I do make some modifications:

• instead of convert re into postfix representation, I use a recursive descend parser.
• wrap all temporary storages into a structure, so I can free resources later.

• russ_nfa.c NFA version
• dfa0.c creating DFA state on-the-fly
• dfa1.c use a bounded memory for caching DFA states
• nfa-posix.y use a reversed NFA, and backward matching

plus, nfa-posix.y has a bug in yylex function. + should be considered as a metacharacter.