Make String#[] not read out-of-bounds if string ends in unicode char

[Papierkorb]

Hello,

While working on fancyline, I noticed that String#[] didn't do proper bound checking,
but relied on Char::Reader for this instead. Char::Reader didn't do bound-checking
either (At all, actually), creating an out-of-bounds memory read.

Sample code

"ß"[1] # => Char(0), expected an IndexError
"ß"[1]? # => Char(0), expected `nil`

# And this does an unchecked memory read:
Char::Reader.new("", pos: 100).current_char
# ^ now always gives `Char(0)`.  Changing it to just raise breaks a ton
# of assumptions made in String.

Both showcased behaviours are fixed by this PR for now.

[straight-shoota]

Is this behaviour really unexpected? A string is delimited by a zero character: "ß" describes a string containing the characters 'ß' and '\0'. So, the character with index 1 is obviously Char::ZERO.

If you use an index that is actually out of bounds (like "b"[2]) this will raise an IndexError (or return nil). There is a out-of-bounds check, you're just off by one.

To me, everything seems to be just as it is supposed to and I don't think this needs a change.

[Papierkorb]

1. The trailing NUL-byte is an implementation detail.
2. container[container.size] is out of bounds for any container type, why should String violate this?
3. "a"[1] is OOB, as expected

This is a bug with unicode handling.

[straight-shoota]

From your description it was not clear that multibyte characters behave differently than ASCII characters. This should obviously be fixed.

Btw. Char::Reader yields \0 and this is part of a public API.

[Papierkorb]

Bump

I'd love to see this being merged, at least so that 0.24 will benefit. It can have security implications. This fix also fixes real-world code on my end.