[postgis-devel] Geodetic Plan

[Paul Ramsey]

On Thu, Jun 18, 2009 at 11:43 PM, <nicklas.aven at jordogskog.no> wrote:
> I have a thought about indexing. Idon't know much about them but I have read some headings from the diskussions about deviding features to get more out of the geometries.
>
> I guess the issues about comparing big to small will be more common when handling "the whole world".
> Wouldn't it be possible to just devide the index.

Never use the word "just".

What you're talking about is basically a "grid index" and there's not
really a "nice" way to implement one. Grid indexes have positive
attributes (they can do things like index really big features more
selectively than box-based indexes, as you note) and negative indexes
(they are not self-tuning).

And once you've pulled your phase-one result set out using the index,
you still need to run explicit tests on it to see if the exact spatial
conditions are met. I think it's better to just establish a best
practice for any spatial database / processing regime (and this is
just as true for 2D and 3D) which is "thou shalt not store objects
that are a significant proportion of your overall domain". If you have
such an object, slice it up *before* you store it.

P.