Throw RangeError on bad numeric arguments to take/drop?

[benjamingr]

Currently, this proposal coerces arguments passed to drop and take to integers without range checks other than for negative numbers because ToIntegerOrInfinity is used to coerce the argument passed in.

This has the potential to create confusing behavior - for example naturals().take(NaN) and naturals().take((x) => x < 15) and naturals().take('aaa') all return an empty iterable.

I think most other places the spec does this get away with it by checking an explicit range (e.g. toString on numbers requiring the radix being at least 2) but admittedly some places don't (e.g. toExponential happily accepting NaNs).

I think it would provide better developer experience to get RangeErrors on NaNs (basically ToNumber the argument first and throw a RangeError on NaNs).

[bakkot]

Most places in the spec don't guard - .slice(NaN) will work, for example, and it's the closest thing to drop and take currently in the language. It's a pretty strong precedent. So this amounts to, do we think it's worth breaking with that precedent?

I think there's a pretty good case to be made for checking for NaN in particular - I don't think there's any good reason to want to treat NaN as zero; it's pretty much always a bug, and I think it's better for bugs to be loud (though much of the design of the language is not consistent with that principle). And this is a new kind of thing, so the consistency argument is less strong than it is for e.g. new methods on Array.

I'd be in favor of throwing when the argument coerces to NaN.

[bathos]

I don't think there's any good reason to want to treat NaN as zero; it's pretty much always a bug

NaN as direct input, probably, but not as undefined. Pretty much everything that accepts an offset, index, or length allows omission/undefined because of ToIntegerOrInfinity: undefined -> NaN -> 0. So it's not just that slice(NaN) works, it's also why slice() works.

[bakkot]

NaN as direct input, probably, but not as undefined

Passing undefined here (or omitting the argument) is also pretty much always a bug.