Well, I’ve committed the new integrated lexer/parser to Subversion.  The state generation time has increased between 20 and 50% over the previous version (depending on which runs you compare), but the state tables do look correct for little things like gateway expressions and discard processing, so I’ll live with it.  And I’m pretty sure I can find a few more optimizations in there, too.

Next up: integrating the new actions and policies into the interpreter.  That should be straight forward.  Then: fixing up error recovery to use the new system.  That probably won’t be.  :-(  However, with that done, it will be down to code generation.  Yeah, I know — I’ve thought that before.  Still, things are looking up.