Boxes and Polygons in ADQL/STC. Questions and recommendation.

[Rob Seaman]

The use case "people trying to get sources that match an image" is a  
bit loose.  For instance, I wrote an IRAF-based astrometry tool to  
query the HST GSC for sources overlapping an image.  A parameter  
expands the boundary of the image in all directions (in pixels or  
arcsecs).  One obvious reason for this is that the tool is being used  
on input images with only poorly known WCS information.  Another  
reason, however, is that typical usage would be applied to a stack of  
images with differing offsets on the sky.  The query is used to  
generate an intermediate catalog applicable to several frames.

In any event, a question here is whether a particular query is  
required to be complete against the image, but may include some  
fraction of hits outside the boundaries of the image - or whether the  
query must be restricted for some reason to hits guaranteed to to rest  
only within the edges (straight, curved or otherwise) of a particular  
area (and then must the query be complete?).  (And what about holes  
that might be cut into the image, etc and so forth?)

Rob
---
On Oct 24, 2009, at 11:56 AM, John Good wrote:

> Roy -
>
>> My point is that I suspect 90% of region queries will, in reality,  
>> be RECT or CONE. We should make sure this works well in the  
>> prototypes, perhaps delay the more complex region implementation  
>> until these two work well?
> Not true.  Cone is probably the most common but the next most common
> is people trying to get sources that match an image, which is hardly  
> ever
> a RECT (and is in fact usually more like a rotated BOX).
>
> - John