[gdal-dev] WCS GetCoverage with AxisOrder swap

[Even Rouault]

On mercredi 8 novembre 2017 14:50:54 CET Ari Jolma wrote:
> Ari Jolma kirjoitti 08.11.2017 klo 13:48:
> >  for MapServer and GeoServer I need to have at least two hack options
> > (NoOffsetSwap and NoGridEnvelopeSwap, maybe there could be only one).
> > I have not yet checked if the rasters returned from those two are ok.
> 
> The rasters that those servers return are ok. I just need the one
> additional hack option for GeoServer to swap the axis names for the
> GetCoverage request.
> 
> I also don't need to use the GridFunction information.

One thing that is confusing, and rings some bells to me as I saw that with GMLJP2 (the
GDAL_JP2K_ALT_OFFSETVECTOR_ORDER config option in gdaljp2metadata.cpp) is the order of the offetVector element

>From your example, we have:

MapServer:
<gml:offsetVector srsName="http://www.opengis.net/def/crs/EPSG/0/4326">0 0.004167</gml:offsetVector>
<gml:offsetVector srsName="http://www.opengis.net/def/crs/EPSG/0/4326">-0.004167 0</gml:offsetVector>

GeoServer:
<gml:offsetVector srsName="http://www.opengis.net/def/crs/EPSG/0/2393">0.0 20.00803035910304</gml:offsetVector>
<gml:offsetVector srsName="http://www.opengis.net/def/crs/EPSG/0/2393">-20.00803035910304 0.0</gml:offsetVector>

Rasdaman
<offsetVector srsName="http://ows.rasdaman.org/def/crs/EPSG/0/4326">-0.02 0</offsetVector>
<offsetVector srsName="http://ows.rasdaman.org/def/crs/EPSG/0/4326">0 0.02</offsetVector> 

So in all 3 cases, the order of the value *inside* a offetVector element is lat/northing long/easting as those CRS as lat-long / northing-easting. OK

BUT Rasdaman does an extra swap of the offsetVector elements themselves.

What is not clear (among many things) is when GridFunction applies. If it applies before linking the raster/grid space to the georeferenced space,
that is the first offsetVector applies to the fastest varying dimension of the final grid,
then I'd say MapServer & GeoServer do the right thing (GeoServer probably closer to correctenss, due to te
presence of the GridFunction). Otherwise, if its applied at the very end of the process, Rasdaman is probably correct.

Practically I'd say if you have
<offsetVector srsName="...">A B</offsetVector>
<offsetVector srsName="...">C D</offsetVector> 
with SRS with lat-long or northing-easting order, with high confidence regarding how "correct" an implementation is:
 * if A != 0 and D > 0 and C=B=0, D the x resolution and A is the y resolution .
 * if A=D=0 and C != 0 and B > 0, then B is the x resolution and C the y resolution.
In other situations, good luck...

-- 
Spatialys - Geospatial professional services
http://www.spatialys.com
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osgeo.org/pipermail/gdal-dev/attachments/20171108/9a786fbe/attachment-0001.html>