[gdal-dev] Setting "INIT_DEST" to nan on windows

[Even Rouault]

Le mardi 10 mai 2011 17:39:52, Kyle Shannon a écrit :
> Even, here is the basic flow of my task.  I am opening several subdatasets
> in a for loop, warping them and populating an internal data structure.
> 
> #include "gdal_priv.h"
> #include "gdalwarper.h"
> #include "cpl_conv.h"
> #include "cpl_port.h"
> #include <string>
> #include <vector>
> int main()
> {
> GDALAllRegister();
> /*
>  * Variable names for the netcdf forecast file
>  */
> 
> std::vector<std::string> varList;
> varList.push_back( "Temperature_height_above_ground" );
> varList.push_back( "V-component_of_wind_height_above_ground" );
> varList.push_back( "U-component_of_wind_height_above_ground" );
> varList.push_back( "Total_cloud_cover" );
> 
> std::string temp;
> GDALDataset* srcDS;
> GDALDataset* wrpDS;
> GDALWarpOptions* psWarpOptions;
> std::string srcWkt, dstWkt;
> dstWkt = "PROJCS[\"WGS 84 / UTM zone 12N\",GEOGCS[\"WGS
> 84\",DATUM[\"WGS_1984\",SPHEROID[\"WGS
> 84\",6378137,298.257223563]],PRIMEM[\"Greenwich\",0],UNIT[\"degree\",0.0174
> 532925199433]],UNIT[\"metre\",1],PROJECTION[\"Transverse_Mercator\"],PARAME
> TER[\"latitude_of_origin\",0],PARAMETER[\"central_meridian\",-111],PARAMETE
> R[\"scale_factor\",0.9996],PARAMETER[\"false_easting\",500000],PARAMETER[\"
> false_northing\",0]]";
> 
> /*
>  * Loop over variable names and open sub datasets
>  */
> 
> for( int i = 0;i < varList.size();i++ ) {
> 
>     temp = "NETCDF:test.nc:" + varList[i];
> 
>     srcDS = (GDALDataset*)GDALOpenShared( temp.c_str(), GA_ReadOnly );
> 
>     srcWkt = srcDS->GetProjectionRef();
> 
>     /*
>      * Grab the first band to get the nodata value for the variable,
>      * assume all bands have the same ndv
>      */
>     GDALRasterBand *poBand = srcDS->GetRasterBand( 1 );
>     int pbSuccess;
>     double dfNoData = poBand->GetNoDataValue( &pbSuccess );
>     psWarpOptions = GDALCreateWarpOptions();
> 
>     int nBandCount = srcDS->GetRasterCount();
>     psWarpOptions->padfDstNoDataReal =
>         (double*) CPLMalloc( sizeof( double ) * nBandCount );
>     psWarpOptions->padfDstNoDataImag =
>         (double*) CPLMalloc( sizeof( double ) * nBandCount );
>     for( int b = 0;b < srcDS->GetRasterCount();b++ ) {
>         psWarpOptions->padfDstNoDataReal[b] = dfNoData;
>         psWarpOptions->padfDstNoDataImag[b] = dfNoData;
>     }
>     if( pbSuccess == false )
>         dfNoData = -9999.0;
>     psWarpOptions->papszWarpOptions =
>     CSLSetNameValue( psWarpOptions->papszWarpOptions,
>                      "INIT_DEST", "NO_DATA" );
> 
>     //set the dskWkt with an internal structure
> 
>     wrpDS = (GDALDataset*) GDALAutoCreateWarpedVRT( srcDS, srcWkt.c_str(),
>                                                     dstWkt.c_str(),
>                                                     GRA_NearestNeighbour,
>                                                     1.0, psWarpOptions );
> 
>     //copy warped data to internal structure
> 
>     GDALDestroyWarpOptions( psWarpOptions );
>     GDALClose((GDALDatasetH) srcDS );
>     GDALClose((GDALDatasetH) wrpDS );
> }
> }
> 

Of course, Valgrind has been helpfull once again (too bad you're on Windows 
;-))

You just forgot : psWarpOptions->nBandCount = nBandCount;

> I am sure I am missing something.
> 
> kss
> 
> /**
>  *
>  * Kyle Shannon
>  * ksshannon at gmail.com
>  *
>  */
> 
> On Tue, May 10, 2011 at 03:51, Even Rouault <even.rouault at mines-
paris.org>wrote:
> > Selon Kyle Shannon <ksshannon at gmail.com>:
> > > I did have to set the padfNoDataImag array as well, so I am one step
> > > closer.  My new issue is ownership of my psWarpOptions.  I assume that
> > 
> > since
> > 
> > > I create it, and the warp api tutorial implies that is it mine, I
> > > destroy
> > 
> > it
> > 
> > > with GDALDestroyWarpOptions().  My problem is that AutoCreateWarpedVRT
> > > appears to be stepping on my warp options when it destroys it's own
> > > copy.
> > > 
> > > I call AutoCreateWarpedVRT() and that function calls
> > > GDALCloneWarpOptions().  When AutoCreateWarpedVRT destroys the copy it
> > 
> > owns,
> > 
> > > it appears to be stepping on my copy, and when I call
> > > GDALDestroyWarpOptions(), I get a seg fault.  Am I doing something
> > > wrong?
> > 
> > Difficult to say without seeing your code. Could you post a small
> > self-sufficient code snippet that demonstrates your issue ?
> > 
> > By reviewing the code, one potential issue would be if you try to free
> > the pTransformerArg field by yourself, as it is freed by the destructor
> > of the VRTWarpedDataset. But if you just use GDALDestroyWarpOptions(),
> > it should work
> > OK
> > 
> > > kss
> > > 
> > > 
> > > 
> > > /**
> > > 
> > >  *
> > >  * Kyle Shannon
> > >  * ksshannon at gmail.com
> > >  *
> > >  */
> > > 
> > > On Mon, May 9, 2011 at 15:43, Even Rouault
> > > 
> > > <even.rouault at mines-paris.org>wrote:
> > > > Le lundi 09 mai 2011 23:30:05, Kyle Shannon a écrit :
> > > > > Thanks Even,
> > > > > If that is the case, are there other values in psWarpOptions I
> > > > > *have*
> > 
> > to
> > 
> > > > > set?  I can't find docs discussing that.
> > > > 
> > > > I don't think so, but you'll just have to try... I'm not  surprised
> > > > the docs
> > > > don't talk about NaN. It is a corner case, and likely not really well
> > > > supported in all places.