[postgis-devel] Boxes cache and type (was: Caching Double-based Boxes)

[Mateusz Łoskot]

On 25 November 2011 09:38, Nicklas Avén <nicklas.aven at jordogskog.no> wrote:
> In my example I posted in last post, the difference is 0.3 meters! It is
> no microunits we are talking about.
>
> The result is also that the ordering is wrong. You don't get the nearest
> point when you expect it.

AFAIU, if error tolerance at 5-th or 6th d.p is enough,
float may be sufficient, but if tolerance is expected at further
positions (e.g. 10th)
double is required. The 6th d.p. of degree is rounds to 10cm, so float may
be inadequate and leading to the issues as you have discovered.

Best regards,
-- 
Mateusz Loskot, http://mateusz.loskot.net