Fix for a crash in String.substring(with:) [SR-2483]

[pushkarnk]

I think this is just one way to fix: https://bugs.swift.org/browse/SR-2483
I've documented some additional findings in the bug report.

[phausler]

could you possibly add a unit test to verify this behavior?

[pushkarnk]

@phausler : I am no longer comfortable with my fix. The API reference says String.init?(_ utf16) returns nil only when the utf16 contains unpaired surrogates - so, the case of it returning nil for "\r\n"[1..<1] does not match the spec and we cannot simply return "". This needs more investigation on the stdlib part. If you don't mind, I'll keep this pull request open until we have a better understanding of how this should be fixed.

[phausler]

Certainly, it might be useful to verify the ground truth of how that behaves in objc - since that is our baseline here

[pushkarnk]

When using substring(with:) if either end of the range breaks a surrogate pair, we crash on Linux but not on Darwin. On Darwin, I found the unpaired surrogate being replaced by the Unicode replacement character. We need to do the same on Linux. Also, for ranges of the type t..<t we need to make sure we return an empty string.

@phausler @parkera : can you please review the new changes?

[pushkarnk]

@swift-ci please test

[pushkarnk]

Updated this to include a handful of corner cases with surrogate pairs and \r\n. I've made sure we behave like Darwin!

[pushkarnk]

@swift-ci please test

[pushkarnk]

After all the prior checks, I think this can be safely unwrapped.

[pushkarnk]

@phausler @parkera Sorry for the delay on this.

I've tried to make substring(with: NSRange) more robust (and behave like Darwin) here, in the process of fixing SR-2483. Can you please review? Thanks.

[parkera]

@swift-ci please test

[pushkarnk]

@parkera @phausler is there anything else you'd want me to consider here?

[parkera]

This took me a while to review because I was digging through the Darwin version, which really defers this operation to CFString. That's where the replacement character substitution happens. In the long term we may want to make NSString.swift do something similar so we don't end up replicating this logic in two places, but in the short term this fix seems appropriate.

[parkera]

@swift-ci test and merge