bug causing out-of-bounds memory access in AddUpdated when updating cached row status

[0xYashwanth]

bug causing out-of-bounds memory access in AddUpdated() when updating cached row status.

In the else if (upd_idx >= 0) branch at line 2691, the code updates res->updated_keyset[upd_idx].status but then incorrectly accesses res->added_tuples + num_fields * upd_add_idx to clear cached data.

When a row's status in the updated_keyset array is modified, the corresponding cached tuple data must be invalidated.
The invariant is: for any index i, the tuple cache entry is stored at base_array + num_fields * i. Since we're modifying updated_keyset[upd_idx], the corresponding tuple cache is at updated_tuples + num_fields * upd_idx.
Using the wrong base array (added_tuples) or wrong index (upd_add_idx = -1) violates this and produces undefined behavior.

Fix

tuple = res->updated_tuples + num_fields * upd_idx;

[davecramer]

is there a way to test this ?