I started to wonder why the test in range_gist_consistent_int() for
RANGESTRAT_CONTAINED_BY was "return true" (ie, search the entire index)
rather than range_overlaps, which is what is tested in the comparable
case in rtree_internal_consistent(). The regression tests showed me
how come: an empty-range index entry should be reported as being
contained by anything, but range_overlaps won't necessarily find such
entries. (The rtree case is all right because there is no equivalent of
an empty range in boxes, circles, or polygons.)
I am not satisfied with this state of affairs though: range_contained_by
really ought to be efficiently indexable, but with the current coding
an index search is nearly useless. Also, so far as I can see, the
current penalty function allows empty-range entries to be scattered
basically anywhere in the index, making a search for "= empty" pretty
inefficient too.
The first solution that comes to mind is to make the penalty and
picksplit functions forcibly segregate empty ranges from others, that is
a split will never put empty ranges together with non-empty ones. Then,
we can assume that a non-empty internal node doesn't represent any empty
leaf entries, and avoid descending to it when it doesn't overlap the
target range. Then the equality-search case could be improved too.
Thoughts, better ideas?
regards, tom lane