[postgis-devel] GBOXF

[Olivier Courtin]

On Dec 1, 2011, at 6:00 PM, Paul Ramsey wrote:

> On Mon, Nov 28, 2011 at 9:45 AM, Olivier Courtin
> <olivier.courtin at oslandia.com> wrote:
>> 
>> On Nov 28, 2011, at 6:38 PM, Paul Ramsey wrote:
>>> One way to ensure consistency would be to make
>>> the GBOX actually be float (GBOXF). I recall you were very happy to
>>> have GBOX replace BOX2FLOAT4. Was that because GBOX is higher
>>> dimensional or because it is double based, or both?
>> 
>> Both in fact,
>> As double bbox are really convenient as our datas are double based.
> 
> What aspect (aside from dimensionality) of the float box was problematic?


I see now only 3 points, and all minors indeed:

1) You avoid all 'stuff' related to double to float conversion.

2)  Upon platform float bbox values could be slightly differents,
      (with real world datas) so unit tests them could be more painfull than double.
     (at least with old unit tests, could be done in a more comprehensive way
      now with cunit)

3)  On a user point of view it looks like 'loosing' precision on the bbox 
      enclosing your datas (meaningfull only with geocentric datas on high scales).

On the other hand, i don't have any metric on the performance overhead to switch from float to double, on a large dataset.



So my primary concern on this point is consistency, 
rather than double vs float.

--
Olivier
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osgeo.org/pipermail/postgis-devel/attachments/20111201/c4a2fcb3/attachment.html>