SubString{UTF8String} and UTF8String have different slicing behaviour

[ivarne]

Found this on stackoverflow

julia> a = "æøå"
"æøå"

julia> a[2:end]
"øå"

julia> b = SubString(a,1)
"æøå"

julia> b[2:end]
ERROR: invalid SubString indexes
 in getindex at ./string.jl:657

I'm not sure what the correct behaviour should be, but having them different seems wrong. Some relevant discussion in #5446

[stevengj]

a[2:end] is wrong, because 2 is not the index of the beginning of a UTF8 character; you really want a[3:end] == a[chr2ind(a,2):end].

[ivarne]

I know the index 2 is invalid. The problem is that the behavior is different for the erroneous index, and I can't see why they should be.

[JeffBezanson]

Range indexing for UTF8String adjusts the start index to the next valid index, while SubString just throws an error in this case. The "adjusting" behavior is probably not necessary.

[ivarne]

Changing range indexing for UTF8String (much better word than slicing, by the way) is a breaking change, that will cause problems. Added appropriate labels. Seems like this should wait for 0.4

[stevengj]

I suspect that most usages that rely on the auto-adjustment are actually bugs (people thinking that character index == string index). So I'm fine with UTF8String throwing an error instead.

[catawbasam]

+1 for throwing the error. The auto-adjustment makes it harder to figure out how UTF8String actually works.