(As reported by Hans Boehm (boehm@acm.org))

Math.hypot says:

"""The computed result must be within 1 ulp of the exact result."""

This means that if the exact result is between adjacent floating point numbers f1 and f2, then the result must be either f1 or f2. I can exhibit a pair of doubles for which this is not the case under OpenJDK 8 through 14.. It's only off by a little.

public class BasicHypotBug {
    public static void main(String[] args) {
        double arg1 = 2.6718173667255144E-307;
        double arg2 = 1.1432573432387167E-308; // Slightly denormal
        // Actual result:
        // 2.6742622187558112649563688387773761534655742E-307
        double resultUpperBound = 2.6742622187558113E-307;
        // Actual value of resultUpperBound:
        // 2.6742622187558112997755783398169632667381058E-307
        double result = Math.hypot(arg1, arg2);
        if (result > resultUpperBound) {
            System.out.println(
                "Result too large: " + result + " > " + resultUpperBound);
        }
    }
}

=>
Result too large: 2.6742622187558117E-307 > 2.6742622187558113E-307