I think that the PVC_RECURSE_WINDOWFUNCS flag shouldn't be used in
make_partial_grouping_target().

First, this function uses the grouping_target (see grouping_planner()) as the
input, and that should only contain the input expressions of window functions
as opposed to the window functions themselves. (make_window_input_target() is
responsible for pulling the input expressions from the window functions.)

Second, if a window function appeared in the result of
make_partial_grouping_target() for any reason, the Agg node would fail to
evaluate it. Am I wrong?

-- 
Antonin Houska
Web: https://www.cybertec-postgresql.com

diff --git a/src/backend/optimizer/plan/planner.c b/src/backend/optimizer/plan/planner.c
index e664eb18c0..c5b9e45098 100644
--- a/src/backend/optimizer/plan/planner.c
+++ b/src/backend/optimizer/plan/planner.c
@@ -5349,7 +5349,6 @@ make_partial_grouping_target(PlannerInfo *root,
 	 */
 	non_group_exprs = pull_var_clause((Node *) non_group_cols,
 									  PVC_INCLUDE_AGGREGATES |
-									  PVC_RECURSE_WINDOWFUNCS |
 									  PVC_INCLUDE_PLACEHOLDERS);
 
 	add_new_columns_to_pathtarget(partial_target, non_group_exprs);

Reply via email to