On Sat, 2020-03-21 at 18:26 -0700, Andres Freund wrote:
> I don't see how? That'd require making the hash bucket addressing
> deal
> with variable sizes, which'd be bad for performance reasons. Since
> there
> can be a aggstate->numtrans AggStatePerGroupDatas for each hash table
> entry, I don't see how to avoid a variable size?
It would not vary for a given hash table. Do you mean the compile-time
specialization (of simplehash.h) would not work any more?
If we aren't storing the "additional" inline in the hash entry, I don't
see any purpose for the argument to BuildTupleHashTableExt(), nor the
purpose of the "entrysize" field of TupleHashTableData.
> If we want to optimize memory usage, I think I'd first go for
> allocating
> the group's "firstTuple" together with all the AggStatePerGroupDatas.
That's a good idea.
Regards,
Jeff Davis