This is clever, and extended my understanding of the RE engine (++), but is it guaranteed to work?

I got interested in why a negative look-ahead was required, and found that negative and positive failing look-behinds work too, but a simple mis-match doesn't, and neither does a failing zero-length positive look-ahead: (?=x). For example, m/(aa)(?{push @a, $1})x/ does not work. Presumably the regex optimiser sees that there is no 'x' in 'aaaa', so it doesn't bother with the step-wise attempts to match the 'a's.

Is it possible a future regex engine will realise that mis-match is inevitable because (?!) will always mis-match, and break this code?