Move away from nightly?

[thomaseizinger]

Are there any plans to move away from Rust nightly?

Specialization doesn't look like it is going to be stabilized any time soon.

[LLFourn]

It is very easy to remove specialization and just put the default (slow) implementation of each primitive as the only one and then enable the faster variants with a nightly feature.

Would that be of help?

[edit] From memory we're talking about a 70 micro second slow down for a secret key to public key (for example). This would make things like: https://docs.rs/sigma_fun/0.3.1/sigma_fun/ext/dl_secp256k1_ed25519_eq/index.html a second or so slower (would need to bench it).

[thomaseizinger]

That would be really nice as secpfun is the only dep forcing us on nightly atm!

I'll see to send a PR for that!

[LLFourn]

So to be clear the things that need to be done are:

1. from the base implementation (the one where everything is marked default for each impl chain) remove default i.e.
   get rid of them for here unless nightly feature enabled from here

secp256kfun/secp256kfun/src/op.rs

Lines 93 to 108 in 26a0f27

	impl<T1, S1, Z1, T2, S2, Z2> PointBinary for (&Point<S1, T1, Z1>, &Point<S2, T2, Z2>) {
	default fn add(self) -> backend::Point {
	let (lhs, rhs) = self;
	ConstantTime::point_add_point(&lhs.0, &rhs.0)
	}
	default fn sub(self) -> backend::Point {
	let (lhs, rhs) = self;
	ConstantTime::point_sub_point(&lhs.0, &rhs.0)
	}
	default fn eq(self) -> bool {
	let (lhs, rhs) = self;
	ConstantTime::point_eq_point(&lhs.0, &rhs.0)
	}
	}

I suggest trying to make a macro that parses a list of functions and outputs two versions (one with not(feature = nightly) and the other with feature = nightly.

2. for each specializing impl just put #[cfg(feature = nightly)] on top of it
3. Put README.md embedding which needs extended_key_value_attributes under nightly as well:

   secp256kfun/secp256kfun/src/lib.rs

   Line 2 in 26a0f27

   #![feature(rustc_attrs, min_specialization, extended_key_value_attributes)]

this should be all we need to run on stable.

[thomaseizinger]

Fixed in #65.