Expand const fn coverage

[vorot93]

Follow up to #318

[parity-cla-bot]

It looks like @vorot93 signed our Contributor License Agreement. 👍

Many thanks,

Parity Technologies CLA Bot

[dvdplm]

What are your thoughts on what criteria to use for deciding what should be const and what shouldn't? E.g. in uint there are a few methods left non-const: low_u128(), bit(), split(), split_u128() – what was the reasoning there?

Another question I have is about the removed #[inline]s: I thought const fn meant it can be transformed to a constant at compile time, but is not guaranteed to be. In other words, that in some code it'd be constified but other times called as a normal function. Am I mistaken?

[vorot93]

@dvdplm my idea is to make pretty much everything const fn unless it is not allowed to be such. That means that the most basic of methods with no mutation can become const fn today.

I missed the methods you listed and will make them const fn right away.

re: inline - we would need a compiler expert here, but I'd expect any function that uses const fn to be classified by internal heuristics as potentially inline-able. This is not something I'm completely sure about though and can restore the attribute.

[dvdplm]

I missed the methods you listed and will make them const fn right away.

Got it, cool. It would be great to have a tool/lint to help us find code that can be constified. :)

[dvdplm]

re: inline - we would need a compiler expert here, but I'd expect any function that uses const fn to be classified by internal heuristics as potentially inline-able. This is not something I'm completely sure about though and can restore the attribute.

Reading up a little about this, my understanding is that a const fn implies a guarantee that the function is callable in a const context; when called in a non-const context the const fn is called like a normal function. E.g. given:

const fn is_even(x: usize) -> bool { x % 2 == 0 }

…then is_even(42) is guaranteed to be const, but if I have fn random() -> usize somewhere and call is_even(random()) then it is not folded.

My conclusion is that we want to keep the #[inline].

[vorot93]

@dvdplm thank you for the write-up. I have added back #[inline] in the last commit.

[vorot93]

By the way, there is an actual clippy lint for potential const fn now :)
https://rust-lang.github.io/rust-clippy/master/index.html#missing_const_for_fn