The ship has cleatly sailed to add parameterized opclasses to 9.4, but let's keep it in mind when we decide on the defaults.
In the absence of parameterizable opclasses, it would be much more flexible to have opclasses that index, keys, values, key-value pairs and paths separately, instead of the current json_ops and json_hash_ops opclasses which index all of those in the same index. That way, if you only e.g. ever query on the existence of a key, you'd only need to index the keys.
I don't understand how we ended up with the current dichotomy of json_ops and json_hash_ops...
