Less errors in RandomBits methods

[fjarri]

A possible change of contract of RandomBits that can eliminate some error/panic cases: in the methods that take bits_precision, treat it as at_least_bits_precision in BoxedUint. That is if RandomBits::random_bits_with_precision() is called with bits_precision == 128 for U256, it will return U256 instead of panicking. Or is that too dynamic?

[tarcieri]

Is there a use case where this would actually make sense? Otherwise it seems like an error condition to me

[fjarri]

In any use case where the user generalizes on T: Integer? BoxedUint methods are already automatically expanding precision to the nearest limb size multiple, this seems to be the same principle.

[tarcieri]

But in particular I'm asking about miscomputing the size and causing the error case. Why is it being miscomputed? Why is it being passed explicitly instead of inferred?

[fjarri]

Because RandomBits has to work for both Uint and BoxedUint. bits_precision is needed in the latter case, but for Uint it duplicates the information that's already contained in the type.

[tarcieri]

But more specifically, what is an example of a case where random_bits_with_precision is being preferred over random_bits, where the caller wants to pass an explicit precision but for some reason can't compute it correctly for the size of the underlying type?

[fjarri]

I admit I don't have a specific example in mind, I see it as more of a quality of life improvement — the general principle being, when it's possible, instead of an error do something that doesn't lead to incorrect behavior.

[tarcieri]

Shouldn't the answer for those cases be "use random_bits instead of random_bits_with_precision"?