Overflow in Average Interpolation

[johanstenberg]

If you check this link:

​https://github.com/OSGeo/gdal/blob/6c14b8248beee10e4993f9c866e05cf0ab3ee0d3/gdal/alg/gdalwarpkernel.cpp#L4567

I'm guessing this can yield an overflow.

[Even Rouault]

I don't think so. This is a double variable.

[johanstenberg]

I don't really know the ins and outs of C++ but in Java for example, Double.MAX_VALUE + Double.MAX_VALUE leads to Double.Infinity. The same code in C++:

double d = numeric_limits<double>::max(); double c = d + d; cout << c << endl;

Leads to "inf" being printed. I just wanted to let you know.

[johanstenberg]

My point here is that if you have a raster with all numeric_limits<double>::max() / 2, it will overflow and not yield the, computable, average.

[Even Rouault]

ah I see. Hou have seen any such raster using the maximum double value or approachig values ? That would be a pain to deal with.

[johanstenberg]

I don't have a raster, I was just reading through the source and thought about it. This algorithm fixes the issue: ​http://www.heikohoffmann.de/htmlthesis/node134.html.