[postgis-devel] GBOXF

[Mark Cave-Ayland]

On 28/11/11 18:02, Sandro Santilli wrote:

>> I will move forward with new bbox work on the above basis, and abandon
>> the doublebox experiment.
>
> Sounds good to me.

As per my previous email, I'm not yet convinced that the tests aren't 
triggering some other bug/code path causing an unusually high slowdown. 
However, we're working blind until we get detailed profiling information.

> Of course a new type will still pose the "float or double" question.
> As well as it was posed (they tell me) by ST_Extent.

Indeed. My main argument for using doubles instead of floats is because 
at the moment, we inflate the bounding box from double to float. This 
means that we can categorically state that a point *outside* of the 
float bounding box is definitely not within the real bounding box of the 
geometry, but we cannot say the inverse, i.e. a point *inside* the float 
bounding box may or may not within the real bounding box of the geometry.

My feeling is that if we use floats then we minimise the usefulness of 
the cache because it reduces the number of use cases that it could be 
used to optimise (and I'm fairly sure that some functions will assume 
that the second predicate in the paragraph above is true, and therefore 
give wrong answers in certain cases).


ATB,

Mark.

-- 
Mark Cave-Ayland - Senior Technical Architect
PostgreSQL - PostGIS
Sirius Corporation plc - control through freedom
http://www.siriusit.co.uk
t: +44 870 608 0063

Sirius Labs: http://www.siriusit.co.uk/labs