startswith and endswith look suspicous

[eschnett]

The definitions for startswith and endswith for strings look suspicious (base/string/util.jl):

startswith(str::AbstractString, chars::Chars) = !isempty(str) && str[start(str)] in chars
endswith(str::AbstractString, chars::Chars) = !isempty(str) && str[end] in chars

1. Why start(str)? Isn't this always 1?
2. Why end? Shouldn't this be endof(str) instead?

[yuyichao]

end is lowered to endof

[nalimilan]

...but indeed 1 and start(str) are equivalent for strings, and I think for all AbstractArrays. Not a big deal though.

[simonster]

If we're going for maximum generality it should be first(str) and last(str). start is definitely wrong since a custom string type could have some other way of defining iteration. I also find the behavior of accepting a Set{Char} or AbstractVector{Char} a bit questionable, since startswith doesn't accept a Vector{T<:AbstractString} or Set{T<:AbstractString}.

[yuyichao]

@simonster I made that mistake some time ago too. first and last returns the element not the index. endsof is the right way to get the last index and IIRC we don't have a generic way to get the start index and for most places we just use 1.

[simonster]

We don't need the index here, just the first and last characters.

[yuyichao]

Ahh. That's right

[stevengj]

I agree that first and last are the right choice, although the current implementation should work for our current string types even when the last character has a multi-unit encoding.