Range::step_by(0) loops forever

[llogiq]

I believe that the step_by(_) method of ranges should panic! if the argument is zero. Currently it loops forever, which is a footgun. At the very least, iterating over such a StepBy should panic! on debug builds. However, I think the earlier the error appears, the better.

Note that this may require us to specialize step_by to types whose Add implementation have a known identity element – at the moment that would mean bounding by the num::Zero trait, which is unstable, also it would probably reduce the usefulness of ranges to numeric types. However, special casing numeric ranges could solve this.

I have also filed an issue at rust-clippy to create a lint for cases where it can be statically detected. Still, that lint cannot detect cases where the argument is calculated at runtime.

The documentation of .step_by(_) should at least warn of the possibility of looping forever or panic once it is implemented.

[Thiez]

There is nothing unsafe about looping forever, so I strongly oppose a panic!(). As a user I would expect (a..b).step_by(c).take(n) to work for all c, without having to special case for c == 0..

[llogiq]

@Thiez: Agreed, this is not inherently unsafe. So maybe panic!() on building the StepBy is not the right option. Still, there are enough options to loop forever already, and as I said, the docs should at least remind the reader that supplying 0 as an argument creates an infinite range.

It may also be useful to be able to issue a warning on debug builds iff a `StepBy { step_by: 0, .. } is actually iterated; perhaps as an additional Debug option.

[alexcrichton]

This is currently intended behavior, and for now I'm going to close this in favor of the discussion in stabilizing step_by as this is definitely a useful data point, but perhaps not big enough to have its own issue right now.