Make `rem a 0` throw an error (just like `mod a 0` resp. `a % 0` does)

[jvoigtlaender]

This fixes https://github.com/elm-lang/core/issues/563. It is in line with the treatment of %/mod as @evancz did here: elm/compiler#262 (comment).

[rtfeldman]

I think #262 is actually outdated; both NaN and Infinity are most definitely a thing in Elm at present:

http://package.elm-lang.org/packages/elm-lang/core/3.0.0/Basics#isNaN
http://package.elm-lang.org/packages/elm-lang/core/3.0.0/Basics#isInfinite

I believe the thinking on NaN and Infinity is "they are part of the IEEE floating point number spec, and adding checks to diverge from that spec has performance costs for operations that get used all over the place, so let's not make that the default."

Pretty sure it would make more sense to remove the check from mod, if anything. 😉

[rtfeldman]

I think #262 is actually outdated; both NaN and Infinity are most definitely a thing in Elm at present:

http://package.elm-lang.org/packages/elm-lang/core/3.0.0/Basics#isNaN
http://package.elm-lang.org/packages/elm-lang/core/3.0.0/Basics#isInfinite

I believe the thinking on NaN and Infinity is "they are part of the IEEE floating point number spec, and adding checks to diverge from that spec has performance costs for operations that get used all over the place, so let's not make that the default."

Pretty sure it would make more sense to remove the check from mod, if anything. 😉

[jvoigtlaender]

@rtfeldman, all the arguments you give apply to Float, and only to Float. Including the arguments about the existence of isNaN and isInfinite. Both of these have type Float -> Bool. But #262 was about Int, not Float. And both mod and rem are operations on Int. So I don't really see how your reasoning applies.

But in any case, I agree that most importantly mod and rem should behave consistently. Whether that is by both allowing division by zero or both not allowing it, well, ...

[jvoigtlaender]

@rtfeldman, all the arguments you give apply to Float, and only to Float. Including the arguments about the existence of isNaN and isInfinite. Both of these have type Float -> Bool. But #262 was about Int, not Float. And both mod and rem are operations on Int. So I don't really see how your reasoning applies.

But in any case, I agree that most importantly mod and rem should behave consistently. Whether that is by both allowing division by zero or both not allowing it, well, ...

[jvoigtlaender]

But if rem a 0 stays allowed and a % 0 becomes allowed, and then naturally a // 0 as well, then isNaN and isInfinite should be adapted to have type number -> Bool rather than Float -> Bool, because they should become applicable to Int as well.

[jvoigtlaender]

But if rem a 0 stays allowed and a % 0 becomes allowed, and then naturally a // 0 as well, then isNaN and isInfinite should be adapted to have type number -> Bool rather than Float -> Bool, because they should become applicable to Int as well.

[jvoigtlaender]

Closing this for the more comprehensive https://github.com/elm-lang/core/issues/590.

@rtfeldman, what do you think about that new issue?

[jvoigtlaender]

Closing this for the more comprehensive https://github.com/elm-lang/core/issues/590.

@rtfeldman, what do you think about that new issue?