CC @gribozavr

This was working in some of the developer snapshots in February-March-ish. So we fixed it then regressed.

The replaceSubrange() call replaces elements before the index that you saved. This invalidates the index, and using that index afterwards is a precondition violation (whether the program will trap or not is unspecified).

@gribozavr what would you suggest then for keeping a pointer to a position in a character view across a mutation? Like I said, this used to work in a Dev snapshot. I'm worried I'm going to have to rewrite my code to linearly traverse the string for every position in every marker every time there's a mutation.

It seems to be an actual problem with swift's String model that we can't actuallly do that, even though I know a valid location prior to mutation and exactly how it will be mutated. We have all of the information to make this possible, but the standard library doesn't expose it.

I think we would need a version of replaceSubrange which returns a new type which I'll call IndexDisplacement. It would know how many internal UTF-16 codepoints have been inserted/removed and where (for a single replaceSubrange operation), and provide a way to translate indexes back and forth. It would allow us to perform this vital operation which we can't today, and with maximum performance.