Re: Fixing GIN for empty/null/full-scan cases

On Jan 18, 2011, at 3:46 PM, Tom Lane wrote:

> The above is "using" the index, but only as a guide to where the rows
> satisfying the partial-index predicate are --- note the lack of any
> index condition in the indexscan node.  That's because the query_int
> query is not in fact compatible with the core-provided index opclass.
> We get much better results using intarray's gin__int_ops opclass:

Ah-hah! Confirmed, thank you. I now get 23.2 ms and an index condition. Much, much better.

Thank you! And I think we can use the improved GIN indexes for this app, once 9.1 drops.

Best,

David