Bison --report=all evidently computes the reduction lookahead sets. If I have understood this right, then it should be possible, as an option, to make a LALR(1) parser that reports an error immediately, as in LR(1). There have been a number of questions here, especially in interactive parsers, which want to know the lookahead sets, and report entered errors immediately.
So this suggests a feature, where the parser gets an extra table with, for each state, these reduction lookahead sets. When, in a state, the $default is encountered, the parser checks if the lookahead token is in the lookahead set, and if not, immediately reports the error. This would make there to be less need for a LR(1) feature, as GLR can handle the fact that LALR(1) accepts a smaller grammar set than LR(1). Hans Aberg
