TODO
* document what works and what doesn't
* document all those functions, classes, etc., that do work.
* export all the symbols correctly.
* import trie from pangram thingie

* DECLAIM - distributed execution of common-lisp agents with intent and mobility

* move stateful datastructures to a stateful: package.
  Question: should I use separate interfaces as in the pure: package,
  or dispatch directly on objects as in usual OO style?

* have generic state-passing updaters in the style of setf.
  (updatef (lookup i m k) v) ==>  ((update-expander lookup) v i m k) ==> (insert i m k v)
  OK, but how does it compose???

   <Fare> should updatef "magically" recurse through all functions that
   have an updater defined?
   <drewc> Fare: what do you mean?
   <Fare> i.e. (updatef (car (cdr '(1 2 3)) 4) ==> (4 3) OR (1 4 3)
   <drewc> oh
   <adeht> Fare: no
   <pkhuong_> Fare: I would say yes.
   * drewc is on the fence
   * `3b votes 4
   <adeht> so (let ((x (cdr '(1 2 3)))) (updatef (car x) 4)) will give
   different results?
   * Fare wonders.
   <Fare> `3b: your answer defeats the whole purpose of updatef!
   <pkhuong_> mm.. true. That is an issue.
   <`3b> Fare: yeah, that was sort of the intent :)
   <caelan> (updatef (car (cdr '(1 2 3)) 4) => (4 3) seems mystifying
   * Fare obviously hadn't thunk it through before implementing
   <`3b> Fare: seems like it might be more useful as a binding form or
   something
   <Fare> `3b: OK. gimme a syntax for it.
   <drewc> binding? it's a monad
   <`3b> Fare: if i tried, i'd probably fall into some DSL pit trying
   to build the ability to do multiple changes at once :)
   <Fare> drewc: what is your take
   <Fare> (updatef (compose 'car 'cdr)  '(1 2 3)) 4) ==> (1 4 3) ?
   <adeht> (update (cadr '(1 2 3)) 4) ==> (1 4 3)
   <adeht> :)
   <drewc> Fare: i'm not sure really. magically composing things seems
   a little too magical to me
   <Fare> so how am I to specify that I'm updating something deep?
   <pkhuong_> Fare: nested updatef.
   <Fare> (update-accessor (x) (car (cdr x)) 1) ?
   <pkhuong_> meh. I think you really do want to update deeply, and
   even setq variables.
   <Fare> (updatef (x #((a 1) (b 2) (c 3))) (car (aref x 1)) 'd) ?
   <Fare> ==> #((a 1) (d 2)(c 3))
   <Fare> drewc: (update (car (aref (updating #((a b) (c d))) 2)) 'e)
   ==> #((a b) (c d) (e nil)) ?
   <pkhuong_> Fare: now, do you want to default to setq on variables?
   <Fare> no, no setq - the whole point is that it should be pure
   * Fare wonders about abawden's quasiquote idea - then it would be
   (update `(car (cdr ,'(1 2 3))) 4) ==> (1 4 3)
   <drewc> Fare: i kind of like the quasiquote idea
   <Fare> easiest way to recurse: go through the first arg and if it's
   a known accessor, recurse.
   <Fare> other way to recurse: find the (updating ) quasiquote-like,
   and there better be only one, and make that the thing to return
   updated.
   <Fare> give a depth argument, recursing through the first
   arguments.
   <Fare> I guess the way to decide is to rewrite a few imperative
   algorithms in functional style and see what works best...
   <pkhuong_> Fare: ime, for that, it's either allowing assignment to
   locals, or tail recursion everywhere.
   <Fare> pkhuong_, so the updatef things would typically be as
   arguments to the tail call.
   <pkhuong_> right.
   <Fare> or as something bound in a monad
   <pkhuong_> ... something like a local binding.
   <Fare> the next thing would be writing a generic updatef protocol
   for updating the slots of a struct or a class
   <Fare> which I suppose means a protocol for extracting initargs to
   copy an instance, or something.
   <Fare> though going through keywords would be... yuck
   <Fare> (let ((x '(1 2 3)) (let-update ((x (car (cdr x)) 4)) x)) ==>
   (1 4 3) ?
   <Fare> (defstruct foo a b) (updatef (foo-a (make-foo :a 1 :b 2)) 3)
   ==> #S(FOO :A 3 :B 2) ?
   <Fare> of course, recursing going through the first argument isn't
   friendly to IPS

   <Fare> updatef [...] my pure answer to setf
   <Fare> except the current version doesn't compose well enough.
   <Fare> a *real* updatef should take three+ arguments: a "zipper"
   interface, a "state" object, and one (or many) values to purely
   insert at the place(s) indicated by the zipper into the structure.
   <Fare> or maybe four. An interface, an object, a key, and values.
   <Fare> (splitting the key from the interface)