Quantity: add dunder methods

[andrewgsavage]

transferred from quantity-dev/quantity-api#6

[lucascolley]

I've added some tentative suggestions but I'm not sure that they are correct — @jorenham do my suggestions look closer to what we would want here?

[andrewgsavage]

I feel like we need to get a type checker going to know what works and doesnt

[lucascolley]

I feel like we need to get a type checker going to know what works and doesnt

Yes, but we then also need a minimal an implementation to check against :)

[andrewgsavage]

I feel like we need to get a type checker going to know what works and doesnt

Yes, but we then also need a minimal implementation to check against :)

does it have to be minimal? would a fork of pint (or another module) work?

[lucascolley]

would a fork of pint (or another module) work?

yeah, I think so!

[jorenham]

I've added some tentative suggestions but I'm not sure that they are correct — @jorenham do my suggestions look closer to what we would want here?

My mental typechecker did not report any errors 👌🏻

[jorenham]

I feel like we need to get a type checker going to know what works and doesnt

Exactly; writing annotations without a type-checker is like programming without a compiler or interpreter

[lucascolley]

the linked issue also suggests:

__floordiv__, __mod__, __divmod__, shifts, &, ^, |, ...

[lucascolley]

type checkers are happy! Any opinions on #5 (comment) @nstarman @neutrinoceros ?

[lucascolley]

alright, let's get this in, we can always revisit other methods as and when need arises. thanks all!

[andrewgsavage]

oh I meant to ask if we want to keep rpow

[lucascolley]

I think so. If we can do x ** q.value, we should also be able to do x ** q.

[andrewgsavage]

that only makes sense for a unitless quantity, you shouldn't be able to do 2 ** Quantity(2, "m"). I think its reasonable for a library to implement this or not implement it

[lucascolley]

ah yeah, good point, we should probably remove it then

[jorenham]

that only makes sense for a unitless quantity, you shouldn't be able to do 2 ** Quantity(2, "m"). I think its reasonable for a library to implement this or not implement it

not sure if it's a thing, but maybe it makes sense for logarithmic units like decibels?

[nstarman]

Agreed. This feels like an API vs implementation issue. If you have access to the runtime, which we do not, then you can distinguish between dimensionful and dimensionless quantities.