Re-allow DISTINCT in pl/pgsql expressions.

I'd omitted this from the grammar in commit c9d529848, figuring that
it wasn't worth supporting.  However we already have one complaint,
so it seems that judgment was wrong.  It doesn't require a huge
amount of code, so add it back.  (I'm still drawing the line at
UNION/INTERSECT/EXCEPT though: those'd require an unreasonable
amount of grammar refactoring, and the single-result-row restriction
makes them near useless anyway.)

Also rethink the documentation: this behavior is a property of
all pl/pgsql expressions, not just assignments.

Discussion: https://postgr.es/m/20210122134106.e94c5cd7@mail.verfriemelt.org

doc/src/sgml/plpgsql.sgml

@@ -917,6 +917,24 @@ PREPARE <replaceable>statement_name</replaceable>(integer, integer) AS SELECT $1
      they are useful to know when trying to diagnose a problem.
      More information appears in <xref linkend="plpgsql-plan-caching"/>.
     </para>
+
+    <para>
+     Since an <replaceable>expression</replaceable> is converted to a
+     <literal>SELECT</literal> command, it can contain the same clauses
+     that an ordinary <literal>SELECT</literal> would, except that it
+     cannot include a top-level <literal>UNION</literal>,
+     <literal>INTERSECT</literal>, or <literal>EXCEPT</literal> clause.
+     Thus for example one could test whether a table is non-empty with
+<programlisting>
+IF count(*) &gt; 0 FROM my_table THEN ...
+</programlisting>
+     since the <replaceable>expression</replaceable>
+     between <literal>IF</literal> and <literal>THEN</literal> is parsed as
+     though it were <literal>SELECT count(*) &gt; 0 FROM my_table</literal>.
+     The <literal>SELECT</literal> must produce a single column, and not
+     more than one row.  (If it produces no rows, the result is taken as
+     NULL.)
+    </para>
   </sect1>
 
   <sect1 id="plpgsql-statements">
@@ -973,20 +991,6 @@ my_array[1:3] := array[1,2,3];
 complex_array[n].realpart = 12.3;
 </programlisting>
     </para>
-
-    <para>
-     It's useful to know that what follows the assignment operator is
-     essentially treated as a <literal>SELECT</literal> command; as long
-     as it returns a single row and column, it will work.  Thus for example
-     one can write something like
-<programlisting>
-total_sales := sum(quantity) from sales;
-</programlisting>
-     This provides an effect similar to the single-row <literal>SELECT
-     ... INTO</literal> syntax described in
-     <xref linkend="plpgsql-statements-sql-onerow"/>.  However, that syntax
-     is more portable.
-    </para>
    </sect2>
 
    <sect2 id="plpgsql-statements-sql-noresult">

src/backend/parser/analyze.c

@@ -2466,7 +2466,7 @@ transformPLAssignStmt(ParseState *pstate, PLAssignStmt *stmt)
 
 	/*
 	 * The rest mostly matches transformSelectStmt, except that we needn't
-	 * consider WITH or DISTINCT, and we build a targetlist our own way.
+	 * consider WITH or INTO, and we build a targetlist our own way.
 	 */
 	qry->commandType = CMD_SELECT;
 	pstate->p_is_insert = false;
@@ -2590,10 +2590,29 @@ transformPLAssignStmt(ParseState *pstate, PLAssignStmt *stmt)
 											EXPR_KIND_GROUP_BY,
 											false /* allow SQL92 rules */ );
 
-	/* No DISTINCT clause */
+	if (sstmt->distinctClause == NIL)
-	Assert(!sstmt->distinctClause);
+	{
 		qry->distinctClause = NIL;
 		qry->hasDistinctOn = false;
+	}
+	else if (linitial(sstmt->distinctClause) == NULL)
+	{
+		/* We had SELECT DISTINCT */
+		qry->distinctClause = transformDistinctClause(pstate,
+													  &qry->targetList,
+													  qry->sortClause,
+													  false);
+		qry->hasDistinctOn = false;
+	}
+	else
+	{
+		/* We had SELECT DISTINCT ON */
+		qry->distinctClause = transformDistinctOnClause(pstate,
+														sstmt->distinctClause,
+														&qry->targetList,
+														qry->sortClause);
+		qry->hasDistinctOn = true;
+	}
 
 	/* transform LIMIT */
 	qry->limitOffset = transformLimitClause(pstate, sstmt->limitOffset,

src/backend/parser/gram.y

@@ -389,7 +389,7 @@ static Node *makeRecursiveViewSelect(char *relname, List *aliases, Node *query);
 				OptTableElementList TableElementList OptInherit definition
 				OptTypedTableElementList TypedTableElementList
 				reloptions opt_reloptions
-				OptWith distinct_clause opt_definition func_args func_args_list
+				OptWith opt_definition func_args func_args_list
 				func_args_with_defaults func_args_with_defaults_list
 				aggr_args aggr_args_list
 				func_as createfunc_opt_list alterfunc_opt_list
@@ -401,6 +401,7 @@ static Node *makeRecursiveViewSelect(char *relname, List *aliases, Node *query);
 				name_list role_list from_clause from_list opt_array_bounds
 				qualified_name_list any_name any_name_list type_name_list
 				any_operator expr_list attrs
+				distinct_clause opt_distinct_clause
 				target_list opt_target_list insert_column_list set_target_list
 				set_clause_list set_clause
 				def_list operator_def_list indirection opt_indirection
@@ -11260,6 +11261,11 @@ select_clause:
  * As with select_no_parens, simple_select cannot have outer parentheses,
  * but can have parenthesized subclauses.
  *
+ * It might appear that we could fold the first two alternatives into one
+ * by using opt_distinct_clause.  However, that causes a shift/reduce conflict
+ * against INSERT ... SELECT ... ON CONFLICT.  We avoid the ambiguity by
+ * requiring SELECT DISTINCT [ON] to be followed by a non-empty target_list.
+ *
  * Note that sort clauses cannot be included at this level --- SQL requires
  *		SELECT foo UNION SELECT bar ORDER BY baz
  * to be parsed as
@@ -11497,8 +11503,13 @@ opt_all_clause:
 			| /*EMPTY*/
 		;
 
+opt_distinct_clause:
+			distinct_clause							{ $$ = $1; }
+			| opt_all_clause						{ $$ = NIL; }
+		;
+
 opt_sort_clause:
-			sort_clause								{ $$ = $1;}
+			sort_clause								{ $$ = $1; }
 			| /*EMPTY*/								{ $$ = NIL; }
 		;
 
@@ -15065,32 +15076,33 @@ role_list:	RoleSpec
  * Therefore the returned struct is a SelectStmt.
  *****************************************************************************/
 
-PLpgSQL_Expr: opt_target_list
+PLpgSQL_Expr: opt_distinct_clause opt_target_list
 			from_clause where_clause
 			group_clause having_clause window_clause
 			opt_sort_clause opt_select_limit opt_for_locking_clause
 				{
 					SelectStmt *n = makeNode(SelectStmt);
 
-					n->targetList = $1;
+					n->distinctClause = $1;
-					n->fromClause = $2;
+					n->targetList = $2;
-					n->whereClause = $3;
+					n->fromClause = $3;
-					n->groupClause = $4;
+					n->whereClause = $4;
-					n->havingClause = $5;
+					n->groupClause = $5;
-					n->windowClause = $6;
+					n->havingClause = $6;
-					n->sortClause = $7;
+					n->windowClause = $7;
-					if ($8)
+					n->sortClause = $8;
-					{
+					if ($9)
-						n->limitOffset = $8->limitOffset;
+					{
-						n->limitCount = $8->limitCount;
+						n->limitOffset = $9->limitOffset;
+						n->limitCount = $9->limitCount;
 						if (!n->sortClause &&
-							$8->limitOption == LIMIT_OPTION_WITH_TIES)
+							$9->limitOption == LIMIT_OPTION_WITH_TIES)
 							ereport(ERROR,
 									(errcode(ERRCODE_SYNTAX_ERROR),
 									 errmsg("WITH TIES cannot be specified without ORDER BY clause")));
-						n->limitOption = $8->limitOption;
+						n->limitOption = $9->limitOption;
 					}
-					n->lockingClause = $9;
+					n->lockingClause = $10;
 					$$ = (Node *) n;
 				}
 		;