Thu Nov 30 17:33:59 EST 2006
on august 9 i had a severe case of coder depression, thinking bad
things about pf. see where we are now with last month's spurt.
done / no longer apply / not so bad
- pre-emptive parser (separate process?) / lisp compatible?
- lock free operation in core
- well thought out serialization: message based comm + persistent state
- proper type/object system (better now)
- are floats/ints/symbols different from packets? (yes)
- write a parser in forth (how to bootstrap this?)
- don't ever use ints again to represent objects!
- don't ever use 2 type descriptions again!
- proper pairs and lisp dot syntax? (no -> requires GC)
still a problem : can't fix this
- operate without libc malloc? / malloc once?
- multimethod dispatch (how to encode types and use that?)
- what about type symbols vs type lists?
- bytecode should not be lists but arrays
- any reason why the stack should be a list?
- need call threading for proper tail recursion (push bit)
- proper exceptions (continuations, dynamic-wind)
so.. dynamic wind i think i sort of fixed. the problem shifted a bit
thanks to tasks: these need task local dynamic variables. i think
blocking task switches is a bit overkill.