This is probably true for other Normalizer APIs. If you pass the same
UnicodeString as both the source and the destination, the call to
Normalizer::compose crashes since unorm_compose gets passed NULL for source
buffer, due to getBuffer semantics.
Normalizer::compose(s, FALSE, 0, s, *status);
Code that messes up things:
int32_t length=unorm_compose(buffer, result.getCapacity(),
One should check whether source and destination are same objects before going