[webmap-discuss] WMS Requests: Rounding BBOX

[Christopher Schmidt]

On Thu, Sep 21, 2006 at 12:20:01PM +0200, Steven M. Ottens wrote:
> Christopher Schmidt wrote:
> > Today, an OpenLayers user asked me why we weren't rounding BBOX values
> > in our tiled image requests, since it was causing problems with his
> > reverse proxy caching to have OpenLayers not do so. The primary reason
> > this isn't done at the moment is that I couldn't pick a good value to
> > use :)
> >
> > 6 decimal places seems to be a generally good value to cut-off at when
> > working in degrees: that gets you down to half-meter accuracy (I 
> > think?). Does anyone have any other thoughts on this? How have you
> > addressed rounding BBOX requests in your own applications?
> 6 decimal places for lat/long is fine. I don't think we need to be able
> to give bounding boxes in mm on lat/long systems anyway. But 6 decimal
> places for projected ones is a bit overkill. Can you make sure that
> those bounding boxes are rounded in a more sensible way?

What is sensible? I know that I can implement it in OpenLayers, but
there needs to be some kind of agreement on what makes sense... the
problem being that what makes sense for you or I doesn't make sense for
everyone. 

For systems using meters, usually two decimal places is overkill... but
there might be someone working with extremely large scale cad drawings
(1:10, or 1:100) where webmap clients make sense... and that 1 cm
difference could be major.

Should rounding be based on the current scale? I think that's the only
way to make sure that requests are never going to lose precision which
is neccesary for correctly displaying the map -- otherwise, when you get
into 1px = 1unit, you'll run into problems with most any rounding...

Does anyone have use cases where rounding has played an effect?

I know that when I was first working with ka-map, I actually did see 
rounding have an effect. THe ka-Map tiling engine rounded scale requests
to the nearest integer. Since ka-Map requires you to enter fixed scales, 
this had never been a problem: but when you try to combined OpenLayers
and ka-map, it did become a problem -- rounding scale to the nearest
integer caused errors that could be as high as half a tile as the tile
got further and further from the geographic meridians. So, I'm aware
that these things do matter... but not sure at what point they matter.
Rounding the scale to four decimal places brought any such differences
to being invisible -- but is that enough for bounding boxes?

-- 
Christopher Schmidt
MetaCarta