The following sample floating-point strings must be rounded to +Infinity
when converted to binary representation. Instead, they are rounded
to Float.MAX_VALUE and Double.MAX_VALUE respectively.

class A {
    public static void main(String[] args) {
        System.out.println(3.4028235677973366E38f);
        System.out.println(Float.valueOf("3.4028235677973366E38"));
        System.out.println(1.79769313486231579979201547673599e308);
        System.out.println(Double.valueOf("1.79769313486231579979201547673599e30
8"));
    }
}

Output:

3.4028235E38
3.4028235E38
1.7976931348623157E308
1.7976931348623157E308

The original bug was filed against javac, which relies on the standard
library to do the conversions. The original description follows:

Name: dkC59003 Date: 07/06/98



The following sample floating-point literals must be rounded to +Infinity
when converted to binary representation, and compile-time error must
be detected by javac. Nevetherless, javac (JDK 1.2b4K) rounds these
constants to Float.MAX_VALUE and Double.MAX_VALUE respectively.

------------------------------------------------------------------

class HugeFPLiteral {

    HugeFPLiteral (float value) {
// The following value equals to:
//
// Float.MAX_VALUE + set_25th_bit_in_mantissa
//
// This value must be rounded to +Infinity according to
// rounding-to-nearest rule, IEEE-754, section 4.3.

float too_huge = 3.4028235677973366E38f; // must imply compiler error
    };

    HugeFPLiteral (double value) {
// The following value equals to:
//
// Double.MAX_VALUE + set_53th_bit_in_mantissa
//
// This value must be rounded to +Infinity according to
// rounding-to-nearest rule, IEEE-754, section 4.3.

double too_huge = 1.79769313486231579979201547673599e308;
    };

}

------------------------------------------------------------------

This bug should be covered by the jmpp test:
    tests/lang/FP/fpl022/fpl02201m
which is to be included into future versions of JCK.

======================================================================