@markusicu, friendly ping (#3604, the UnicodeSet parser rewrite, is waiting behind this one).

As for ^, - is treated normally at the beginning of a UnicodeSet and as a set elsewhere.

I didn't understand "as a set elsewhere". Did you mean is treated as syntax to form a range elsewhere? Eg [a-b]?
BTW I think we only had the "is normal at the start" to match regexes.

I didn't understand "as a set elsewhere". Did you mean is treated as syntax to form a range elsewhere? Eg [a-b]?

The context here is that the symbol table can map characters to sets, e.g., it can say “0 means [a-z]”. And this can be done to syntax characters, e.g., “- means [1-9]”. The effect of mapping syntax characters varies, and this is what this pull request tests. Obviously this is madness and we should disallow the mapping of syntax characters in a future version of ICU, but for now I am testing exactly what madness happens so I can preserve it in my rewrite of the parser.