Open Radar

 
Summary:
It is expected that String.canBeConverted returning true should always correspond to [NSString dataUsingEncoding:] returning non-nil, however there are cases where this is not true. If you construct a UTF-16 string with only a first surrogate character and try to convert it to UTF-8, String.canBeConverted will be true, but -dataUsingEncoding: will be nil. This is either a bug in one or the other method (and possibly also in the method that allowed this string to be generated; I'm not certain if it's valid).

Furthermore, string.data(using:.utf8) is nil when string.utf8 is not. This is surprising at least.

Steps to Reproduce:
        const unsigned char bytes[] = {0xDC, 0};
        NSData *data = [NSData dataWithBytes:bytes length:2];
        NSString *string = [[NSString alloc] initWithData:data encoding: NSUTF16StringEncoding];
        NSLog(@"%@", string); // \udc00
        NSLog(@"%d", [string canBeConvertedToEncoding:NSUTF8StringEncoding]); // 1
        NSData *utf8 = [string dataUsingEncoding:NSUTF8StringEncoding];
        NSLog(@"%@", utf8); // (null)
        NSLog(@"%s", string.UTF8String); // ""
        NSLog(@"%d", string.UTF8String == nil); // 1

Note that the Swift version of this is also incorrect, but different:

let lowSurrogateFirst = Data([0xDC, 0])
let string = String(bytes: lowSurrogateFirst, encoding: .utf16BigEndian)! // replacement-character
string.canBeConverted(to: .utf8) // true
string.data(using: .utf8) // nil
string.utf8 // replacement character
Data(string.utf8) as NSData // <efbfbd> (replacement character)


Expected Results:
At a minimum, canBeConverted should always be false if data(using:) will return to be nil.

It's unclear whether NSString or String is more correct in how they accept this string of bytes. Since this sequence is not valid UTF-16, I would expect init to fail rather than injecting a replacement character.