获得（非零）double或float的最小步长

Question

For the purposes of boundary value analysis in testing, I've been looking to get the smallest step sizes for doubles or floats in .Net. This will allow me to get the next larger and next smaller value that can be represented in a double or float.

double.epsilon , "applies only to Double instances that have a value of zero or an exponent of -1022".

Let's say that I have an arbitrary double or float. What are the next larger and next smaller value that I can represent in .Net?

Based on the MSDN article, I have implemented the following code, which looks plausible:

private static int GetExponentWithMinorAdjustment(double value)
{
    //int indent = result.Length;

    // Convert the double to an 8-byte array.
    byte[] bytes = BitConverter.GetBytes(value);

    int exponent = (bytes[7] & 0x07F) << 4;
    exponent = exponent | ((bytes[6] & 0xF0) >> 4);
    int adjustment = exponent != 0 ? 1023 : 1022;

    return exponent != 0 ? exponent - 1 : exponent; ;
}

private static double NextLargerNumber(double value)
{
    // Cannot use double.epsilon * Math.Pow(2, GetExponentWithMinorAdjustment(value)) as this results in infinity
    return value + Math.Pow(2, GetExponentWithMinorAdjustment(value) - 1074));
}

The problem with the above code is that it looks shaky. Notice the ternary operator (inspired from the MSDN article); what's the purpose of that adjustment?

Also, while the output of several tests looks OK for positive double values, the results for negative values do not seem as good (see output image):

What would an acceptable, correct and ideal solution to this problem look like?

Answer 1

Use unsafe code to map the bits directly to an unsigned int, increment or decrement, convert back.

Might require additional special handling at the boundaries of 0 and the special values for infinity, Nan and so on.