On: What would it take to add refinement types to Rust...

[vlovich123]

Might be annoying if you’re working with floating point but given that each conversion introduces error, it’s probably good to be explicit and recommend internally to be consistent

[stouset → vlovich123]

Thanks to generics you could theoretically work with whatever underlying type makes the most sense for your use-case.

    let v1 : Inches<Ratio> = Ratio::new(5, 8).into();
    let v2 : Inches<Ratio> = Ratio::new(3, 8).into();

    let v3 : Feet<_> = (v1 + v2).into();

    assert_equal!(v3, Ratio::new(1, 12));

[throwawaymaths → stouset]

no that's not the point. if you let rust automagically decide when and where to apply conversions you could easily wind up in a situation where you have more operations than you need, which increases numerical error, and also be a bitch to uncover or refactor to minimize conversions.

[stouset → throwawaymaths]

Rust doesn’t automatically apply conversions, full stop.

And with a thoughtful approach to the API, you could avoid numerical error entirely by using integral types.

[amenghra → stouset]

That’s a good starting point. The temperature delta issue remains: adding and subtracting temperatures should create a different unit. Thankfully there aren’t a lot of non-linear units.