Summary of this thread so far:
There was a lot of discussion comparing this with Tomas Vondra's Hash
Join patch. The conclusion was that while it would be nice to be able to
dump transition state to disk, for aggregates like array_agg, the patch
is fine as it is. Dumping transition states would require much more
work, and this is already useful without it. Moreover, solving the
array_agg problem later won't require a rewrite; rather, it'll build on
top of this.
You listed a number of open items in the original post, and these are
still outstanding:
> * costing
> * EXPLAIN details for disk usage
> * choose number of partitions intelligently
> * performance testing
I think this is enough for this commitfest - we have consensus on the
design. For the next one, please address those open items, and resubmit.
- Heikki