On Thu, Sep 17, 2020 at 02:12:12PM -0300, Ranier Vilela wrote: >Hi, > >In case gd->any_hashable is FALSE, grouping_is_hashable is never called. >In this case, the planner could use HASH for groupings, but will never know. >
The condition is pretty simple - if the query has grouping sets, look at grouping sets, otherwise look at groupClause. For this to be an issue the query would need to have both grouping sets and (independent) group clause - which AFAIK is not possible.
Uh?
(parse->groupClause != NIL) If different from NIL we have ((independent) group clause), grouping_is_hashable should check?
For hashing to be worth considering, at least one grouping set has to be hashable - otherwise it's pointless.
Granted, you could have something like
GROUP BY GROUPING SETS ((a), (b)), c
which I think essentially says "add c to every grouping set" and that will be covered by the any_hashable check.
I am not going into the merit of whether or not it is worth hashing. grouping_is_hashable as a last resort, must decide.
>Apparently gd pointer, will never be NULL there, verified with Assert(gd != >NULL). >
Um, what? If you add the assert right before the if condition, you won't even be able to do initdb. It's pretty clear it'll crash for any query without grouping sets.
CID 1412604 (#1 of 1): Dereference after null check (FORWARD_NULL)13.var_deref_model:Passing null pointer gd to create_ordinary_grouping_paths, which dereferences it. [show details]