David Rowley <dgrowle...@gmail.com> writes:
> I've pushed the patch now.
So I tried adapting my patch to not make a copy of the input slot,
and it didn't work: I was still getting assertion failures about
the slot not being a MinimalTupleSlot as expected. On investigation
it appears your patch did not fully adjust BuildTupleHashTableExt
for variable input-slot type. You need the attached as well.
I'm not sure why the existing regression tests didn't catch this.
But it may not be worth searching for a test case, because my patch
will be one ...
regards, tom lane
diff --git a/src/backend/executor/execGrouping.c b/src/backend/executor/execGrouping.c
index 4a8f72305c..7491e53c03 100644
--- a/src/backend/executor/execGrouping.c
+++ b/src/backend/executor/execGrouping.c
@@ -236,9 +236,9 @@ BuildTupleHashTableExt(PlanState *parent,
hash_iv);
/* build comparator for all columns */
- /* XXX: should we support non-minimal tuples for the inputslot? */
hashtable->tab_eq_func = ExecBuildGroupingEqual(inputDesc, inputDesc,
- &TTSOpsMinimalTuple, &TTSOpsMinimalTuple,
+ inputOps,
+ &TTSOpsMinimalTuple,
numCols,
keyColIdx, eqfuncoids, collations,
allow_jit ? parent : NULL);
