Support for LR(k) grammars.
Rationale: Many languages are hard to express with LR(1) grammars. For instance grammar for many modern programming languages can be written easily and naturally using an LR(k) grammar, for more than 1 token of lookup. Of course the set of accepted languages of LR(1) is the same as of LR(k). But still, the 'natural' LR(k) grammar has to be converted by hand into equivalent LR(1) form. The resulting LR(1) grammar is often very hard to write, read, and maintain.
I request a feature where Menhir could use LR(k) grammars as input. Then either the parsing algorithm would need to be modified to do farther lookups, or Menhir could rewrite the grammar automatically to LR(1), parse as it does today, and transform the resulting parse tree back to the form corresponding to the specified LR(k) grammar.