Robert Haas <robertmhaas@gmail.com> writes:
> I'm having trouble figuring out under what set of circumstances spgist
> is expected to be the best available alternative. It only supports a
> small subset of the data types that GiST does, so I suppose the point
> is that it should be faster for the cases that it does handle. And,
> considering that this is all brand-new code, the fact that it's almost
> keeping pace with btree on both pattern-matching and equality
> comparisons is certainly respectable -- but I so far haven't found any
> cases where it's a clear win. There's limited glory in being the
> almost-fastest way of indexing for a certain class of queries.
> Admittedly, I haven't tried the point-in-box stuff yet.
I think the text opclass is mostly meant as sample code/proof-of-concept.
What spgist is likely to be good for is queries that btree can't cope
with at all. But it's cute that we can make it handle LIKE for a few
more lines of code, so I'm fine with adding that.
regards, tom lane
