You have spotted a problem with ML-style type systems - they are not great
at encoding the kind of "phasing" of initialization that you refer to
(neither most are other type systems, though of course one could imagine a
type system that allowed you to express this sort of thing).  However, the
"constants" you refer to can encode quite complex computations (including
allocation of data structures, e.g. let x = [2;3;5]]), which is a big
advantage compared to, say, C, but it does not cover everything.  So in
practice people tend to make do with constant values until the
initialization sequence gets too complicated, then move to references, but
with the knowledge that these references are set once-for-all.  Hiding the
references behind functions is, as you say, often a good idea.

Basically, don't be afraid of using references for this kind of thing - it's
one of the reasons they are in the language.

Cheers,
Don


-----Original Message-----
From: David.Mentre@irisa.fr [mailto:David.Mentre@irisa.fr]
Sent: 20 June 2000 17:00
To: caml-list@inria.fr
Subject: How to cleanly encode "quasi-constants"?


Hello dear camlists,

I'm still learning caml and functionnal programming, so this is maybe a
newbie question. Please don't be too rude.

In my program, I have variables that can only be modified once
(potentially at initialization time after parsing command line and done
some computation) and then are never modified for a very long time (rest
of program execution). I call them "quasi-constants", they have a
write-once semantic. They are used throught the program so it is very
painfull to propagate them along function calls (i.e. these variables
need a global scope, at least from a programmer point of view).

Right now, I'm using constants (let var = ...) or mutable references
(let var := ...) to encode such things.

However I'm not satisfied. With constants, I obviously can only modify
them at compile time. With mutables, I can modify them but potentially
at any time, meaning potential bugs. So my question is: 

 What is the Right Way(tm) to encode such a thing in an ML like
 language and more specifically in OCaml?

I was thinking of using a functor (return a module with functions
referencing these quasi-constants and quasi-constants values given at
module creation). Is it right? It is also possible to use functions,
however is seems impracticable for a real program (this approach creates
a function many pages long):

# let create_manipulator v =
  let f1 () = v 
  and f2 x = v+x in (f1, f2);;
  val create_manipulator : int -> (unit -> int) * (int -> int) = <fun>
# let (g1, g2) = create_manipulator 3;;
val g1 : unit -> int = <fun>
val g2 : int -> int = <fun>
# g1 ();;
- : int = 3


Two side questions:

 - is there such a problem tackled in a big program ? (ocaml compiler?)

 - it seems to me that Arg.Set and Arg.Clear types have this semantic
   (an option is only set at parse time).



Best regards,
david
-- 
 David.Mentre@irisa.fr -- http://www.irisa.fr/prive/dmentre/
 Opinions expressed here are only mine.