Fix parse_cte.c's failure to examine sub-WITHs in DML statements.

makeDependencyGraphWalker thought that only SelectStmt nodes could
contain a WithClause.  Which was true in our original implementation
of WITH, but astonishingly we missed updating this code when we added
the ability to attach WITH to INSERT/UPDATE/DELETE (and later MERGE).
Moreover, since it was coded to deliberately block recursion to a
WithClause, even updating raw_expression_tree_walker didn't save it.

The upshot of this was that we didn't see references to outer CTE
names appearing within an inner WITH, and would neither complain about
disallowed recursion nor account for such references when sorting CTEs
into a usable order.  The lack of complaints about this is perhaps not
so surprising, because typical usage of WITH wouldn't hit either case.
Still, it's pretty broken; failing to detect recursion here leads to
assert failures or worse later on.

Fix by factoring out the processing of sub-WITHs into a new function
WalkInnerWith, and invoking that for all the statement types that
can have WITH.

Bug: #18878
Reported-by: Yu Liang <luy70@psu.edu>
Author: Tom Lane <tgl@sss.pgh.pa.us>
Discussion: https://postgr.es/m/18878-a26fa5ab6be2f2cf@postgresql.org
Backpatch-through: 13

src/backend/parser/parse_cte.c

@@ -88,6 +88,7 @@ static void analyzeCTE(ParseState *pstate, CommonTableExpr *cte);
 /* Dependency processing functions */
 static void makeDependencyGraph(CteState *cstate);
 static bool makeDependencyGraphWalker(Node *node, CteState *cstate);
+static void WalkInnerWith(Node *stmt, WithClause *withClause, CteState *cstate);
 static void TopologicalSort(ParseState *pstate, CteItem *items, int numitems);
 
 /* Recursion validity checker functions */
@@ -731,26 +732,108 @@ makeDependencyGraphWalker(Node *node, CteState *cstate)
 	if (IsA(node, SelectStmt))
 	{
 		SelectStmt *stmt = (SelectStmt *) node;
-		ListCell   *lc;
 
 		if (stmt->withClause)
 		{
-			if (stmt->withClause->recursive)
+			/* Examine the WITH clause and the SelectStmt */
+			WalkInnerWith(node, stmt->withClause, cstate);
+			/* We're done examining the SelectStmt */
+			return false;
+		}
+		/* if no WITH clause, just fall through for normal processing */
+	}
+	else if (IsA(node, InsertStmt))
+	{
+		InsertStmt *stmt = (InsertStmt *) node;
+
+		if (stmt->withClause)
+		{
+			/* Examine the WITH clause and the InsertStmt */
+			WalkInnerWith(node, stmt->withClause, cstate);
+			/* We're done examining the InsertStmt */
+			return false;
+		}
+		/* if no WITH clause, just fall through for normal processing */
+	}
+	else if (IsA(node, DeleteStmt))
+	{
+		DeleteStmt *stmt = (DeleteStmt *) node;
+
+		if (stmt->withClause)
+		{
+			/* Examine the WITH clause and the DeleteStmt */
+			WalkInnerWith(node, stmt->withClause, cstate);
+			/* We're done examining the DeleteStmt */
+			return false;
+		}
+		/* if no WITH clause, just fall through for normal processing */
+	}
+	else if (IsA(node, UpdateStmt))
+	{
+		UpdateStmt *stmt = (UpdateStmt *) node;
+
+		if (stmt->withClause)
+		{
+			/* Examine the WITH clause and the UpdateStmt */
+			WalkInnerWith(node, stmt->withClause, cstate);
+			/* We're done examining the UpdateStmt */
+			return false;
+		}
+		/* if no WITH clause, just fall through for normal processing */
+	}
+	else if (IsA(node, MergeStmt))
+	{
+		MergeStmt  *stmt = (MergeStmt *) node;
+
+		if (stmt->withClause)
+		{
+			/* Examine the WITH clause and the MergeStmt */
+			WalkInnerWith(node, stmt->withClause, cstate);
+			/* We're done examining the MergeStmt */
+			return false;
+		}
+		/* if no WITH clause, just fall through for normal processing */
+	}
+	else if (IsA(node, WithClause))
 	{
 		/*
-				 * In the RECURSIVE case, all query names of the WITH are
-				 * visible to all WITH items as well as the main query. So
-				 * push them all on, process, pop them all off.
+		 * Prevent raw_expression_tree_walker from recursing directly into a
+		 * WITH clause.  We need that to happen only under the control of the
+		 * code above.
 		 */
-				cstate->innerwiths = lcons(stmt->withClause->ctes,
-										   cstate->innerwiths);
-				foreach(lc, stmt->withClause->ctes)
+		return false;
+	}
+	return raw_expression_tree_walker(node,
+									  makeDependencyGraphWalker,
+									  (void *) cstate);
+}
+
+/*
+ * makeDependencyGraphWalker's recursion into a statement having a WITH clause.
+ *
+ * This subroutine is concerned with updating the innerwiths list correctly
+ * based on the visibility rules for CTE names.
+ */
+static void
+WalkInnerWith(Node *stmt, WithClause *withClause, CteState *cstate)
+{
+	ListCell   *lc;
+
+	if (withClause->recursive)
+	{
+		/*
+		 * In the RECURSIVE case, all query names of the WITH are visible to
+		 * all WITH items as well as the main query.  So push them all on,
+		 * process, pop them all off.
+		 */
+		cstate->innerwiths = lcons(withClause->ctes, cstate->innerwiths);
+		foreach(lc, withClause->ctes)
 		{
 			CommonTableExpr *cte = (CommonTableExpr *) lfirst(lc);
 
 			(void) makeDependencyGraphWalker(cte->ctequery, cstate);
 		}
-				(void) raw_expression_tree_walker(node,
+		(void) raw_expression_tree_walker(stmt,
 										  makeDependencyGraphWalker,
 										  (void *) cstate);
 		cstate->innerwiths = list_delete_first(cstate->innerwiths);
@@ -758,11 +841,11 @@ makeDependencyGraphWalker(Node *node, CteState *cstate)
 	else
 	{
 		/*
-				 * In the non-RECURSIVE case, query names are visible to the
-				 * WITH items after them and to the main query.
+		 * In the non-RECURSIVE case, query names are visible to the WITH
+		 * items after them and to the main query.
 		 */
 		cstate->innerwiths = lcons(NIL, cstate->innerwiths);
-				foreach(lc, stmt->withClause->ctes)
+		foreach(lc, withClause->ctes)
 		{
 			CommonTableExpr *cte = (CommonTableExpr *) lfirst(lc);
 			ListCell   *cell1;
@@ -772,28 +855,11 @@ makeDependencyGraphWalker(Node *node, CteState *cstate)
 			cell1 = list_head(cstate->innerwiths);
 			lfirst(cell1) = lappend((List *) lfirst(cell1), cte);
 		}
-				(void) raw_expression_tree_walker(node,
+		(void) raw_expression_tree_walker(stmt,
 										  makeDependencyGraphWalker,
 										  (void *) cstate);
 		cstate->innerwiths = list_delete_first(cstate->innerwiths);
 	}
-			/* We're done examining the SelectStmt */
-			return false;
-		}
-		/* if no WITH clause, just fall through for normal processing */
-	}
-	if (IsA(node, WithClause))
-	{
-		/*
-		 * Prevent raw_expression_tree_walker from recursing directly into a
-		 * WITH clause.  We need that to happen only under the control of the
-		 * code above.
-		 */
-		return false;
-	}
-	return raw_expression_tree_walker(node,
-									  makeDependencyGraphWalker,
-									  (void *) cstate);
 }
 
 /*

src/test/regress/expected/with.out

@@ -2031,6 +2031,14 @@ WITH RECURSIVE x(n) AS (
 ERROR:  ORDER BY in a recursive query is not implemented
 LINE 3:   ORDER BY (SELECT n FROM x))
                    ^
+-- and this
+WITH RECURSIVE x(n) AS (
+  WITH sub_cte AS (SELECT * FROM x)
+  DELETE FROM graph RETURNING f)
+	SELECT * FROM x;
+ERROR:  recursive query "x" must not contain data-modifying statements
+LINE 1: WITH RECURSIVE x(n) AS (
+                       ^
 CREATE TEMPORARY TABLE y (a INTEGER);
 INSERT INTO y SELECT generate_series(1, 10);
 -- LEFT JOIN

src/test/regress/sql/with.sql

@@ -930,6 +930,13 @@ WITH RECURSIVE x(n) AS (
   ORDER BY (SELECT n FROM x))
 	SELECT * FROM x;
 
+-- and this
+WITH RECURSIVE x(n) AS (
+  WITH sub_cte AS (SELECT * FROM x)
+  DELETE FROM graph RETURNING f)
+	SELECT * FROM x;
+
+
 CREATE TEMPORARY TABLE y (a INTEGER);
 INSERT INTO y SELECT generate_series(1, 10);