Integer overflow in Java bindings for huge image reading

[ehouzay]

It's impossible to read huge images using Java binding (and quite sure other binding even if I don't test it). Integer overflow exception error comes in the console.

This error comes from SWIG and internal buffer size control.

Image size are computes as integer leading to integer overflow with big images.

Have a look in Dataset.i in swig/include...

int ComputeDatasetRasterIOSize (int buf_xsize, int buf_ysize, int nPixelSize,
                                int nBands, int* bandMap, int nBandMapArrayLength,
                                int nPixelSpace, int nLineSpace, int nBandSpace,
                                int bSpacingShouldBeMultipleOfPixelSize );

[Even Rouault]

Could you show the exact values of the parameters of the ReadRaster() call you make ? But I believe the error message is justified. You can't currently read more than 2 GB at once within GDAL and that's the purpose of these checks. This might work in some places, but for sure not all drivers are ready for this. So you can try removing the checks and it might work, but the best practice is that you should split your read into smaller and more manageable pieces.

[ehouzay]

I have tested this on GeoTiff images of 12000x12000 in float (DEM) and Landsat RGB images in MrSID with images of 40000x40000 on a 64 bits computer with 16Gb of RAM...

In C/C++ all is OK. The problem comes from the bidding.

I use GDAL 1.7.2 recompiled from scratch.

As mentionned in the http://trac.osgeo.org/gdal/wiki/Release/1.7.2-News, this version support more than 2Gb allocation in the RasterIO function, called in the binding.

[Even Rouault]

I've checked. The Java language can't support more than 2 GB objects... There's nothing we can do about that unfortunately