Implement constant-expression simplification per Bernard

Frankpitt, plus some improvements from yours truly.  The simplifier depends
on the proiscachable field of pg_proc to tell it whether a function is
safe to pre-evaluate --- things like nextval() are not, for example.
Update pg_proc.h to contain reasonable cacheability information; as of
6.5.* hardly any functions were marked cacheable.  I may have erred too
far in the other direction; see recent mail to pghackers for more info.
This update does not force an initdb, exactly, but you won't see much
benefit from the simplifier until you do one.

src/backend/executor/execQual.c

@@ -7,7 +7,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/executor/execQual.c,v 1.60 1999/09/24 00:24:23 tgl Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/executor/execQual.c,v 1.61 1999/09/26 02:28:15 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -956,8 +956,8 @@ ExecEvalFunc(Expr *funcClause,
 static Datum
 ExecEvalNot(Expr *notclause, ExprContext *econtext, bool *isNull)
 {
-	Datum		expr_value;
 	Node	   *clause;
+	Datum		expr_value;
 	bool		isDone;
 
 	clause = lfirst(notclause->args);
@@ -995,67 +995,47 @@ ExecEvalOr(Expr *orExpr, ExprContext *econtext, bool *isNull)
 	List	   *clauses;
 	List	   *clause;
 	bool		isDone;
-	bool		IsNull;
-	Datum		const_value = 0;
+	bool		AnyNull;
+	Datum		clause_value;
 
-	IsNull = false;
 	clauses = orExpr->args;
+	AnyNull = false;
 
 	/*
-	 * we use three valued logic functions here... we evaluate each of the
-	 * clauses in turn, as soon as one is true we return that value.  If
-	 * none is true and  none of the clauses evaluate to NULL we return
-	 * the value of the last clause evaluated (which should be false) with
-	 * *isNull set to false else if none is true and at least one clause
-	 * evaluated to NULL we set *isNull flag to true -
+	 * If any of the clauses is TRUE, the OR result is TRUE regardless
+	 * of the states of the rest of the clauses, so we can stop evaluating
+	 * and return TRUE immediately.  If none are TRUE and one or more is
+	 * NULL, we return NULL; otherwise we return FALSE.  This makes sense
+	 * when you interpret NULL as "don't know": if we have a TRUE then the
+	 * OR is TRUE even if we aren't sure about some of the other inputs.
+	 * If all the known inputs are FALSE, but we have one or more "don't
+	 * knows", then we have to report that we "don't know" what the OR's
+	 * result should be --- perhaps one of the "don't knows" would have been
+	 * TRUE if we'd known its value.  Only when all the inputs are known
+	 * to be FALSE can we state confidently that the OR's result is FALSE.
 	 */
 	foreach(clause, clauses)
 	{
-
 		/*
 		 * We don't iterate over sets in the quals, so pass in an isDone
 		 * flag, but ignore it.
 		 */
-		const_value = ExecEvalExpr((Node *) lfirst(clause),
+		clause_value = ExecEvalExpr((Node *) lfirst(clause),
 									econtext,
 									isNull,
 									&isDone);
-
 		/*
-		 * if the expression evaluates to null, then we remember it in the
-		 * local IsNull flag, if none of the clauses are true then we need
-		 * to set *isNull to true again.
+		 * if we have a non-null true result, then return it.
 		 */
 		if (*isNull)
-		{
-			IsNull = *isNull;
-
-			/*
-			 * Many functions don't (or can't!) check if an argument is
-			 * NULL or NOT_NULL and may return TRUE (1) with *isNull TRUE
-			 * (an_int4_column <> 1: int4ne returns TRUE for NULLs). Not
-			 * having time to fix the function manager I want to fix OR:
-			 * if we had 'x <> 1 OR x isnull' then when x is NULL TRUE was
-			 * returned by the 'x <> 1' clause ... but ExecQualClause says
-			 * that the qualification should *fail* if isnull is TRUE for
-			 * any value returned by ExecEvalExpr. So, force this rule
-			 * here: if isnull is TRUE then the clause failed. Note:
-			 * nullvalue() & nonnullvalue() always sets isnull to FALSE
-			 * for NULLs. - vadim 09/22/97
-			 */
-			const_value = 0;
+			AnyNull = true;		/* remember we got a null */
+		else if (DatumGetInt32(clause_value) != 0)
+			return clause_value;
 	}
 
-		/*
-		 * if we have a true result, then we return it.
-		 */
-		if (DatumGetInt32(const_value) != 0)
-			return const_value;
-	}
-
-	/* IsNull is true if at least one clause evaluated to NULL */
-	*isNull = IsNull;
-	return const_value;
+	/* AnyNull is true if at least one clause evaluated to NULL */
+	*isNull = AnyNull;
+	return (Datum) false;
 }
 
 /* ----------------------------------------------------------------
@@ -1067,49 +1047,43 @@ ExecEvalAnd(Expr *andExpr, ExprContext *econtext, bool *isNull)
 {
 	List	   *clauses;
 	List	   *clause;
-	Datum		const_value = 0;
 	bool		isDone;
-	bool		IsNull;
-
-	IsNull = false;
+	bool		AnyNull;
+	Datum		clause_value;
 
 	clauses = andExpr->args;
+	AnyNull = false;
 
 	/*
-	 * we evaluate each of the clauses in turn, as soon as one is false we
-	 * return that value.  If none are false or NULL then we return the
-	 * value of the last clause evaluated, which should be true.
+	 * If any of the clauses is FALSE, the AND result is FALSE regardless
+	 * of the states of the rest of the clauses, so we can stop evaluating
+	 * and return FALSE immediately.  If none are FALSE and one or more is
+	 * NULL, we return NULL; otherwise we return TRUE.  This makes sense
+	 * when you interpret NULL as "don't know", using the same sort of
+	 * reasoning as for OR, above.
 	 */
 	foreach(clause, clauses)
 	{
-
 		/*
 		 * We don't iterate over sets in the quals, so pass in an isDone
 		 * flag, but ignore it.
 		 */
-		const_value = ExecEvalExpr((Node *) lfirst(clause),
+		clause_value = ExecEvalExpr((Node *) lfirst(clause),
 									econtext,
 									isNull,
 									&isDone);
-
 		/*
-		 * if the expression evaluates to null, then we remember it in
-		 * IsNull, if none of the clauses after this evaluates to false we
-		 * will have to set *isNull to true again.
+		 * if we have a non-null false result, then return it.
 		 */
 		if (*isNull)
-			IsNull = *isNull;
-
-		/*
-		 * if we have a false result, then we return it, since the
-		 * conjunction must be false.
-		 */
-		if (DatumGetInt32(const_value) == 0)
-			return const_value;
+			AnyNull = true;		/* remember we got a null */
+		else if (DatumGetInt32(clause_value) == 0)
+			return clause_value;
 	}
 
-	*isNull = IsNull;
-	return const_value;
+	/* AnyNull is true if at least one clause evaluated to NULL */
+	*isNull = AnyNull;
+	return (Datum) (! AnyNull);
 }
 
 /* ----------------------------------------------------------------
@@ -1126,7 +1100,7 @@ ExecEvalCase(CaseExpr *caseExpr, ExprContext *econtext, bool *isNull)
 {
 	List	   *clauses;
 	List	   *clause;
-	Datum		const_value = 0;
+	Datum		clause_value;
 	bool		isDone;
 
 	clauses = caseExpr->args;
@@ -1144,7 +1118,7 @@ ExecEvalCase(CaseExpr *caseExpr, ExprContext *econtext, bool *isNull)
 		 * We don't iterate over sets in the quals, so pass in an isDone
 		 * flag, but ignore it.
 		 */
-		const_value = ExecEvalExpr((Node *) wclause->expr,
+		clause_value = ExecEvalExpr(wclause->expr,
 									econtext,
 									isNull,
 									&isDone);
@@ -1154,27 +1128,25 @@ ExecEvalCase(CaseExpr *caseExpr, ExprContext *econtext, bool *isNull)
 		 * case statement is satisfied.  A NULL result from the test is
 		 * not considered true.
 		 */
-		if (DatumGetInt32(const_value) != 0 && ! *isNull)
+		if (DatumGetInt32(clause_value) != 0 && ! *isNull)
 		{
-			const_value = ExecEvalExpr((Node *) wclause->result,
+			return ExecEvalExpr(wclause->result,
 								econtext,
 								isNull,
 								&isDone);
-			return (Datum) const_value;
 		}
 	}
 
 	if (caseExpr->defresult)
 	{
-		const_value = ExecEvalExpr((Node *) caseExpr->defresult,
+		return ExecEvalExpr(caseExpr->defresult,
 							econtext,
 							isNull,
 							&isDone);
 	}
-	else
-		*isNull = true;
 
-	return const_value;
+	*isNull = true;
+	return (Datum) 0;
 }
 
 /* ----------------------------------------------------------------
@@ -1357,7 +1329,6 @@ bool
 ExecQual(List *qual, ExprContext *econtext)
 {
 	List	   *clause;
-	bool		result;
 
 	/*
 	 * debugging stuff
@@ -1378,27 +1349,17 @@ ExecQual(List *qual, ExprContext *econtext)
 	 * a "qual" is a list of clauses.  To evaluate the qual, we evaluate
 	 * each of the clauses in the list.
 	 *
-	 * ExecQualClause returns true when we know the qualification *failed* so
-	 * we just pass each clause in qual to it until we know the qual
+	 * ExecQualClause returns true when we know the qualification *failed*
+	 * so we just pass each clause in qual to it until we know the qual
 	 * failed or there are no more clauses.
 	 */
-	result = false;
 
 	foreach(clause, qual)
 	{
-		result = ExecQualClause((Node *) lfirst(clause), econtext);
-		if (result == true)
-			break;
+		if (ExecQualClause((Node *) lfirst(clause), econtext))
+			return false;		/* qual failed, so return false */
 	}
 
-	/*
-	 * if result is true, then it means a clause failed so we return
-	 * false.  if result is false then it means no clause failed so we
-	 * return true.
-	 */
-	if (result == true)
-		return false;
-
 	return true;
 }
 

src/backend/nodes/equalfuncs.c

@@ -7,7 +7,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/nodes/equalfuncs.c,v 1.48 1999/08/21 03:48:57 tgl Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/nodes/equalfuncs.c,v 1.49 1999/09/26 02:28:21 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -223,6 +223,22 @@ _equalAggref(Aggref *a, Aggref *b)
 	return true;
 }
 
+static bool
+_equalSubLink(SubLink *a, SubLink *b)
+{
+	if (a->subLinkType != b->subLinkType)
+		return false;
+	if (a->useor != b->useor)
+		return false;
+	if (!equal(a->lefthand, b->lefthand))
+		return false;
+	if (!equal(a->oper, b->oper))
+		return false;
+	if (!equal(a->subselect, b->subselect))
+		return false;
+	return true;
+}
+
 static bool
 _equalArray(Array *a, Array *b)
 {
@@ -393,7 +409,7 @@ _equalSubPlan(SubPlan *a, SubPlan *b)
 	if (a->plan_id != b->plan_id)
 		return false;
 
-	if (!equal(a->sublink->oper, b->sublink->oper))
+	if (!equal(a->sublink, b->sublink))
 		return false;
 
 	return true;
@@ -713,6 +729,9 @@ equal(void *a, void *b)
 		case T_Aggref:
 			retval = _equalAggref(a, b);
 			break;
+		case T_SubLink:
+			retval = _equalSubLink(a, b);
+			break;
 		case T_Func:
 			retval = _equalFunc(a, b);
 			break;

src/backend/optimizer/plan/planmain.c

@@ -7,7 +7,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/optimizer/plan/planmain.c,v 1.44 1999/09/13 00:17:25 tgl Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/optimizer/plan/planmain.c,v 1.45 1999/09/26 02:28:27 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -66,24 +66,41 @@ query_planner(Query *root,
 	List	   *level_tlist;
 	Plan	   *subplan;
 
+	/*
+	 * Simplify constant expressions in both targetlist and qual.
+	 *
+	 * Note that at this point the qual has not yet been converted to
+	 * implicit-AND form, so we can apply eval_const_expressions directly.
+	 * Also note that we need to do this before SS_process_sublinks,
+	 * because that routine inserts bogus "Const" nodes.
+	 */
+	tlist = (List *) eval_const_expressions((Node *) tlist);
+	qual = (List *) eval_const_expressions((Node *) qual);
+
+	/*
+	 * Canonicalize the qual, and convert it to implicit-AND format.
+	 */
+	qual = canonicalize_qual((Expr *) qual, true);
+#ifdef OPTIMIZER_DEBUG
+	printf("After canonicalize_qual()\n");
+	pprint(qual);
+#endif
+
+	/* Replace uplevel vars with Param nodes */
 	if (PlannerQueryLevel > 1)
 	{
-		/* should copy be made ? */
 		tlist = (List *) SS_replace_correlation_vars((Node *) tlist);
 		qual = (List *) SS_replace_correlation_vars((Node *) qual);
 	}
+	/* Expand SubLinks to SubPlans */
 	if (root->hasSubLinks)
 		qual = (List *) SS_process_sublinks((Node *) qual);
 
-	qual = canonicalize_qual((Expr *) qual, true);
-#ifdef OPTIMIZER_DEBUG
-	printf("After canonicalize_qual()\n");
-	pprint(qual);
-#endif
-
 	/*
 	 * Pull out any non-variable qualifications so these can be put in the
-	 * topmost result node.
+	 * topmost result node.  (Any *really* non-variable quals will probably
+	 * have been optimized away by eval_const_expressions().  What we're
+	 * looking for here is quals that depend only on outer-level vars...)
 	 */
 	qual = pull_constant_clauses(qual, &constant_qual);
 

src/backend/optimizer/plan/planner.c

@@ -7,7 +7,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/optimizer/plan/planner.c,v 1.68 1999/09/18 19:07:00 tgl Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/optimizer/plan/planner.c,v 1.69 1999/09/26 02:28:27 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -301,7 +301,28 @@ union_planner(Query *parse)
 	 */
 	if (parse->havingQual)
 	{
-		List	   *ql;
+		/*--------------------
+		 * Require the havingQual to contain at least one aggregate function
+		 * (else it could have been done as a WHERE constraint).  This check
+		 * used to be much stricter, requiring an aggregate in each clause of
+		 * the CNF-ified qual.  However, that's probably overly anal-retentive.
+		 * We now do it first so that we will not complain if there is an
+		 * aggregate but it gets optimized away by eval_const_expressions().
+		 * The agg itself is never const, of course, but consider
+		 *		SELECT ... HAVING xyz OR (COUNT(*) > 1)
+		 * where xyz reduces to constant true in a particular query.
+		 * We probably should not refuse this query.
+		 *--------------------
+		 */
+		if (pull_agg_clause(parse->havingQual) == NIL)
+			elog(ERROR, "SELECT/HAVING requires aggregates to be valid");
+
+		/* Simplify constant expressions in havingQual */
+		parse->havingQual = eval_const_expressions(parse->havingQual);
+
+		/* Convert the havingQual to implicit-AND normal form */
+		parse->havingQual = (Node *)
+			canonicalize_qual((Expr *) parse->havingQual, true);
 
 		/* Replace uplevel Vars with Params */
 		if (PlannerQueryLevel > 1)
@@ -323,20 +344,6 @@ union_planner(Query *parse)
 												  parse->targetList))
 				elog(ERROR, "Sub-SELECT in HAVING clause must use only GROUPed attributes from outer SELECT");
 		}
-
-		/* convert the havingQual to implicit-AND normal form */
-		parse->havingQual = (Node *)
-			canonicalize_qual((Expr *) parse->havingQual, true);
-
-		/*
-		 * Require an aggregate function to appear in each clause of the
-		 * havingQual (else it could have been done as a WHERE constraint).
-		 */
-		foreach(ql, (List *) parse->havingQual)
-		{
-			if (pull_agg_clause(lfirst(ql)) == NIL)
-				elog(ERROR, "SELECT/HAVING requires aggregates to be valid");
-		}
 	}
 
 	/*

src/include/optimizer/clauses.h

@@ -6,7 +6,7 @@
  *
  * Copyright (c) 1994, Regents of the University of California
  *
- * $Id: clauses.h,v 1.29 1999/08/22 20:14:56 tgl Exp $
+ * $Id: clauses.h,v 1.30 1999/09/26 02:28:44 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -52,6 +52,8 @@ extern void get_rels_atts(Node *clause, int *relid1,
 			  AttrNumber *attno1, int *relid2, AttrNumber *attno2);
 extern void CommuteClause(Expr *clause);
 
+extern Node *eval_const_expressions(Node *node);
+
 extern bool expression_tree_walker(Node *node, bool (*walker) (),
 								   void *context);
 extern Node *expression_tree_mutator(Node *node, Node * (*mutator) (),