Add IteratorSize for Char

[bkamins]

See discussion in https://discourse.julialang.org/t/broadcasting-and-single-characters/16836.
I am not sure if the current behavior is indented, or what I propose in this PR is intended.

This can be classified as breaking change or bug fix - I am not sure what is the policy for such PRs after Julia 1.0 release.

[bkamins]

CI fails in one case and it seems unrelated (but I am not 100% sure, as the place of the error is strange and it should not happen).

[StefanKarpinski]

Seems reasonable to me. Does anyone with strong opinions about broadcasting want to chime in? @mbauman, @JeffBezanson?

[nalimilan]

+1 for fixing broadcasting, but maybe Chars should remain HasLength (like strings and symbols) rather than HasShape{0} (like numbers)? AFAICT making them HasShape{0} would change the result of [x for x in 'a'] from a one-element vector to a zero-dimensional array. If you only define broadcastable(c::Char) = c you should fix broadcasting without changing comprehensions.

[bkamins]

Agreed this is a change - that is why I have said that I am not sure what is the intended behavior.

I understood that the general design intention was that [x for x in 'a'] and [x for x in 1] should behave the same (both producing 0-dimensional array).

Also I think that we should have that identity.('a') and [x for x in 'a'] produce the same thing (what would be the reason for a different behavior)?

[StefanKarpinski]

I really don’t want to make characters iterable. We wanted to make numbers non-iterable but it was too disruptive so let’s not make another scalar type iterable for no reason.

[KristofferC]

But they already are?

[bkamins]

My understanding is that "we are", but only half-way. This PR makes Char consistent with numbers.

I would also prefer Char to be treated purely as a scalar, but this would be even more breaking.

[mbauman]

Given that Chars are already iterable and already define size(c::AbstractChar) = () and axes similarly, adding this trait simply brings everything into sync.

[StefanKarpinski]

Sigh.

[bkamins]

Any decision on this change?

[bkamins]

@KristofferC Shall I make a PR to news, or this is now done in some other way?

[StefanKarpinski]

Yes, PR to add an entry to that file would be perfect. Thank you!