... and no, I don't know of any fast Thompson NFA simulations, but I don't see why they shouldn't be possible. They have a very simple "next" function, modulo t

... and no, I don't know of any fast Thompson NFA simulations, but I don't see why they shouldn't be possible. They have a very simple "next" function, modulo the awfulness of getting past epsilons, but that seems to be roughly parallel to the awfulness of computing arbitrary 'next' functions in Glushkov-land. I'm not aware of anyone that's actually tried.