Jack Orenstein <jao@xxxxxxxxxxxx> writes: > I guess the top-level question is this: Is it possible in principle for the > btree index to subsume GIN index capabilities? Well, you could write a completely different index AM that happened to share the same on-disk representation, perhaps. But it would be a different AM. btree is built for scalar values and has none of the mechanisms that GIN does for breaking down input values into components-to-be-indexed, nor for analyzing complex query operators to find out what indexable search conditions are implied. regards, tom lane