[gdal-dev] WMS MINRESOLUTION can lead to overflow of raster dimensions

[Timothy Astle]

I'm trying to open a WMS dataset using the minimally documented 
MINRESOLUTION option. (http://www.gdal.org/frmt_wms.html)

When providing a way-too-fine resolution, I manage to create a raster 
that exceeds GDALs 32-bit integer based dimensions.  Below shows what 
I'm seeing.  You can see that one of the dimensions has overflowed.

gdal_translate.exe -of PNG 
"http://nowcoast.noaa.gov:80/wms/com.esri.wms.Esrimap/obs?SERVICE=WMS&VERSION=1.1.1&REQUEST=GetMap&LAYERS=RAS_RIDGE_NEXRAD&SRS=EPSG:4326&BBOX=-180,-90,180,90&transparent=true&MINRESOLUTION=8.98315e-008" 
my.png

ERROR 1: Invalid dataset dimensions : -2147483648 x 2003751468
GDALOpen failed - 1
Invalid dataset dimensions : -2147483648 x 2003751468

I'm wondering if it'd make sense to handle the kind of input that would 
cause this problem to surface.  Looking at the calculations made in the 
wmsdriver, when a MINRESOLUTION is provided the following code is where 
the issue happens:

     if (osMinResolution.size() != 0)
     {
         double dfMinResolution = CPLAtofM(osMinResolution);

         while (nOverviewCount > 20)
         {
             nOverviewCount --;
             dfMinResolution *= 2;
         }

         nXSize = (int) ((dfMaxX - dfMinX) / dfMinResolution + 0.5);
         nYSize = (int) ((dfMaxY - dfMinY) / dfMinResolution + 0.5);
     }

Would it make sense to do something as follows?  (Note:  I haven't tried 
this yet, I'm just soliciting early feedback so please bare with me )


     if (osMinResolution.size() != 0)
     {
         double dfMinResolution = CPLAtofM(osMinResolution);

         while (nOverviewCount > 20)
         {
             nOverviewCount --;
             dfMinResolution *= 2;
         }

         // *** Changes here ***
         // Determine a suitable resolution that doesn't overflow max int.
         double dXSize = ((dfMaxX - dfMinX) / dfMinResolution + 0.5);
         double dYSize = ((dfMaxY - dfMinY) / dfMinResolution + 0.5);

         while (dXSize > INT_MAX || dYSize > INT_MAX)
         {
             nOverviewCount --;
             dfMinResolution *= 2;

             dXSize = ((dfMaxX - dfMinX) / dfMinResolution + 0.5);
             dYSize = ((dfMaxY - dfMinY) / dfMinResolution + 0.5);
         }

         nXSize = (int) dXSize;
         nYSize = (int) dYSize;
     }

Any thoughts?


Tim Astle