[m-dev.] for review: polymorphic ground insts

[David Overton]

On Tue, Oct 10, 2000 at 10:43:59PM +1100, Fergus Henderson wrote:
> On 29-Sep-2000, David Overton <dmo at cs.mu.OZ.AU> wrote:
> > It's been a long time, but I've finally got back to this and got it
> > working.  I've fixed the problem with existential types
> 
> Well, I'm not really convinced that this change handles existential
> types consistently with the way it handles non-existential types.
> But if it is only the corner cases that it mishandles, and these cases
> occur rarely or not at all in practice, and the symptom is just that
> the mode checker rejects the program, and the code in question (in
> maybe_get_cons_id_arg_types in modes.m) is marked with an XXX, then I
> think it is OK to commit it.
> 
> The situations I'm worried about are things like
> 
> 	:- type t ---> some [T] f(T).
> 	:- inst t ---> f(ground).
> 
> 	:- pred p(t::in, t::out(t)) is det.
> 	p(X, X).
> 
> Is this allowed?

This case will work fine.

> What about if I change the definition of type `t' to
> 
> 	:- type t ---> f(int).
> 
> ?

This case also works.

Note: neither of these examples work in the current compiler.

The problematic case would be if you change the definition of the inst
`t' to something like

	:- inst t ---> f(bound(....)).

There is no way of checking whether this bound ints contains all the
functors for the existential type T.

> 
> Also, I found it confusing that the name maybe_get_cons_id_arg_types
> was used for two different procedures (/4 in modes.m, /5 in type_util.m).
> I think it would be a good idea to rename one of them.

I've renamed the one in type_util.m to `get_cons_id_non_existential_arg_types'.

> 
> > compiler/type_util.m:
> > 	Add a new predicate, type_util__cons_id_arg_types which
> > 	nondeterministically returns the cons_ids and argument types for a
> > 	given type.
> 
> The log message there doesn't describe all the changes.

I'll add this sentence:

	Add a new predicate type_util__get_consid_non_existential_arg_types
	which is the same as type_util__get_existential_arg_types except
	that it fails rather than aborting for existenially types arguments.

> 
> > @@ -814,20 +850,24 @@
> >  	).
> >  
> >  type_util__cons_id_arg_types(ModuleInfo, VarType, ConsId, ArgTypes) :-
> > +	type_to_type_id(VarType, TypeId, TypeArgs),
> > +	module_info_types(ModuleInfo, Types),
> > +	map__lookup(Types, TypeId, TypeDefn),
> > +	hlds_data__get_type_defn_body(TypeDefn, TypeDefnBody),
> > +	TypeDefnBody = du_type(_, ConsTags, _, _),
> > +	map__member(ConsTags, ConsId, _),
> > +	
> >  	module_info_ctors(ModuleInfo, Ctors),
> > -
> > -	map__member(Ctors, ConsId, ConsDefns),
> > +	map__lookup(Ctors, ConsId, ConsDefns),
> >  	list__member(ConsDefn, ConsDefns),
> >  	
> > -	type_to_type_id(VarType, TypeId, TypeArgs),
> >  	ConsDefn = hlds_cons_defn(ExistQVars0, _, Args, TypeId, _),
> > -	module_info_types(ModuleInfo, Types),
> > -	map__lookup(Types, TypeId, TypeDefn),
> > -	hlds_data__get_type_defn_tparams(TypeDefn, TypeDefnParams),
> > -	term__term_list_to_var_list(TypeDefnParams, TypeDefnVars),
> >  
> >  	% XXX handle ExistQVars
> >  	ExistQVars0 = [],
> > +
> > +	hlds_data__get_type_defn_tparams(TypeDefn, TypeDefnParams),
> > +	term__term_list_to_var_list(TypeDefnParams, TypeDefnVars),
> 
> Can you explain the purpose of that change?

This predicate non-deterministically returns a ConsId and ArgTypes for
a type, VarType.  My original version was extremely inefficient
(profiling showed that it contributed about 15% to the execution time
of the compiler, IIRC) since it did a search right through the
cons_table looking for constructors belonging to the type.  The
revised version looks up the constructors in the type_table instead.

> 
> Apart from that, the relative diff looks fine.
> 

Okay, I'll make those changes and commit.


David
--------------------------------------------------------------------------
mercury-developers mailing list
Post messages to:       mercury-developers at cs.mu.oz.au
Administrative Queries: owner-mercury-developers at cs.mu.oz.au
Subscriptions:          mercury-developers-request at cs.mu.oz.au
--------------------------------------------------------------------------