On Mon, 2004-12-13 at 06:09, Michael Walter wrote:
> On 12 Dec 2004 16:33:29 +1100, skaller <skaller@users.sourceforge.net> wrote:
> > Yes, that indeed is my intention. Basically, any non-transparent
> > non-function code can be made purely functional and transparent
> > with a simple transformation, yet it doesn't by this transformation
> > get any easier to reason about the code.
> Well, actually monadic I/O makes is easier to reason about code than
> languages which allow for non-transparent side effects in functions. I
> think you were making a really good argument pro monads yourself -- in
> your C interpreter monad you can easily see what's making use of the C
> interpreter (any function "in"/lifted to the C interpreter monad), and
> what's not. Same for I/O -- you can assume that only functions
> "in"/lifted to the IO monad (simply said, every value of type IO a)
> are able to do I/O. This clearly allows easier reasoning about code.


This is a good point, and it seems to support my thesis.

What you seem to be saying is that in mixed monadic/plain code,
the abstraction boundary is plain, which helps reasoning.

So the monadic model is good, because it plainly separates
the 'purely functional' non-monadic code, from the
monadic code, which admits two distinct views -- at the lower
level it is pure, whilst the higher level interpretation
might not be -- depending both on the monad being used,
and also how it is used.

Compare with Ocaml which does not localise non-transparent
code (using references, Hashtbl, etc).

-- 
John Skaller, mailto:skaller@users.sf.net
voice: 061-2-9660-0850, 
snail: PO BOX 401 Glebe NSW 2037 Australia
Checkout the Felix programming language http://felix.sf.net