On Wed, Jun 18, 2014 at 8:51 AM, Heikki Linnakangas
<hlinnakangas@vmware.com> wrote:
>
> I liked Greg's sketch of what the opclass support functions would be. It
> doesn't seem significantly more complicated than what's in the patch now.
Which was
On Tue, Jun 17, 2014 at 8:48 PM, Greg Stark <stark@mit.edu> wrote:
> An aggregate to generate a min/max "bounding box" from several values
> A function which takes an "bounding box" and a new value and returns
> the new "bounding box"
> A function which tests if a value is in a "bounding box"
> A function which tests if a "bounding box" overlaps a "bounding box"
Which I'd generalize a bit further by renaming "bounding box" with
"compressed set", and allow it to be parameterized.
So, you have:
An aggregate to generate a "compressed set" from several values
A function which adds a new value to the "compressed set" and returns
the new "compressed set"
A function which tests if a value is in a "compressed set"
A function which tests if a "compressed set" overlaps another
"compressed set" of equal type
If you can define different compressed sets, you can use this to
generate both min/max indexes as well as bloom filter indexes. Whether
we'd want to have both is perhaps questionable, but having the ability
to is probably desirable.
One problem with such a generalized implementation would be, that I'm
not sure in-place modification of the "compressed set" on-disk can be
assumed to be safe on all cases. Surely, for strictly-enlarging sets
it would, but while min/max and bloom filters both fit the bill, it's
not clear that one can assume this for all structures.
Adding also a "in-place updateable" bit to the "type" would perhaps
inflate the complexity of the patch due to the need to provide both
code paths?
