Path: utzoo!attcan!uunet!ginosko!uakari.primate.wisc.edu!ames!uhccux!munnari.oz.au!cs.mu.oz.au!ok
From: ok@cs.mu.oz.au (Richard O'Keefe)
Newsgroups: comp.lang.prolog
Subject: Re: hierarchy among modules in prolog
Keywords: modules, hierarchy, Quintus prolog
Message-ID: <2326@munnari.oz.au>
Date: 8 Oct 89 07:48:14 GMT
References: <5641@portia.Stanford.EDU>
Sender: news@cs.mu.oz.au
Lines: 112

In article <5641@portia.Stanford.EDU>, sreerang@portia.Stanford.EDU (Sreeranga Rajan) writes:
> I would like to know if it is possible to make the visibility of
> assertions in modules hierarchical in Prolog.

Please say more.  What is it that you want to do?
There are two separate questions:  the structure of the space of module
names, and the set of visibility graphs you can construct.


> The Prolog I am using being Quintus Prolog running on SunOS,
> seems to have a flat module system.

The space of module names in Quintus Prolog is indeed flat:
there is a one-to-one correspondence between modules and atoms.
However the space of module names were structured, there would
be *some* "flat" set which could be placed in correspondence with it.

As for the visibility graphs you can construct in Quintus Prolog,
think of modules (M) and predicates (M:P/N) as being nodes in a
bipartite graph, where there is an arc from a predicate (M:P/N)
to the module (M) it is defined in and an arc from a module (M)
to every predicate (M':P'/N') which is visible in M.  Any such
graph is possible.

Suppose that you want to have some sort of tree-structured scheme,
where a module may have other modules as "children", and everything
that is visible in a parent is visible in all its children.

It is already the case that a module in Quintus Prolog may re-export
something that it has imported from another module, so there would
be no problem with having the parent export something which is
actually defined in a child.  The trick is to get the children to
see the parent's private predicates.

Actually, there isn't any trick.  A module can forcibly import something
from another module even if the other module didn't export it.  You do
get a warning message.  So the children could simply import everything
they need from the parent.  We would have

% File parent.pl	% File child1.pl	% File child2.pl
:- module(parent, [	:- module(child1, [	:- module(child2, [
	mine/1,			onea/1,			twoa/1,
	onea/1,			oneb/2			twob/2
	twob/2		   ]).			   ]).
   ]).
:- use_module(child1).	:- use_module(parent, [	:- use_module(parent, [
:- use_module(child2).		needed/1,		needed/1,
				vital/3			useful/2
			   ]).			   ]).

define mine/1		define onea/1		define twoa/1
   and needed/1		   and oneb/2		   and twob/2
   and useful/2
   and vital/3

Yes, there is a cycle here: parent uses child1 uses parent.  But that's
just fine.  use_module is like ensure_loaded; both are just as happy as
can be with usage cycles, and do exactly the right thing.

Just because there isn't a hierarchy in the NAMES of the modules
doesn't mean there can't be a hierarchy in the CONTENTS!

Suppose that you really don't want to get these error messages.
You can restructure the parent module as two modules.

	% File parent.pl
	:- module(parent, [
		mine/1,
		onea/1,
		twob/2
	   ]).
	:- use_module(local,  [mine/1]),
	   use_module(child1, [onea/1]),
	   use_module(child2, [twob/2]).
	end_of_file.

	% File local.pl
	:- module(parent, [
		mine/1,
		needed/1,
		useful/2,
		vital/3
	   ]).
	/* former body of parent.pl goes here */

The children are then modified to use the 'local' module rather than
the 'parent' module.

Now it really doesn't take a lot of ingenuity to write a little
preprocessor that will generate these headers automatically.  In fact,
the headers and the clauses don't have to be in the same files.
For example, child1.pl could look like

	% File child1.pl
	
	:- module(child1, [
		onea/1,
		oneb/2
	   ]).
	:- use_module(local, [
		needed/1,
		vital/3
	   ]).
	:- compile('child1.body').	% no :-module header in that!
	end_of_file.

So your preprocessor doesn't need to alter any files containing declarations
or rules, it can just generate new header files.

"Hierarchical" could mean almost anything, so there's a good chance that
the sketch above doesn't address your problem.  Tell us more about what
you're trying to do.