alternative module systems

alternative module systems

[Markus Mottl]

Hello,

having just looked at a few examples of the new module system as
implemented by Claudio Russo in Moscow ML, I wonder whether people at INRIA
have already considered it?

It supports higher order functors, mutually recursive modules, even first
class modules. The module examples in the Moscow ML distribution also
demonstrate the new capabilites using the bootstrapping methods explained
in Okasaki's book on "Purely Functional Datastructures". This is really
neat stuff!

I have no idea whether there are any caveats to this solution, but it looks
pretty general. Any comments whether something similar could be implemented
in OCaml? As it seems, the solution is a true superset of the previous
module system used in SML.

Best regards,
Markus Mottl

-- 
Markus Mottl, mottl@miss.wu-wien.ac.at, http://miss.wu-wien.ac.at/~mottl

Re: alternative module systems

[Brian Rogoff]

On Mon, 14 Aug 2000, Markus Mottl wrote:
> Hello,
> 
> having just looked at a few examples of the new module system as
> implemented by Claudio Russo in Moscow ML, I wonder whether people at INRIA
> have already considered it?
> 
> It supports higher order functors, mutually recursive modules, even first
> class modules. The module examples in the Moscow ML distribution also
> demonstrate the new capabilites using the bootstrapping methods explained
> in Okasaki's book on "Purely Functional Datastructures". This is really
> neat stuff!

Yes it is. I wonder what else those features will enable. In particular,
the use of first class structures to configure a system at run
time looks really useful to me. This is one of those things I'd normally
be disposed towards doing with classes. I find it disturbing though, since
my mental model of modules is that they are "second class". 

Also, there is the age old (in web years :) desire to have mutually
recursive types which span module boundaries, and it is possible in this 
system. I guess there have been papers describing other solutions, but 
a having a real implementation is a win.

> I have no idea whether there are any caveats to this solution, but it looks
> pretty general. Any comments whether something similar could be implemented
> in OCaml? As it seems, the solution is a true superset of the previous
> module system used in SML.

It's a superset of the OCaml module system too, at least that part that is 
comparable with SML (obviously SML doesn't have classes and such). 

-- Brian

Re: alternative module systems

[Markus Mottl]

> Yes it is. I wonder what else those features will enable. In particular,
> the use of first class structures to configure a system at run
> time looks really useful to me. This is one of those things I'd normally
> be disposed towards doing with classes. I find it disturbing though, since
> my mental model of modules is that they are "second class". 

It also took me a while to wind my brain around this concept. Given that
modules have some nice properties that do not hold for classes (objects),
being able to create them at runtime and even pass them around as values
looks quite promising. To my knowledge, the described approach is said to
be formally very clean.

> It's a superset of the OCaml module system too, at least that part that is 
> comparable with SML (obviously SML doesn't have classes and such). 

Indeed, it looks as if one could switch "below the hud" without breaking
code, doesn't it?

Best regards,
Markus Mottl

-- 
Markus Mottl, mottl@miss.wu-wien.ac.at, http://miss.wu-wien.ac.at/~mottl