Pseudo-regular parsing combined with hand-written pratt-parsers is a nice sweet-spot: easy to read, very robust, and produce great error messages. The technique

Pseudo-regular parsing combined with hand-written pratt-parsers is a nice sweet-spot: easy to read, very robust, and produce great error messages. The technique was, I am told, moderately popular in the 80s and 90s. (My familiarity with them comes from a language called Spad.)