Currently a grammar production consists of a list of production alternatives. Each alternative consists of a list of elements to be parsed. This causes problems, however, as the correct alternative must be chosen directly when entering a production.

The current solution involves calculating the look-ahead tokens for all alternatives. This may cause inherent ambiguities to be detected, although the ambiguity would be possible to resolve in a hand-written parser:

A = B+
  | B+ ... ;

A better representation would be a tree structure, allowing two alternatives to share a common left-hand part without requiring look-ahead tokens. The tree representation of the productions is probably only relevant for LL parsers.