RepeatCall shouldn't implement DoubleEndedIterator

[Stebalien]

For example, the following iterator is definitely not reservable:

RepeatCall::new({
    let mut counter = 0;
    move || {
        counter += 1;
        counter
    }
});

[bluss]

I guess you're right about that, most implementations do not make sense as double ended. You can of course run into this problem in other places, like in a regular .iter().map() too. I'm not sure what the final decision will be. After all, .map() is double ended.

[Stebalien]

That's... odd. I've submitted a bug to discuss it (rust-lang/rust#30632). It kind of makes sense because map (and filter) are generally pure functions but this behavior should, at least, be noted.

However, this case is pretty clear; RepeatCall always produces an infinite sequence and infinite sequences only have one end.

[bluss]

I disagree actually, infinite sequences can have two ends (For example .cycle()). Maybe cycle is a bad argument, since I wrote it 😄

[bluss]

Oh no, cycle is not double ended! My evil plans,... Ok I guess I managed to scare you there by mistake.

[bluss]

On the whole, sideeffectful iterators can always cause trouble like this. I'm leaning towards wontfixing this bug for this reason, it has use cases where double endedness is useful.

[Stebalien]

Infinite sequences have one end.

-- ZFC

Yes, some sequences can go backwards from the starting point but that's not double ended, it's no-ended (doubly infinite).

[bluss]

I thought this issue could stay open until we make a decision (0.5 is next breaking change opportunity).

Currently I lean WONTFIX.

[bluss]

In a crazy turn of events, I'm removing the DEI implementation in 0.5