Subtraction fails for large numbers

[seanmkauffman]

Applying a numeric function (at least subtraction, possibly others) to numbers large enough to overflow an int will fail. The following code should output 10.0 but actually outputs 0.0.

import ca.uqac.lif.cep.util.Numbers;
import ca.uqac.lif.cep.functions.Function;

public class TestDiff {
	public static void main(String[] args) {
		Function negation = Numbers.subtraction;
		Object[] out = new Object[1];
		negation.evaluate(new Object[]{1590785415514L, 1590785415504L}, out);
		System.out.println("The return value of the function is: " + out[0]);
	}
}

[sylvainhalle]

This is "expected", in the sense that the Numbers utility class converts everything into float internally. See for example in the case of Subtraction:

beepbeep-3/Core/src/ca/uqac/lif/cep/util/Numbers.java

Line 637 in 6314e68

return x.floatValue() - y.floatValue();

Sadly, Java does not allow operations directly on instances of Number, so a conversion to some concrete type must be done. I chose float out of a (sparsely justified) belief that operations on floats are faster than on doubles. A possible solution would be simply to switch all functions in the Number class to doubles internally.

Is there a specific use case that requires you to use such large numbers?

[seanmkauffman]

The use case is to calculate the difference between millisecond-granularity timestamps, such as those returned by System.currentTimeMillis(). In my case, there is a workaround, because I know that the timestamps in question will fall within a certain period. So, I can just mask off everything except the lower several bits. However, I could foresee circumstances where this isn't possible.

If there was a warning that told the user they lost information from the cast, that might be a helpful solution.