Simplify Math.signbit. Don't ignore sign bit for NaNs

[MaxGraey]

I don't think we should ignore sign for NaNs due to signed -NaN is actually valid and canonical.

• I've read the contributing guidelines

[dcodeIO]

Does this align with https://github.com/tc39/proposal-Math.signbit?

[MaxGraey]

Does this align with https://github.com/tc39/proposal-Math.signbit?

proposal-Math.signbit doesn't specify this. But anyway that proposal on Stage 1 and I don't think it will be move forward in near future. Other implrmrntation in different languages use simply NaN-agnostic version so I suggest follow the same way

[dcodeIO]

My thinking is mostly that the introduction of Math.signbit is based upon exactly that proposal, so I think it should do exactly what is proposed. Code relying on different semantics has better options in AS anyway and does not necessarily have to use that function.

[MaxGraey]

That proposal is not standard and nobody except me it seems interested to move that proposal forward. Anyway even if AS implementation will be slightly different I don't think it broke any code now and in future. First of all signed nans is really rare case, also I don't think somebody will build his logic around sign / non-sign NaNs=) Also proposed behaviour align with other languages which not exclude sign NaNs and just extract sign bit

[MaxGraey]

But I'm ok to leave old behaviour (and close this PR) as more JavaScript way which prefer don't distinguish NaN sign)

[dcodeIO]

Is an interesting situation. Like, the motivation to propose it for JS is that there is no good alternative, and NaN canonicalization is merely a side-effect. Our motivation, however, is to provide it for compatibility with JS, but we don't really do NaN canonicalization. Now, users probably won't care about NaN canonicalization that much, they just want the sign bit. In turn, the question relevant to us becomes "what is the sign of NaN?" independently of canonicalization, leading us here. Interesting.

[dcodeIO]

The radical but also most straight-forward solution here is to document that "AssemblyScript doesn't do NaN canonicalization", then merging this PR. So, slightly in favor of making this change. Can you add a new comment?

[MaxGraey]

For example Object.is(NaN, -NaN) will return false as expected, so Math.signbit with this changes will be only one way to figure outing sign of NaN without bit manipulations

[MaxGraey]

AssemblyScript doesn't do NaN canonicalization

It's not formally correct due to -NaN also canonical and propagated "as is". This also became a discovery for me recently) So perhaps it could be AssemblyScript implementation sensitive to NaN sign which may differ to https://github.com/tc39/proposal-Math.signbit

[github-actions]

🎉 This PR is included in version 0.14.12 🎉

The release is available on:

• npm package (@latest dist-tag)
• GitHub release

Your semantic-release bot 📦🚀