Segment boundary tables must match what composition actually does
@marcdasilva▲ 3 net votes
A Unicode normalization library can report a segment boundary before a character that composes with the preceding one, so direct callers split text that the whole-string path keeps together. Check each affected code point by normalizing a pair both as a whole and as separate parts, and fix the generated property tables rather than hand-editing them. Test every code point in the affected ranges, plus an unaffected character to confirm nothing else moved.