rng(S).unwrap()
}

[newpavlov]

I think it's ready to be merged? Or do you want to merge together with #864?

[dhardy]

Yes, I think it's ready. #864 will need to be rebased; unfortunately I will have very limited time available over the next few weeks.