We've run into following java function that models some variation of NULL arithmetic:

public static Long add(Long value1, Long value2)
{
  return value1 == null ? value2 : value2 == null ? value1 : value1 + value2;
}

When this function runs the outcome is different from what we have expected.

Here is a quiz:

1. What is outcome of add(1, 2):

   1. 3;
   2. null;
   3. other.

2. What is outcome of add(3, null):

   1. 3;
   2. null;
   3. other.

3. What is outcome of add(null, 4):

   1. null;
   2. 4;
   3. other.

4. What is outcome of add(null, null):

   1. null;
   2. 0;
   3. other.

Our assumptions were:

1. add(1, 2) == 3;
2. add(3, null) == 3;
3. add(null, 4) == 4;
4. add(null, null) == null;

Java works differently:

1. add(1, 2) == 3;
2. add(3, null) throws NullPointerException;
3. add(null, 4) throws NullPointerException;
4. add(null, null) throws NullPointerException;

The problem is with compile time type of ternary ?: operator. Compiler decides it's long, while we intuitively expected Long. Java casts null to long (which results into NPE), and then to Long.

Correct code would be:

public static Long add(Long value1, Long value2)
{
  if (value1 == null)
  {
    return value2;
  }
  else if (value2 == null)
  {
    return value;
  }
  else
  {
    return value1 + value2;
  }
}

This version does not cast anything to long, and works as we originally expected.

Honestly, we're a bit anexious about this subtle difference of if-then-else and ?: operator.