Tue Dec 5 10:59:36 EST 2006
yeah... reasons are probably external, but i had a bit of a bad day
yesterday doing the stuff described above. reflecting on that a bit, i
can see the following problems: PF does not use linear memory for
threaded code. this gives problems when you have to look
back/forward. nothing that can't be solved, and i think writing the
interpreter such that there's a single atom per functional unit should
fix it (this means, encoding instructions in the type descriptor
bits). disentangling this however is quite a job, since forth.c is
quite a mess.
i must say that after a couple of weeks of fixing deep problems, i'm
getting overconfident. but, i'm starting to believe again in PF being
a solid base for other experimental work. i managed not to break any
important things in the stream/scheduler rewrite, which is a good
thing. gives me hope to iron out some more conceptual inconsistencies.