On Wed, Nov 14, 2001 at 07:33:06AM -0700, Jay Kominek wrote:
Obviously {mi pu ca'a broda} does work. So the key difference is that
'pu' cannot follow 'ca'a' within a single time construction.
I've investigated why jbofi'e doesn't reduce ca'a to a standalone term
(inferring 'ku') when 'pu' is the lookahead token. Essentially, it's
because it thinks it's starting to see something like
mi ca'a pu gi le nanmu gi le ninmu cu broda
(whatever that might mean), i.e. it's expecting to see 'gi' after pu to
make everything work out OK. This expression parses OK on the official
v2.33 parser. (I don't actually have the official v3 parser running yet
- I've never got round to sorting the build out.)
I need to away and think quite hard about this problem. The full extent
of the bug is not yet clear to me - I presume I'm missing one of the
pre-parser stages that the official parser has.