a good efficency hack for fgrep is to use a jump table indexed by the input character for the root node in the trie of keywords. I know steve tjiang did this at Bell Labs in 1984; does anyone know of any other people who did this or similar? please mail research!andrew or andrew at research.att.com