Any guidance on keeping reproducibility from seeded RNGs?

[cuviper]

In rust-num/num-bigint#53, the author is offering a faster way to generate the random bits for a BigUint. This is great, but it occurs to me that this will change what value is generated by a seeded RNG, and this could be considered a kind of breaking change. Do you have any guidance in this area?

[cuviper]

I guess this is the value-stability discussed in #408, whereby we should call this a breaking change?

[dhardy]

This is something we haven't finalised our policy on yet. Since num-bigint is not yet stable (like Rand) I guess we shouldn't worry too much.

IMO value-breaking changes should at least be mentioned in the changelog and should not land in "patch" versions, but I can't tell other crate authors what to do.

Value-stability is not just about random number generation, but is probably more important here than elsewhere, so I suppose it would be useful if we have an official policy on this that other crates can refer to (should they wish)?

[vks]

I don't think it could be considered a semver breaking change, but I agree that we should not break reproducibility in patch versions (except maybe for bugfixes).