[gdal-dev] HDF-EOS vs. GDAL: order of dimensions

[Ivan Shmakov]

	One more issue with HDF-EOS support in GDAL:

$ gdalinfo \
      'HDF4_EOS:EOS_SWATH:"AIRS.....061.L2.RetStd.v4.0.9.102.D07281140036.hdf":L2_Standard_atmospheric&surface_product:latAIRS' \
      | grep -F latAIRS
  SUBDATASET_25_NAME=HDF4_EOS:EOS_SWATH:"AIRS.....061.L2.RetStd.v4.0.9.102.D07281140036.hdf":L2_Standard_atmospheric&surface_product:latAIRS
  SUBDATASET_25_DESC=[45x30x3x3] latAIRS (32-bit floating-point)
$ gdal_translate \
      'HDF4_EOS:EOS_SWATH:"AIRS.....061.L2.RetStd.v4.0.9.102.D07281140036.hdf":L2_Standard_atmospheric&surface_product:latAIRS' \
      airs.lat-airs.tiff
Input file size is 3, 3
0Warning 1: Lost metadata writing to GeoTIFF ... too large to fit in tag.
...10...20...30...40...50...60...70...80...90...100 - done.
$ 

	Surely the source image has to be treaten as 45x30.

	Apparently, GDAL tries to ``guess'' the spatial dimensions of an
	HDF-EOS dataset:

--cut: frmts/hdf4/hdf4imagedataset.cpp: HDF4ImageDataset::GetImageDimensions ()--
    // Search for the starting "X" and "Y" in the names or take
    // the last two dimensions as X and Y sizes
    iXDim = nDimCount - 1;
    iYDim = nDimCount - 2;

    for ( i = 0; i < nDimCount; i++ )
    {
        if ( EQUALN( papszDimList[i], "X", 1 ) && iBandDim != i )
            iXDim = i;
        else if ( EQUALN( papszDimList[i], "Y", 1 ) && iBandDim != i )
            iYDim = i;
    }
--cut: frmts/hdf4/hdf4imagedataset.cpp: HDF4ImageDataset::GetImageDimensions ()--

	And gets it wrong, since the dimensions are:

--cut: ncdump-hdf -h AIRS.....061.L2.RetStd.v4.0.9.102.D07281140036.hdf--
    "\t\t\t\tDataFieldName=\"latAIRS\"\n",
    "\t\t\t\tDataType=DFNT_FLOAT32\n",
    "\t\t\t\tDimList=(\"GeoTrack\",\"GeoXTrack\",\"AIRSTrack\",\"AIRSXTrack\")\n",
    "\t\t\tEND_OBJECT=DataField_40\n",
--cut: ncdump-hdf -h AIRS.....061.L2.RetStd.v4.0.9.102.D07281140036.hdf--

	Any opinions on how the spatial dimensions detection should be
	done to handle such cases?  May it be of any help to try to also
	inspect the geolocation fields and ``guess'' the X, Y dimensions
	only among the dimensions referred by both the geolocation
	fields and the field being handled?  Or should there be a way to
	manually override any detection logic and specify the X, Y
	dimensions explicitly?