Why is `NonZeroScalar::invert()` fallible?

[fjarri]

It feels like, since we know that it is nonzero, there should be some fn invert(&self) -> NonZeroScalar, or at least -> Scalar, instead of -> CtOption<Scalar>, which will never have the None variant. Isn't guaranteed invertibility one of the major uses of a NonZeroScalar type?

[tarcieri]

The issue is Invert is a trait and the current definition uses CtOption<Self::Output> as a return type.

It would probably make sense to get that CtOption out of the trait signature and allow NonZeroScalar's impl to be infallible. Other impls could choose to set type Output = CtOption<Self>.

Unfortunately this would require a breaking change to the elliptic-curve crate so it can't happen until the next release.

[fjarri]

Or perhaps we could have Invert (infallible) and TryInvert (defaulting to Invert if it's defined)?

[tarcieri]

RustCrypto/traits#894 implemented the proposed changes:

• CtOption is no longer a mandatory part of the return type for Invert
• The blanket impl of Invert returns a CtOption
• The impl of Invert for NonZeroScalar now infallibly returns a NonZeroScalar