[m-dev.] For review: Add implementation of reference types (global heap)

[Warwick Harvey]

Fergus Henderson <fjh at cs.mu.OZ.AU> writes:
> > +/* MR_make_permanent:
> > +**
> 
> Should be
> 
> 	/*
> 	** ...
> 	**

Fair enough.  I would have done it that way, but instead just copied the 
style used for deep_copy().  I'll fix that too while I'm at it.  :-)

> > :- pragma inline(new_nb_reference/2).
> > :- pragma c_code(new_nb_reference(X::in, Ref::out), will_not_call_mercury, "
> > 	incr_hp(Ref, 1);
> > 	save_transient_registers();
> > 	*(Word *)Ref = MR_make_long_lived(X, (Word *) TypeInfo_for_T,
> > 			(Word *) Ref);
> > 	restore_transient_registers();
> > ").
> > 
> > :- pragma inline(value/2).
> > :- pragma c_code(value(Ref::in, X::out), will_not_call_mercury, "
> > 	X = *(Word *) Ref;
> > ").
> > 
> > :- pragma inline(update/2).
> > :- pragma c_code(update(Ref::in, X::in), will_not_call_mercury, "
> > 	save_transient_registers();
> > 	*(Word *)Ref = MR_make_long_lived(X, (Word *) TypeInfo_for_T,
> > 			(Word *) Ref);
> > 	restore_transient_registers();
> > ").
> 
> If you were concerned about efficiency, you could wrap `#ifndef
> CONSERVATIVE_GC ... #endif' around those calls to
> save_transient_registers() and restore_transient_registers().

I was concerned about this initially, because it implies that this code (in 
extras) knows some details about the implementation of this function (in 
runtime).  But I suppose it would not be unreasonable to document this as 
part of the function's interface.

Thanks for your other comments, they were quite valuable.  I'll post the 
relative diff when it's done.

Warwick