Vinzent Steinberg

I'm a bit concerned that the trait bounds are leaking implementation details. If we change the implementation and require different traits, it can easily become a breaking change. > I'm...

> The alternative is to only macro-impl the primitives here, then implement bigint separately, and leave others to their own. That's not necessarily the greatest, but how many other Integer...

I rebased on master and made `GcdResult` extendable as suggested by @strake. @cuviper Should I move the added functions to a new `NumberTheory` trait?

> Extended GCD was already merged in #19, so this should now build on top of that if possible. My implementation is not equivalent: It does not use `Clone`, but...

@cuviper Which option do you prefer? 1. Add `NumRef` and `RefNum` to the relevant `Integer` methods (avoiding unnecessary allocations but requiring a breaking change). 2. Rewrite the new methods to...

> `self.denom` is compared to `Zero::zero` instead of calling `is_zero()`. This is fixed. > This might mean that Rust needs some general guideline on how to treat integer overflow. I...

2 sounds like the best option for now.

At least in debug mode, this panics now. Can this be closed, or should it panic in release mode as well?

I'm not sure whether this helps, but the [average](https://crates.io/crates/average) crate provides constant-memory algorithms to calculate various statistics.

What is the status of this? It seems like what is suggested has been implemented in `PartialPivLu`, but not for other decompositions, like Cholesky. *Edit:* Actually it is implemented on...