you're right. inclusion/intersection/etc. aren't actually computed via DFA but instead are computed directly on the regular expression representation itself. and large disjunctions (with 256 branches) are what is very heavy.
(it's possible to instead do these operations on DFAs but at the time i found it hard to get from an automata back to a reasonable-looking regular expression.)
(it's possible to instead do these operations on DFAs but at the time i found it hard to get from an automata back to a reasonable-looking regular expression.)