String constructor is bad

[TotalVerb]

The String constructor here leaves a lot to be desired:

"""
    String(v::Vector{UInt8})

Wrap a vector of bytes encoding string data as UTF-8 in a `String` object.
The resulting `String` object takes ownership of the array.
"""
String(s::Vector{UInt8}) =
    ccall(:jl_pchar_to_string, Ref{String}, (Ptr{UInt8},Int), s, length(s))

• It makes a copy, so it's very slow. At least in Julia 0.3, no copy was made.
• The behaviour is different from the docstring. The String object doesn't take ownership of anything.
• It's unsafe. String([0x80]); is allowed, even though is is an invalid string.
• It overwrites the String inner constructor, making the inner constructor useless and emitting a warning at build time.

Obviously not all of these can be fixed (points 1 and 3 would move it in different directions), but I think it ought to pick a direction:

• String should be safe, and check its input (along with making a copy). The method is updated to check validity of UTF8. The docstring is updated. An unsafe_array_to_string replaces the current behaviour, and additionally does not make a copy. The inner constructor is deleted.
• String should not be safe, and does not check input, and does not make a copy. The method is deleted and the docstring moved to the inner constructor.

[ararslan]

Ref #16107 (general direction for String) and #16590 (recent developments)

[stevengj]

I think there was a conscious decision by @StefanKarpinski to allow invalid UTF-8 data, so that String containers could be used to pass through arbitrary binary data. This is common to many UTF-8 implementations, including Python's. (The decoding algorithm to extract chars/codepoints must still check for invalid data, of course. There have been a lot of proposals for how to decode such data.)

[stevengj]

See also #16470.

[stevengj]

I think @StefanKarpinski may have accidentally changed the String(::Vector{UInt8}) semantics in #16453. When he renamed the bytestring(::Vector{UInt8}) function to String(::Vector{UInt8}), he overrode the inner String constructor (which doesn't make a copy) from #16212.

Probably bytestring(data::Vector{UInt8}) should have been deprecated to String(copy(data)).