You learn something every day. I would have assumed a < b < c would have compared the result of a < b (i.e. 0 or 1 for true or false) with c. (I have a tendency to assume left-associativity except with what I know to be right-associative.) It's interesting to see which operators perlop says can't be chained like that (the ones listed as nonassociative).

Seems a little suboptimal to handle it with yacc, rather than something that could give a better error message.