Attached are two patches to implement and document anonymous composite

types for Table Functions, as previously proposed on HACKERS. Here is a
brief explanation:

1. Creates a new pg_type typtype: 'p' for pseudo type (currently either
     'b' for base or 'c' for catalog, i.e. a class).

2. Creates new builtin type of typtype='p' named RECORD. This is the
     first of potentially several pseudo types.

3. Modify FROM clause grammer to accept:
     SELECT * FROM my_func() AS m(colname1 type1, colname2 type1, ...)
     where m is the table alias, colname1, etc are the column names, and
     type1, etc are the column types.

4. When typtype == 'p' and the function return type is RECORD, a list
     of column defs is required, and when typtype != 'p', it is
disallowed.

5. A check was added to ensure that the tupdesc provide via the parser
     and the actual return tupdesc match in number and type of
attributes.

When creating a function you can do:
     CREATE FUNCTION foo(text) RETURNS setof RECORD ...

When using it you can do:
     SELECT * from foo(sqlstmt) AS (f1 int, f2 text, f3 timestamp)
       or
     SELECT * from foo(sqlstmt) AS f(f1 int, f2 text, f3 timestamp)
       or
     SELECT * from foo(sqlstmt) f(f1 int, f2 text, f3 timestamp)

Included in the patches are adjustments to the regression test sql and
expected files, and documentation.

p.s.
     This potentially solves (or at least improves) the issue of builtin
     Table Functions. They can be bootstrapped as returning RECORD, and
     we can wrap system views around them with properly specified column
     defs. For example:

     CREATE VIEW pg_settings AS
       SELECT s.name, s.setting
       FROM show_all_settings()AS s(name text, setting text);

     Then we can also add the UPDATE RULE that I previously posted to
     pg_settings, and have pg_settings act like a virtual table, allowing
     settings to be queried and set.


Joe Conway

doc/src/sgml/ref/select.sgml

 <!--
-$Header: /cvsroot/pgsql/doc/src/sgml/ref/select.sgml,v 1.54 2002/04/23 02:07:16 tgl Exp $
+$Header: /cvsroot/pgsql/doc/src/sgml/ref/select.sgml,v 1.55 2002/08/04 19:48:09 momjian Exp $
 PostgreSQL documentation
 -->
 
@@ -40,6 +40,12 @@ where <replaceable class="PARAMETER">from_item</replaceable> can be:
 ( <replaceable class="PARAMETER">select</replaceable> )
     [ AS ] <replaceable class="PARAMETER">alias</replaceable> [ ( <replaceable class="PARAMETER">column_alias_list</replaceable> ) ]
 |
+<replaceable class="PARAMETER">table_function_name</replaceable> ( [ <replaceable class="parameter">argtype</replaceable> [, ...] ] )
+    [ AS ] <replaceable class="PARAMETER">alias</replaceable> [ ( <replaceable class="PARAMETER">column_alias_list</replaceable> | <replaceable class="PARAMETER">column_definition_list</replaceable> ) ]
+|
+<replaceable class="PARAMETER">table_function_name</replaceable> ( [ <replaceable class="parameter">argtype</replaceable> [, ...] ] )
+    AS ( <replaceable class="PARAMETER">column_definition_list</replaceable> )
+|
 <replaceable class="PARAMETER">from_item</replaceable> [ NATURAL ] <replaceable class="PARAMETER">join_type</replaceable> <replaceable class="PARAMETER">from_item</replaceable>
     [ ON <replaceable class="PARAMETER">join_condition</replaceable> | USING ( <replaceable class="PARAMETER">join_column_list</replaceable> ) ]
   </synopsis>
@@ -82,7 +88,7 @@ where <replaceable class="PARAMETER">from_item</replaceable> can be:
       <term><replaceable class="PARAMETER">from_item</replaceable></term>
       <listitem>
        <para>
-        A table reference, sub-SELECT, or JOIN clause.  See below for details.
+        A table reference, sub-SELECT, table function, or JOIN clause.  See below for details.
        </para>
       </listitem>
      </varlistentry>
@@ -156,6 +162,23 @@ where <replaceable class="PARAMETER">from_item</replaceable> can be:
        </para>
       </listitem>
      </varlistentry>
+
+     <varlistentry>
+      <term><replaceable class="PARAMETER">table function</replaceable></term>
+      <listitem>
+       <para>
+	A table function can appear in the FROM clause.  This acts as though
+	its output were created as a temporary table for the duration of
+	this single SELECT command. An alias may also be used. If an alias is
+	written, a column alias list can also be written to provide	substitute names
+	for one or more columns of the table function. If the table function has been
+	defined as returning the RECORD data type, an alias, or the keyword AS, must
+    also be present, followed by a column definition list in the form
+	( <replaceable class="PARAMETER">column_name</replaceable> <replaceable class="PARAMETER">data_type</replaceable> [, ... ] ).
+	The column definition list must match the actual number and types returned by the function.
+       </para>
+      </listitem>
+     </varlistentry>
      
      <varlistentry>
       <term><replaceable class="PARAMETER">join_type</replaceable></term>
@@ -380,6 +403,19 @@ where <replaceable class="PARAMETER">from_item</replaceable> can be:
     grouping, aggregation, or sorting in a single query.
    </para>
 
+   <para>
+    A FROM item can be a table function (i.e. a function that returns
+    multiple rows and columns).  When a table function is created, it may
+	be defined to return a named scalar or composite data type (an existing
+	scalar data type, or a table or view name), or it may be defined to return
+	a RECORD data type. When a table function is defined to return RECORD, it
+	must be followed in the FROM clause by an alias, or the keyword AS alone,
+	and then by a parenthesized list of column names and types. This provides
+	a query-time composite type definition. The FROM clause composite type
+	must match the actual composite type returned from the function or an
+	ERROR will be generated.
+   </para>
+
    <para>
     Finally, a FROM item can be a JOIN clause, which combines two simpler
     FROM items.  (Use parentheses if necessary to determine the order
@@ -925,6 +961,43 @@ SELECT actors.name
  Warren Beatty
  Westward
  Woody Allen
+</programlisting>
+  </para>
+
+  <para>
+   This example shows how to use a table function, both with and without
+   a column definition list.
+
+<programlisting>
+distributors:
+ did |     name
+-----+--------------
+ 108 | Westward
+ 111 | Walt Disney
+ 112 | Warner Bros.
+ ...
+
+CREATE FUNCTION distributors(int)
+  RETURNS SETOF distributors AS '
+  SELECT * FROM distributors WHERE did = $1;
+  ' LANGUAGE SQL;
+
+SELECT * FROM distributors(111);
+ did |    name
+-----+-------------
+ 111 | Walt Disney
+(1 row)
+
+CREATE FUNCTION distributors_2(int)
+  RETURNS SETOF RECORD AS '
+  SELECT * FROM distributors WHERE did = $1;
+  ' LANGUAGE SQL;
+
+SELECT * FROM distributors_2(111) AS (f1 int, f2 text);
+ f1  |     f2
+-----+-------------
+ 111 | Walt Disney
+(1 row)
 </programlisting>
   </para>
  </refsect1>

src/backend/access/common/tupdesc.c

@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/access/common/tupdesc.c,v 1.83 2002/08/02 18:15:04 tgl Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/access/common/tupdesc.c,v 1.84 2002/08/04 19:48:09 momjian Exp $
  *
  * NOTES
  *	  some of the executor utility code such as "ExecTypeFromTL" should be
@@ -24,6 +24,7 @@
 #include "catalog/namespace.h"
 #include "catalog/pg_type.h"
 #include "nodes/parsenodes.h"
+#include "parser/parse_relation.h"
 #include "parser/parse_type.h"
 #include "utils/builtins.h"
 #include "utils/syscache.h"
@@ -600,46 +601,53 @@ RelationNameGetTupleDesc(char *relname)
 TupleDesc
 TypeGetTupleDesc(Oid typeoid, List *colaliases)
 {
-	Oid			relid = typeidTypeRelid(typeoid);
-	TupleDesc	tupdesc;
+	char		functyptype = typeid_get_typtype(typeoid);
+	TupleDesc	tupdesc = NULL;
 
 	/*
 	 * Build a suitable tupledesc representing the output rows
 	 */
-	if (OidIsValid(relid))
+	if (functyptype == 'c')
 	{
 		/* Composite data type, i.e. a table's row type */
-		Relation	rel;
-		int			natts;
-
-		rel = relation_open(relid, AccessShareLock);
-		tupdesc = CreateTupleDescCopy(RelationGetDescr(rel));
-		natts = tupdesc->natts;
-		relation_close(rel, AccessShareLock);
+		Oid			relid = typeidTypeRelid(typeoid);
 
-		/* check to see if we've given column aliases */
-		if(colaliases != NIL)
+		if (OidIsValid(relid))
 		{
-			char	   *label;
-			int			varattno;
+			Relation	rel;
+			int			natts;
 
-			/* does the List length match the number of attributes */
-			if (length(colaliases) != natts)
-				elog(ERROR, "TypeGetTupleDesc: number of aliases does not match number of attributes");
+			rel = relation_open(relid, AccessShareLock);
+			tupdesc = CreateTupleDescCopy(RelationGetDescr(rel));
+			natts = tupdesc->natts;
+			relation_close(rel, AccessShareLock);
 
-			/* OK, use the aliases instead */
-			for (varattno = 0; varattno < natts; varattno++)
+			/* check to see if we've given column aliases */
+			if(colaliases != NIL)
 			{
-				label = strVal(nth(varattno, colaliases));
-
-				if (label != NULL)
-					namestrcpy(&(tupdesc->attrs[varattno]->attname), label);
-				else
-					MemSet(NameStr(tupdesc->attrs[varattno]->attname), 0, NAMEDATALEN);
+				char	   *label;
+				int			varattno;
+
+				/* does the List length match the number of attributes */
+				if (length(colaliases) != natts)
+					elog(ERROR, "TypeGetTupleDesc: number of aliases does not match number of attributes");
+
+				/* OK, use the aliases instead */
+				for (varattno = 0; varattno < natts; varattno++)
+				{
+					label = strVal(nth(varattno, colaliases));
+
+					if (label != NULL)
+						namestrcpy(&(tupdesc->attrs[varattno]->attname), label);
+					else
+						MemSet(NameStr(tupdesc->attrs[varattno]->attname), 0, NAMEDATALEN);
+				}
 			}
 		}
+		else
+			elog(ERROR, "Invalid return relation specified for function");
 	}
-	else
+	else if (functyptype == 'b')
 	{
 		/* Must be a base data type, i.e. scalar */
 		char	   *attname;
@@ -664,6 +672,11 @@ TypeGetTupleDesc(Oid typeoid, List *colaliases)
 						   0,
 						   false);
 	}
+	else if (functyptype == 'p' && typeoid == RECORDOID)
+		elog(ERROR, "Unable to determine tuple description for function"
+						" returning \"record\"");
+	else
+		elog(ERROR, "Unknown kind of return type specified for function");
 
 	return tupdesc;
 }

src/backend/catalog/pg_proc.c

@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/catalog/pg_proc.c,v 1.82 2002/08/02 18:15:05 tgl Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/catalog/pg_proc.c,v 1.83 2002/08/04 19:48:09 momjian Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -25,6 +25,7 @@
 #include "miscadmin.h"
 #include "parser/parse_coerce.h"
 #include "parser/parse_expr.h"
+#include "parser/parse_relation.h"
 #include "parser/parse_type.h"
 #include "tcop/tcopprot.h"
 #include "utils/builtins.h"
@@ -33,7 +34,7 @@
 #include "utils/syscache.h"
 
 
-static void checkretval(Oid rettype, List *queryTreeList);
+static void checkretval(Oid rettype, char fn_typtype, List *queryTreeList);
 Datum fmgr_internal_validator(PG_FUNCTION_ARGS);
 Datum fmgr_c_validator(PG_FUNCTION_ARGS);
 Datum fmgr_sql_validator(PG_FUNCTION_ARGS);
@@ -367,94 +368,113 @@ checkretval(Oid rettype, List *queryTreeList)
 	 */
 	tlistlen = ExecCleanTargetListLength(tlist);
 
-	/*
-	 * For base-type returns, the target list should have exactly one
-	 * entry, and its type should agree with what the user declared. (As
-	 * of Postgres 7.2, we accept binary-compatible types too.)
-	 */
 	typerelid = typeidTypeRelid(rettype);
-	if (typerelid == InvalidOid)
+
+	if (fn_typtype == 'b')
 	{
-		if (tlistlen != 1)
-			elog(ERROR, "function declared to return %s returns multiple columns in final SELECT",
-				 format_type_be(rettype));
+		/*
+		 * For base-type returns, the target list should have exactly one
+		 * entry, and its type should agree with what the user declared. (As
+		 * of Postgres 7.2, we accept binary-compatible types too.)
+		 */
 
-		restype = ((TargetEntry *) lfirst(tlist))->resdom->restype;
-		if (!IsBinaryCompatible(restype, rettype))
-			elog(ERROR, "return type mismatch in function: declared to return %s, returns %s",
-				 format_type_be(rettype), format_type_be(restype));
+		if (typerelid == InvalidOid)
+		{
+			if (tlistlen != 1)
+				elog(ERROR, "function declared to return %s returns multiple columns in final SELECT",
+					 format_type_be(rettype));
 
-		return;
-	}
+			restype = ((TargetEntry *) lfirst(tlist))->resdom->restype;
+			if (!IsBinaryCompatible(restype, rettype))
+				elog(ERROR, "return type mismatch in function: declared to return %s, returns %s",
+					 format_type_be(rettype), format_type_be(restype));
 
-	/*
-	 * If the target list is of length 1, and the type of the varnode in
-	 * the target list matches the declared return type, this is okay.
-	 * This can happen, for example, where the body of the function is
-	 * 'SELECT func2()', where func2 has the same return type as the
-	 * function that's calling it.
-	 */
-	if (tlistlen == 1)
-	{
-		restype = ((TargetEntry *) lfirst(tlist))->resdom->restype;
-		if (IsBinaryCompatible(restype, rettype))
 			return;
-	}
-
-	/*
-	 * By here, the procedure returns a tuple or set of tuples.  This part
-	 * of the typechecking is a hack. We look up the relation that is the
-	 * declared return type, and scan the non-deleted attributes to ensure
-	 * that they match the datatypes of the non-resjunk columns.
-	 */
-	reln = heap_open(typerelid, AccessShareLock);
-	relnatts = reln->rd_rel->relnatts;
-	rellogcols = 0;				/* we'll count nondeleted cols as we go */
-	colindex = 0;
+		}
 
-	foreach(tlistitem, tlist)
+		/*
+		 * If the target list is of length 1, and the type of the varnode in
+		 * the target list matches the declared return type, this is okay.
+		 * This can happen, for example, where the body of the function is
+		 * 'SELECT func2()', where func2 has the same return type as the
+		 * function that's calling it.
+		 */
+		if (tlistlen == 1)
+		{
+			restype = ((TargetEntry *) lfirst(tlist))->resdom->restype;
+			if (IsBinaryCompatible(restype, rettype))
+				return;
+		}
+	}
+	else if (fn_typtype == 'c')
 	{
-		TargetEntry *tle = (TargetEntry *) lfirst(tlistitem);
-		Form_pg_attribute attr;
-		Oid			tletype;
-		Oid			atttype;
+		/*
+		 * By here, the procedure returns a tuple or set of tuples.  This part
+		 * of the typechecking is a hack. We look up the relation that is the
+		 * declared return type, and scan the non-deleted attributes to ensure
+		 * that they match the datatypes of the non-resjunk columns.
+		 */
+		reln = heap_open(typerelid, AccessShareLock);
+		relnatts = reln->rd_rel->relnatts;
+		rellogcols = 0;				/* we'll count nondeleted cols as we go */
+		colindex = 0;
+
+		foreach(tlistitem, tlist)
+		{
+			TargetEntry *tle = (TargetEntry *) lfirst(tlistitem);
+			Form_pg_attribute attr;
+			Oid			tletype;
+			Oid			atttype;
+
+			if (tle->resdom->resjunk)
+				continue;
+
+			do {
+				colindex++;
+				if (colindex > relnatts)
+					elog(ERROR, "function declared to return %s does not SELECT the right number of columns (%d)",
+						 format_type_be(rettype), rellogcols);
+				attr = reln->rd_att->attrs[colindex - 1];
+			} while (attr->attisdropped);
+			rellogcols++;
 
-		if (tle->resdom->resjunk)
-			continue;
+			tletype = exprType(tle->expr);
+			atttype = attr->atttypid;
+			if (!IsBinaryCompatible(tletype, atttype))
+				elog(ERROR, "function declared to return %s returns %s instead of %s at column %d",
+					 format_type_be(rettype),
+					 format_type_be(tletype),
+					 format_type_be(atttype),
+					 rellogcols);
+		}
 
-		do {
+		for (;;)
+		{
 			colindex++;
 			if (colindex > relnatts)
-				elog(ERROR, "function declared to return %s does not SELECT the right number of columns (%d)",
-					 format_type_be(rettype), rellogcols);
-			attr = reln->rd_att->attrs[colindex - 1];
-		} while (attr->attisdropped);
-		rellogcols++;
-
-		tletype = exprType(tle->expr);
-		atttype = attr->atttypid;
-		if (!IsBinaryCompatible(tletype, atttype))
-			elog(ERROR, "function declared to return %s returns %s instead of %s at column %d",
-				 format_type_be(rettype),
-				 format_type_be(tletype),
-				 format_type_be(atttype),
-				 rellogcols);
-	}
+				break;
+			if (!reln->rd_att->attrs[colindex - 1]->attisdropped)
+				rellogcols++;
+		}
 
-	for (;;)
-	{
-		colindex++;
-		if (colindex > relnatts)
-			break;
-		if (!reln->rd_att->attrs[colindex - 1]->attisdropped)
-			rellogcols++;
-	}
+		if (tlistlen != rellogcols)
+			elog(ERROR, "function declared to return %s does not SELECT the right number of columns (%d)",
+				 format_type_be(rettype), rellogcols);
 
-	if (tlistlen != rellogcols)
-		elog(ERROR, "function declared to return %s does not SELECT the right number of columns (%d)",
-			 format_type_be(rettype), rellogcols);
+		heap_close(reln, AccessShareLock);
 
-	heap_close(reln, AccessShareLock);
+		return;
+	}
+	else if (fn_typtype == 'p' && rettype == RECORDOID)
+	{
+		/*
+		 * For RECORD return type, defer this check until we get the
+		 * first tuple.
+		 */
+		return;
+	}
+	else
+		elog(ERROR, "Unknown kind of return type specified for function");
 }
 
 
@@ -553,6 +573,7 @@ fmgr_sql_validator(PG_FUNCTION_ARGS)
 	bool		isnull;
 	Datum		tmp;
 	char	   *prosrc;
+	char		functyptype;
 
 	tuple = SearchSysCache(PROCOID, funcoid, 0, 0, 0);
 	if (!HeapTupleIsValid(tuple))
@@ -569,8 +590,11 @@ fmgr_sql_validator(PG_FUNCTION_ARGS)
 
 	prosrc = DatumGetCString(DirectFunctionCall1(textout, tmp));
 
+	/* check typtype to see if we have a predetermined return type */
+	functyptype = typeid_get_typtype(proc->prorettype);
+
 	querytree_list = pg_parse_and_rewrite(prosrc, proc->proargtypes, proc->pronargs);
-	checkretval(proc->prorettype, querytree_list);
+	checkretval(proc->prorettype, functyptype, querytree_list);
 
 	ReleaseSysCache(tuple);
 	PG_RETURN_BOOL(true);

src/backend/executor/functions.c

@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/executor/functions.c,v 1.52 2002/06/20 20:29:28 momjian Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/executor/functions.c,v 1.53 2002/08/04 19:48:09 momjian Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -194,7 +194,8 @@ init_sql_fcache(FmgrInfo *finfo)
 	 * get the type length and by-value flag from the type tuple
 	 */
 	fcache->typlen = typeStruct->typlen;
-	if (typeStruct->typrelid == InvalidOid)
+
+	if (typeStruct->typtype == 'b')
 	{
 		/* The return type is not a relation, so just use byval */
 		fcache->typbyval = typeStruct->typbyval;

src/backend/executor/nodeFunctionscan.c

@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/executor/nodeFunctionscan.c,v 1.3 2002/07/20 05:16:58 momjian Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/executor/nodeFunctionscan.c,v 1.4 2002/08/04 19:48:09 momjian Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -31,6 +31,7 @@
 #include "executor/nodeFunctionscan.h"
 #include "parser/parsetree.h"
 #include "parser/parse_expr.h"
+#include "parser/parse_relation.h"
 #include "parser/parse_type.h"
 #include "storage/lmgr.h"
 #include "tcop/pquery.h"
@@ -39,14 +40,11 @@
 #include "utils/tuplestore.h"
 
 static TupleTableSlot *FunctionNext(FunctionScan *node);
-static TupleTableSlot *function_getonetuple(TupleTableSlot *slot,
-											Node *expr,
-											ExprContext *econtext,
-											TupleDesc tupdesc,
-											bool returnsTuple,
+static TupleTableSlot *function_getonetuple(FunctionScanState *scanstate,
 											bool *isNull,
 											ExprDoneCond *isDone);
 static FunctionMode get_functionmode(Node *expr);
+static bool tupledesc_mismatch(TupleDesc tupdesc1, TupleDesc tupdesc2);
 
 /* ----------------------------------------------------------------
  *						Scan Support
@@ -62,9 +60,6 @@ static TupleTableSlot *
 FunctionNext(FunctionScan *node)
 {
 	TupleTableSlot	   *slot;
-	Node			   *expr;
-	ExprContext		   *econtext;
-	TupleDesc			tupdesc;
 	EState			   *estate;
 	ScanDirection		direction;
 	Tuplestorestate	   *tuplestorestate;
@@ -78,11 +73,8 @@ FunctionNext(FunctionScan *node)
 	scanstate = (FunctionScanState *) node->scan.scanstate;
 	estate = node->scan.plan.state;
 	direction = estate->es_direction;
-	econtext = scanstate->csstate.cstate.cs_ExprContext;
 
 	tuplestorestate = scanstate->tuplestorestate;
-	tupdesc = scanstate->tupdesc;
-	expr = scanstate->funcexpr;
 
 	/*
 	 * If first time through, read all tuples from function and pass them to
@@ -108,10 +100,7 @@ FunctionNext(FunctionScan *node)
 
 			isNull = false;
 			isDone = ExprSingleResult;
-			slot = function_getonetuple(scanstate->csstate.css_ScanTupleSlot,
-										expr, econtext, tupdesc,
-										scanstate->returnsTuple,
-										&isNull, &isDone);
+			slot = function_getonetuple(scanstate, &isNull, &isDone);
 			if (TupIsNull(slot))
 				break;
 
@@ -169,7 +158,8 @@ ExecInitFunctionScan(FunctionScan *node, EState *estate, Plan *parent)
 	RangeTblEntry	   *rte;
 	Oid					funcrettype;
 	Oid					funcrelid;
-	TupleDesc			tupdesc;
+	char				functyptype;
+	TupleDesc			tupdesc = NULL;
 
 	/*
 	 * FunctionScan should not have any children.
@@ -209,25 +199,36 @@ ExecInitFunctionScan(FunctionScan *node, EState *estate, Plan *parent)
 	rte = rt_fetch(node->scan.scanrelid, estate->es_range_table);
 	Assert(rte->rtekind == RTE_FUNCTION);
 	funcrettype = exprType(rte->funcexpr);
-	funcrelid = typeidTypeRelid(funcrettype);
+
+	/*
+	 * Now determine if the function returns a simple or composite type,
+	 * and check/add column aliases.
+	 */
+	functyptype = typeid_get_typtype(funcrettype);
 
 	/*
 	 * Build a suitable tupledesc representing the output rows
 	 */
-	if (OidIsValid(funcrelid))
+	if (functyptype == 'c')
 	{
-		/*
-		 * Composite data type, i.e. a table's row type
-		 * Same as ordinary relation RTE
-		 */
-		Relation	rel;
+		funcrelid = typeidTypeRelid(funcrettype);
+		if (OidIsValid(funcrelid))
+		{
+			/*
+			 * Composite data type, i.e. a table's row type
+			 * Same as ordinary relation RTE
+			 */
+			Relation	rel;
 
-		rel = relation_open(funcrelid, AccessShareLock);
-		tupdesc = CreateTupleDescCopy(RelationGetDescr(rel));
-		relation_close(rel, AccessShareLock);
-		scanstate->returnsTuple = true;
+			rel = relation_open(funcrelid, AccessShareLock);
+			tupdesc = CreateTupleDescCopy(RelationGetDescr(rel));
+			relation_close(rel, AccessShareLock);
+			scanstate->returnsTuple = true;
+		}
+		else
+			elog(ERROR, "Invalid return relation specified for function");
 	}
-	else
+	else if (functyptype == 'b')
 	{
 		/*
 		 * Must be a base data type, i.e. scalar
@@ -244,6 +245,21 @@ ExecInitFunctionScan(FunctionScan *node, EState *estate, Plan *parent)
 						   false);
 		scanstate->returnsTuple = false;
 	}
+	else if (functyptype == 'p' && funcrettype == RECORDOID)
+	{
+		/*
+		 * Must be a pseudo type, i.e. record
+		 */
+		List *coldeflist = rte->coldeflist;
+
+		tupdesc = BuildDescForRelation(coldeflist);
+		scanstate->returnsTuple = true;
+	}
+	else
+		elog(ERROR, "Unknown kind of return type specified for function");
+
+	scanstate->fn_typeid = funcrettype;
+	scanstate->fn_typtype = functyptype;
 	scanstate->tupdesc = tupdesc;
 	ExecSetSlotDescriptor(scanstate->csstate.css_ScanTupleSlot,
 						  tupdesc, false);
@@ -404,17 +420,20 @@ ExecFunctionReScan(FunctionScan *node, ExprContext *exprCtxt, Plan *parent)
  * Run the underlying function to get the next tuple
  */
 static TupleTableSlot *
-function_getonetuple(TupleTableSlot *slot,
-					 Node *expr,
-					 ExprContext *econtext,
-					 TupleDesc tupdesc,
-					 bool returnsTuple,
+function_getonetuple(FunctionScanState *scanstate,
 					 bool *isNull,
 					 ExprDoneCond *isDone)
 {
-	HeapTuple			tuple;
-	Datum				retDatum;
-	char				nullflag;
+	HeapTuple		tuple;
+	Datum			retDatum;
+	char			nullflag;
+	TupleDesc		tupdesc = scanstate->tupdesc;
+	bool			returnsTuple = scanstate->returnsTuple;
+	Node		   *expr = scanstate->funcexpr;
+	Oid				fn_typeid = scanstate->fn_typeid;
+	char			fn_typtype = scanstate->fn_typtype;
+	ExprContext	   *econtext = scanstate->csstate.cstate.cs_ExprContext;
+	TupleTableSlot *slot = scanstate->csstate.css_ScanTupleSlot;
 
 	/*
 	 * get the next Datum from the function
@@ -435,6 +454,16 @@ function_getonetuple(TupleTableSlot *slot,
 			 * function returns pointer to tts??
 			 */
 			slot = (TupleTableSlot *) retDatum;
+
+			/*
+			 * if function return type was RECORD, we need to check to be
+			 * sure the structure from the query matches the actual return
+			 * structure
+			 */
+			if (fn_typtype == 'p' && fn_typeid == RECORDOID)
+				if (tupledesc_mismatch(tupdesc, slot->ttc_tupleDescriptor))
+					elog(ERROR, "Query specified return tuple and actual"
+									" function return tuple do not match");
 		}
 		else
 		{
@@ -467,3 +496,26 @@ get_functionmode(Node *expr)
 	 */
 	return PM_REPEATEDCALL;
 }
+
+static bool
+tupledesc_mismatch(TupleDesc tupdesc1, TupleDesc tupdesc2)
+{
+	int			i;
+
+	if (tupdesc1->natts != tupdesc2->natts)
+		return true;
+
+	for (i = 0; i < tupdesc1->natts; i++)
+	{
+		Form_pg_attribute attr1 = tupdesc1->attrs[i];
+		Form_pg_attribute attr2 = tupdesc2->attrs[i];
+
+		/*
+		 * We really only care about number of attributes and data type
+		 */
+		if (attr1->atttypid != attr2->atttypid)
+			return true;
+	}
+
+	return false;
+}

src/backend/nodes/copyfuncs.c

@@ -15,7 +15,7 @@
  * Portions Copyright (c) 1994, Regents of the University of California
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/nodes/copyfuncs.c,v 1.199 2002/08/04 04:31:44 momjian Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/nodes/copyfuncs.c,v 1.200 2002/08/04 19:48:09 momjian Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -1482,6 +1482,7 @@ _copyRangeTblEntry(RangeTblEntry *from)
 	newnode->relid = from->relid;
 	Node_Copy(from, newnode, subquery);
 	Node_Copy(from, newnode, funcexpr);
+	Node_Copy(from, newnode, coldeflist);
 	newnode->jointype = from->jointype;
 	Node_Copy(from, newnode, joinaliasvars);
 	Node_Copy(from, newnode, alias);
@@ -1707,6 +1708,7 @@ _copyRangeFunction(RangeFunction *from)
 
 	Node_Copy(from, newnode, funccallnode);
 	Node_Copy(from, newnode, alias);
+	Node_Copy(from, newnode, coldeflist);
 
 	return newnode;
 }

src/backend/nodes/equalfuncs.c

@@ -20,7 +20,7 @@
  * Portions Copyright (c) 1994, Regents of the University of California
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/nodes/equalfuncs.c,v 1.146 2002/08/04 04:31:44 momjian Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/nodes/equalfuncs.c,v 1.147 2002/08/04 19:48:09 momjian Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -1607,6 +1607,8 @@ _equalRangeVar(RangeVar *a, RangeVar *b)
 		return false;
 	if (!equal(a->alias, b->alias))
 		return false;
+	if (!equal(a->coldeflist, b->coldeflist))
+		return false;
 
 	return true;
 }
@@ -1742,6 +1744,8 @@ _equalRangeTblEntry(RangeTblEntry *a, RangeTblEntry *b)
 		return false;
 	if (!equal(a->funcexpr, b->funcexpr))
 		return false;
+	if (!equal(a->coldeflist, b->coldeflist))
+		return false;
 	if (a->jointype != b->jointype)
 		return false;
 	if (!equal(a->joinaliasvars, b->joinaliasvars))

src/backend/nodes/outfuncs.c

@@ -5,7 +5,7 @@
  * Portions Copyright (c) 1996-2002, PostgreSQL Global Development Group
  * Portions Copyright (c) 1994, Regents of the University of California
  *
- *	$Header: /cvsroot/pgsql/src/backend/nodes/outfuncs.c,v 1.165 2002/07/18 17:14:19 momjian Exp $
+ *	$Header: /cvsroot/pgsql/src/backend/nodes/outfuncs.c,v 1.166 2002/08/04 19:48:09 momjian Exp $
  *
  * NOTES
  *	  Every (plan) node in POSTGRES has an associated "out" routine which
@@ -1004,6 +1004,8 @@ _outRangeTblEntry(StringInfo str, RangeTblEntry *node)
 		case RTE_FUNCTION:
 			appendStringInfo(str, ":funcexpr ");
 			_outNode(str, node->funcexpr);
+			appendStringInfo(str, ":coldeflist ");
+			_outNode(str, node->coldeflist);
 			break;
 		case RTE_JOIN:
 			appendStringInfo(str, ":jointype %d :joinaliasvars ",

src/backend/nodes/readfuncs.c

@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/nodes/readfuncs.c,v 1.126 2002/07/18 17:14:19 momjian Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/nodes/readfuncs.c,v 1.127 2002/08/04 19:48:09 momjian Exp $
  *
  * NOTES
  *	  Most of the read functions for plan nodes are tested. (In fact, they
@@ -1545,6 +1545,10 @@ _readRangeTblEntry(void)
 		case RTE_FUNCTION:
 			token = pg_strtok(&length); /* eat :funcexpr */
 			local_node->funcexpr = nodeRead(true);		/* now read it */
+
+			token = pg_strtok(&length); /* eat :coldeflist */
+			local_node->coldeflist = nodeRead(true);	/* now read it */
+
 			break;
 
 		case RTE_JOIN:

src/backend/parser/gram.y

@@ -11,7 +11,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/parser/gram.y,v 2.354 2002/08/04 06:51:23 thomas Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/parser/gram.y,v 2.355 2002/08/04 19:48:09 momjian Exp $
  *
  * HISTORY
  *	  AUTHOR			DATE			MAJOR EVENT
@@ -215,7 +215,8 @@ static void doNegateFloat(Value *v);
 				target_list, update_target_list, insert_column_list,
 				insert_target_list, def_list, opt_indirection,
 				group_clause, TriggerFuncArgs, select_limit,
-				opt_select_limit, opclass_item_list, trans_options
+				opt_select_limit, opclass_item_list, trans_options,
+				tableFuncElementList
 
 %type <range>	into_clause, OptTempTableName
 
@@ -256,8 +257,8 @@ static void doNegateFloat(Value *v);
 
 %type <vsetstmt> set_rest
 
-%type <node>	OptTableElement, ConstraintElem
-%type <node>	columnDef
+%type <node>	OptTableElement, ConstraintElem, tableFuncElement
+%type <node>	columnDef, tableFuncColumnDef
 %type <defelt>	def_elem
 %type <node>	def_arg, columnElem, where_clause, insert_column_item,
 				a_expr, b_expr, c_expr, r_expr, AexprConst,
@@ -4448,6 +4449,34 @@ table_ref:	relation_expr
 				{
 					RangeFunction *n = makeNode(RangeFunction);
 					n->funccallnode = $1;
+					n->coldeflist = NIL;
+					$$ = (Node *) n;
+				}
+			| func_table AS '(' tableFuncElementList ')'
+				{
+					RangeFunction *n = makeNode(RangeFunction);
+					n->funccallnode = $1;
+					n->coldeflist = $4;
+					$$ = (Node *) n;
+				}
+			| func_table AS ColId '(' tableFuncElementList ')'
+				{
+					RangeFunction *n = makeNode(RangeFunction);
+					Alias *a = makeNode(Alias);
+					n->funccallnode = $1;
+					a->aliasname = $3;
+					n->alias = a;
+					n->coldeflist = $5;
+					$$ = (Node *) n;
+				}
+			| func_table ColId '(' tableFuncElementList ')'
+				{
+					RangeFunction *n = makeNode(RangeFunction);
+					Alias *a = makeNode(Alias);
+					n->funccallnode = $1;
+					a->aliasname = $2;
+					n->alias = a;
+					n->coldeflist = $4;
 					$$ = (Node *) n;
 				}
 			| func_table alias_clause
@@ -4455,6 +4484,7 @@ table_ref:	relation_expr
 					RangeFunction *n = makeNode(RangeFunction);
 					n->funccallnode = $1;
 					n->alias = $2;
+					n->coldeflist = NIL;
 					$$ = (Node *) n;
 				}
 			| select_with_parens
@@ -4703,6 +4733,39 @@ where_clause:
 		;
 
 
+tableFuncElementList:
+			tableFuncElementList ',' tableFuncElement
+				{
+					if ($3 != NULL)
+						$$ = lappend($1, $3);
+					else
+						$$ = $1;
+				}
+			| tableFuncElement
+				{
+					if ($1 != NULL)
+						$$ = makeList1($1);
+					else
+						$$ = NIL;
+				}
+			| /*EMPTY*/							{ $$ = NIL; }
+		;
+
+tableFuncElement:
+			tableFuncColumnDef					{ $$ = $1; }
+		;
+
+tableFuncColumnDef:	ColId Typename
+				{
+					ColumnDef *n = makeNode(ColumnDef);
+					n->colname = $1;
+					n->typename = $2;
+					n->constraints = NIL;
+
+					$$ = (Node *)n;
+				}
+		;
+
 /*****************************************************************************
  *
  *	Type syntax

src/backend/parser/parse_clause.c

@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/parser/parse_clause.c,v 1.94 2002/06/20 20:29:32 momjian Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/parser/parse_clause.c,v 1.95 2002/08/04 19:48:10 momjian Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -515,7 +515,7 @@ transformRangeFunction(ParseState *pstate, RangeFunction *r)
 	 * OK, build an RTE for the function.
 	 */
 	rte = addRangeTableEntryForFunction(pstate, funcname, funcexpr,
-										r->alias, true);
+										r, true);
 
 	/*
 	 * We create a RangeTblRef, but we do not add it to the joinlist or

src/include/catalog/catversion.h

@@ -37,7 +37,7 @@
  * Portions Copyright (c) 1996-2002, PostgreSQL Global Development Group
  * Portions Copyright (c) 1994, Regents of the University of California
  *
- * $Id: catversion.h,v 1.144 2002/08/02 18:15:09 tgl Exp $
+ * $Id: catversion.h,v 1.145 2002/08/04 19:48:10 momjian Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -53,6 +53,6 @@
  */
 
 /*							yyyymmddN */
-#define CATALOG_VERSION_NO	200208011
+#define CATALOG_VERSION_NO	200208041
 
 #endif

src/include/catalog/pg_type.h

@@ -8,7 +8,7 @@
  * Portions Copyright (c) 1996-2002, PostgreSQL Global Development Group
  * Portions Copyright (c) 1994, Regents of the University of California
  *
- * $Id: pg_type.h,v 1.125 2002/07/24 19:11:13 petere Exp $
+ * $Id: pg_type.h,v 1.126 2002/08/04 19:48:10 momjian Exp $
  *
  * NOTES
  *	  the genbki.sh script reads this file and generates .bki
@@ -60,10 +60,10 @@ CATALOG(pg_type) BOOTSTRAP
 	bool		typbyval;
 
 	/*
-	 * typtype is 'b' for a basic type and 'c' for a catalog type (ie a
-	 * class). If typtype is 'c', typrelid is the OID of the class' entry
-	 * in pg_class. (Why do we need an entry in pg_type for classes,
-	 * anyway?)
+	 * typtype is 'b' for a basic type, 'c' for a catalog type (ie a
+	 * class), or 'p' for a pseudo type. If typtype is 'c', typrelid is the
+	 * OID of the class' entry in pg_class. (Why do we need an entry in
+	 * pg_type for classes, anyway?)
 	 */
 	char		typtype;
 
@@ -501,6 +501,16 @@ DATA(insert OID = 2209 ( _regoperator  PGNSP PGUID -1 f b t \054 0 2204 array_in
 DATA(insert OID = 2210 ( _regclass     PGNSP PGUID -1 f b t \054 0 2205 array_in array_out i x f 0 -1 0 _null_ _null_ ));
 DATA(insert OID = 2211 ( _regtype      PGNSP PGUID -1 f b t \054 0 2206 array_in array_out i x f 0 -1 0 _null_ _null_ ));
 
+/*
+ * pseudo-types 
+ *
+ * types with typtype='p' are special types that represent classes of types
+ * that are not easily defined in advance. Currently there is only one pseudo
+ * type -- record. The record type is used to specify that the value is a
+ * tuple, but of unknown structure until runtime. 
+ */
+DATA(insert OID = 2249 ( record        PGNSP PGUID  4 t p t \054 0 0 oidin oidout          i p f 0 -1 0 _null_ _null_ ));
+#define RECORDOID		2249
 
 /*
  * prototypes for functions in pg_type.c

src/include/nodes/execnodes.h

@@ -7,7 +7,7 @@
  * Portions Copyright (c) 1996-2002, PostgreSQL Global Development Group
  * Portions Copyright (c) 1994, Regents of the University of California
  *
- * $Id: execnodes.h,v 1.70 2002/06/20 20:29:49 momjian Exp $
+ * $Id: execnodes.h,v 1.71 2002/08/04 19:48:10 momjian Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -509,11 +509,17 @@ typedef struct SubqueryScanState
  *		Function nodes are used to scan the results of a
  *		function appearing in FROM (typically a function returning set).
  *
- *		functionmode			function operating mode:
+ *		functionmode		function operating mode:
  *							- repeated call
  *							- materialize
  *							- return query
+ *		tupdesc				function's return tuple description
  *		tuplestorestate		private state of tuplestore.c
+ *		funcexpr			function expression being evaluated
+ *		returnsTuple		does function return tuples?
+ *		fn_typeid			OID of function return type
+ *		fn_typtype			return Datum type, i.e. 'b'ase,
+ *							'c'atalog, or 'p'seudo
  * ----------------
  */
 typedef enum FunctionMode
@@ -525,12 +531,14 @@ typedef enum FunctionMode
 
 typedef struct FunctionScanState
 {
-	CommonScanState csstate;	/* its first field is NodeTag */
+	CommonScanState csstate;		/* its first field is NodeTag */
 	FunctionMode	functionmode;
 	TupleDesc		tupdesc;
 	void		   *tuplestorestate;
-	Node		   *funcexpr;	/* function expression being evaluated */
-	bool			returnsTuple; /* does function return tuples? */
+	Node		   *funcexpr;
+	bool			returnsTuple;
+	Oid				fn_typeid;
+	char			fn_typtype;
 } FunctionScanState;
 
 /* ----------------------------------------------------------------

src/include/nodes/parsenodes.h

@@ -7,7 +7,7 @@
  * Portions Copyright (c) 1996-2002, PostgreSQL Global Development Group
  * Portions Copyright (c) 1994, Regents of the University of California
  *
- * $Id: parsenodes.h,v 1.197 2002/08/04 04:31:44 momjian Exp $
+ * $Id: parsenodes.h,v 1.198 2002/08/04 19:48:10 momjian Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -400,6 +400,8 @@ typedef struct RangeFunction
 	NodeTag		type;
 	Node	   *funccallnode;	/* untransformed function call tree */
 	Alias	   *alias;			/* table alias & optional column aliases */
+	List	   *coldeflist;		/* list of ColumnDef nodes for runtime
+								 * assignment of RECORD TupleDesc */
 } RangeFunction;
 
 /*
@@ -527,6 +529,8 @@ typedef struct RangeTblEntry
 	 * Fields valid for a function RTE (else NULL):
 	 */
 	Node	   *funcexpr;		/* expression tree for func call */
+	List	   *coldeflist;		/* list of ColumnDef nodes for runtime
+								 * assignment of RECORD TupleDesc */
 
 	/*
 	 * Fields valid for a join RTE (else NULL/zero):

src/include/parser/parse_relation.h

@@ -7,7 +7,7 @@
  * Portions Copyright (c) 1996-2002, PostgreSQL Global Development Group
  * Portions Copyright (c) 1994, Regents of the University of California
  *
- * $Id: parse_relation.h,v 1.35 2002/08/02 18:15:09 tgl Exp $
+ * $Id: parse_relation.h,v 1.36 2002/08/04 19:48:11 momjian Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -44,7 +44,7 @@ extern RangeTblEntry *addRangeTableEntryForSubquery(ParseState *pstate,
 extern RangeTblEntry *addRangeTableEntryForFunction(ParseState *pstate,
 													char *funcname,
 													Node *funcexpr,
-													Alias *alias,
+													RangeFunction *rangefunc,
 													bool inFromCl);
 extern RangeTblEntry *addRangeTableEntryForJoin(ParseState *pstate,
 						  List *colnames,
@@ -61,5 +61,6 @@ extern List *expandRelAttrs(ParseState *pstate, RangeTblEntry *rte);
 extern int	attnameAttNum(Relation rd, const char *attname, bool sysColOK);
 extern Name attnumAttName(Relation rd, int attid);
 extern Oid	attnumTypeId(Relation rd, int attid);
+extern char typeid_get_typtype(Oid typeid);
 
 #endif   /* PARSE_RELATION_H */

src/test/regress/expected/type_sanity.out

@@ -16,7 +16,7 @@
 SELECT p1.oid, p1.typname
 FROM pg_type as p1
 WHERE (p1.typlen <= 0 AND p1.typlen != -1) OR
-    (p1.typtype != 'b' AND p1.typtype != 'c') OR
+    (p1.typtype != 'b' AND p1.typtype != 'c' AND p1.typtype != 'p') OR
     NOT p1.typisdefined OR
     (p1.typalign != 'c' AND p1.typalign != 's' AND
      p1.typalign != 'i' AND p1.typalign != 'd') OR
@@ -60,7 +60,7 @@ WHERE (p1.typtype = 'c' AND p1.typrelid = 0) OR
 -- NOTE: as of 7.3, this check finds SET, smgr, and unknown.
 SELECT p1.oid, p1.typname
 FROM pg_type as p1
-WHERE p1.typtype != 'c' AND p1.typname NOT LIKE '\\_%' AND NOT EXISTS
+WHERE p1.typtype = 'b' AND p1.typname NOT LIKE '\\_%' AND NOT EXISTS
     (SELECT 1 FROM pg_type as p2
      WHERE p2.typname = ('_' || p1.typname)::name AND
            p2.typelem = p1.oid);

src/test/regress/sql/type_sanity.sql

@@ -19,7 +19,7 @@
 SELECT p1.oid, p1.typname
 FROM pg_type as p1
 WHERE (p1.typlen <= 0 AND p1.typlen != -1) OR
-    (p1.typtype != 'b' AND p1.typtype != 'c') OR
+    (p1.typtype != 'b' AND p1.typtype != 'c' AND p1.typtype != 'p') OR
     NOT p1.typisdefined OR
     (p1.typalign != 'c' AND p1.typalign != 's' AND
      p1.typalign != 'i' AND p1.typalign != 'd') OR
@@ -55,7 +55,7 @@ WHERE (p1.typtype = 'c' AND p1.typrelid = 0) OR
 
 SELECT p1.oid, p1.typname
 FROM pg_type as p1
-WHERE p1.typtype != 'c' AND p1.typname NOT LIKE '\\_%' AND NOT EXISTS
+WHERE p1.typtype = 'b' AND p1.typname NOT LIKE '\\_%' AND NOT EXISTS
     (SELECT 1 FROM pg_type as p2
      WHERE p2.typname = ('_' || p1.typname)::name AND
            p2.typelem = p1.oid);