Revised diff for parallel conjunction

[Fergus Henderson]

Thomas Charles CONWAY <conway at cs.mu.oz.au> writes:

>+	;	par_conj(hlds_goals, store_map)
>+					% parallel conjunction
>+					% The store_map specifies the locations
>+					% in which live variables should be
>+					% stored at the start of the parallel
>+					% conjunction.

At the start?  Are you sure that's correct?

>+	% instmap__unify(NonLocalVars, InstMapNonlocalvarPairss):
>+	%       Unify the `InstMaps' in the list of pairs resulting
>+	%	from different branches of a parallel conjunction and
>+	%	update the instantiatedness of all the nonlocal variables.
>+	%	The variable locking that is done when modechecking
>+	%	the individual conjuncts ensures that variables have
>+	%	at most one producer.

This is not ideal -- ideally it would be possible for
different parallel conjuncts to bind different parts of
a single variable.

That is sufficient for now, but it might be worth adding a comment
about this.

>+instmap__unify(NonLocals, InstMapList, ModeInfo0, ModeInfo) :-
...
>+			% having got the first instmapping, to use as
>+			% an accumulator, all instmap__unify_2 which
>+			% unifies each of the nonlocals from each instmap
>+			% with the corresponding inst in the accumulator.

s/all/call/ ?

>+	;	join_and_terminate(lval)
>+			% Signal that this thread of execution has finished in
>+			% the current parallel conjunction, then terminate it.
>+			% The synchronisation term specified by the
>+			% given lval. (See the documentation in par_conj_gen.m
>+			% and runtime/context.mod for further information about
>+			% synchronisation terms.)

s/specified/is specified/

>+	;	join_and_continue(lval, label)
>+			% Signal that this thread of execution has finished
>+			% in the current parallel conjunction, then branch to
>+			% the given label. The synchronisation
>+			% term specified by the given lval.

s/specified/is specified/

>+	;	mode_error_par_conj(merge_errors)
>+			% different arms of a parallel conj result in
>+			% mutually exclusive bindings
[...]
>+	;	mode_error_bind_var(var, inst, inst, lock_context)
> 			% attempt to bind a non-local variable inside
>+			% a negated context, or attempt to re-bind a variable
>+			% in a parallel conjunct
[...]
>+	;	mode_error_parallel_var(var, inst, inst)
>+			% attempt to bind a non-local variable that has already
>+			% been bound in another parallel conjunct.

I still don't understand the difference between these three possible
errors.  Certainly the parallel conjunct case of mode_error_bind_var
seems identical to mode_error_parallel_var.
And what exactly do you mean by "mutually exclusive"?

>+		{ LockContext = lambda_goal },
>+		prog_out__write_context(Context),
>+		io__write_string(
>+		"  scope error: attempt to bind a nonlocal variable inside\n"),
>+		prog_out__write_context(Context),
>+		io__write_string("  a pred/lambda goal.\n")

I suggest you just say "lambda goal" rather than "pred/lambda goal".

>+++ modecheck_unify.m	1997/10/21 05:28:13
>@@ -381,10 +381,11 @@
> 	% lock the non-locals
> 	% (a lambda goal is not allowed to bind any of the non-local
> 	% variables, since it could get called more than once)
>+	% `negation' as the locked_var_context is a place holder.

That comment is obsolete.

>+:- pred modecheck_par_conj_list(list(hlds_goal), list(hlds_goal),
>+		set(var), list(pair(instmap, set(var))), mode_info, mode_info).
>+:- mode modecheck_par_conj_list(in, out, in, out,
>+		mode_info_di, mode_info_uo) is det.
>+
>+modecheck_par_conj_list([], [], _NonLocals, []) --> [].
>+modecheck_par_conj_list([Goal0|Goals0], [Goal|Goals], NonLocals, 
>+		[InstMap - GoalNonLocals|InstMaps]) -->
>+	mode_info_dcg_get_instmap(InstMap0),
>+	{ Goal0 = _ - GoalInfo },
>+	{ goal_info_get_nonlocals(GoalInfo, GoalNonLocals) },
>+	mode_info_get_parallel_vars(PVars0),
>+	{ set__init(Bound0) },
>+	mode_info_set_parallel_vars([NonLocals - Bound0|PVars0]),
>+
>+	modecheck_goal(Goal0, Goal),
>+	mode_info_get_parallel_vars(PVars1),
>+	(
>+		{ PVars1 = [_ - Bound1|PVars2] },
>+		(
>+			{ PVars2 = [OuterNonLocals - OuterBound0|PVars3] },
>+			{ set__intersect(OuterNonLocals, Bound1, Bound) },
>+			{ set__union(OuterBound0, Bound, OuterBound) },
>+			{ PVars = [OuterNonLocals - OuterBound|PVars3] },
>+			mode_info_set_parallel_vars(PVars)
>+		;
>+			{ PVars2 = [] },
>+			mode_info_set_parallel_vars(PVars2)
>+		)
>+	;
>+		{ PVars1 = [] },
>+		{ error("lost parallel vars") }
>+	),
>+	mode_info_dcg_get_instmap(InstMap),
>+	mode_info_set_instmap(InstMap0),
>+	mode_info_lock_vars(Bound1, par_conj),
>+	modecheck_par_conj_list(Goals0, Goals, NonLocals, InstMaps),
>+	mode_info_unlock_vars(Bound1).

Some comments here would be helpful.

>-modecheck_set_var_inst(Var0, FinalInst, ModeInfo0, ModeInfo) :-
>+modecheck_set_var_inst(Var0, FinalInst, ModeInfo00, ModeInfo) :-
...
>+	(
>+		PVars0 = [],
>+		ModeInfo = ModeInfo3
>+	;
>+		PVars0 = [NonLocals - Bound0|PVars1],
>+		( set__member(Var0, NonLocals) ->
>+			set__insert(Bound0, Var0, Bound),
>+			PVars = [NonLocals - Bound|PVars1]
>+		;
>+			PVars = PVars0
>+		),
>+		mode_info_set_parallel_vars(PVars, ModeInfo3, ModeInfo)
> 	).

A comment here would be helpful.

--
Fergus Henderson <fjh at cs.mu.oz.au>   |  "I have always known that the pursuit
WWW: <http://www.cs.mu.oz.au/~fjh>   |  of excellence is a lethal habit"
PGP: finger fjh at 128.250.37.3         |     -- the last words of T. S. Garp.