copy byte arrays upon string construction

[StefanKarpinski]

Strings are immutable by interface even though manipulation of their underlying data is not prevented. This violates that abstraction:

julia> bytes = linspace(uint8('A'),uint8('F'))
[65,66,67,68,69,70]

julia> str = ASCIIString(bytes)
"ABCDEF"

julia> bytes[1] += 32
[97,66,67,68,69,70]

julia> str
"aBCDEF"

Here, changing a mutable object affects the state of a string object whose interface is immutable.

[JeffBezanson]

This is extremely nasty. How can we avoid allocating two byte arrays for every string constructed?
And it is still possible to pull out str.data and mutate it, so it's not like this overhead fixes the problem completely.
I'm inclined to pretend this bug is a feature.

[StefanKarpinski]

I'd be ok with that. It basically just need to be stated in BIG RED LETTERS that if you pass a byte array to ASCIIString or UTF8String that you no longer own that array. It's gone. Belongs to the string.

In general, I feel like the private vs. public contract in Julia is this: an object's state is private — if you access it directly and change it, you had better know what you're doing; if you just manipulate an object with functions that are provided to manipulate it, then that's safe because those are its public interface. So the fact that str's state can be mutated "illegally" by accessing and changing str.data doesn't really bother me.

[JeffBezanson]

Fine. Close it!

[StefanKarpinski]

We could also make copying and non-copying versions with the default being copying.