On Mon, Mar 10, 2014 at 2:19 PM, Peter Geoghegan <pg@heroku.com> wrote:
On Mon, Mar 10, 2014 at 3:00 AM, Alexander Korotkov <aekorotkov@gmail.com> wrote: > I din't get comment about "leftmost" element. There is absolutely no > distinguish between array elements. All elements are extracted into same > keys independent of their indexes. It seems to have no change since I wrote > hstore_hash_ops. Could you share test case to illustrate what you mean?
I don't have time to post that at the moment, but offhand I *think* your confusion may be due to the fact that the json_hash_ops opclass (as I call it) was previously consistent with the behavior of the other GIN opclass (the default). The problem is that they (well, at least the default GIN and GiST opclasses) were inconsistent with how the containment operator behaved in respect of jsonb array elements generally.
Here is the commit on our feature branch where I fixed the problem for the default GIN opclass:
If it doesn't explain the problem, you may still wish to comment on the correctness of this fix. I am still waiting on feedback from Oleg and Teodor.
Apparently, there is bug in calculation of hashes. Array elements were hashed incrementally while each of them should be hashed separately. That cause an effect of distinguishing array elements by their indexes. Not sure about when this bug was added.