Implementing Compare for FXCop rule CA2225: Operator overloads have named alternates.

I'm getting this FXCop Error because I overloaded the >= and <= operators on a certain type:

"Provide a method named 'Compare' as a friendly for operator 'Foo.operator >=(Foo, Foo)'."

I've been looking all over the place but I can't find a good example of how to implement this method. I know how to do the actual comparison, but FXCop wants me to create a Compare() method for the >, <, >= and <= operators. That means I have to somehow tell the Compare() method wether I want to check if the two parameters are greater than, smaller than, greater than or equal to or smaller than or equal to each others. I guess I could create a ComparisonOperator enum and pass that, but surely there's a more convenient way to do it?

Ah, perfect. I knew there had to be a simpler way but I couldn't think of it. Thanks!

Note that although -1, 0, and 1 are normally used, they don't have to be. For example:

int Compare(int a, int b)
{
  return b - a;
} 

Fullfils the criteria of the contract without any conditionals

10 hours ago, evildictaitor wrote

Fullfils the criteria of the contract without any conditionals

Not necessarily, as the "wrong" values of a and b in that case can cause unexpected exceptions that shouldn't be passed on to the caller.

1 hour ago, AndyC wrote

*snip*

Not necessarily, as the "wrong" values of a and b in that case can cause unexpected exceptions that shouldn't be passed on to the caller.

I could see this being the case with nullable ints, for instance, but care to elaborate on other scenarios?

Also, shouldn't it be the other way around, a-b? Say a = 5 and b = 10, then a-b =  5-10 = -5, less than zero hence a is smaller than b.

1 hour ago, Bas wrote

*snip*

I could see this being the case with nullable ints, for instance, but care to elaborate on other scenarios?

Also, shouldn't it be the other way around, a-b? Say a = 5 and b = 10, then a-b =  5-10 = -5, less than zero hence a is smaller than b.

I think you're probably right about the order. As to situations in which it goes wrong, consider the case where a = Int32.MinValue and b > 0, thus a-b ends up with an overflow. 

Yeah, the subtraction approach doesn't really work because of overflow problems (which in C# won't cause exceptions by default, just incorrect results; you need to use "checked" to get an exception).

Also, don't just provide a method named compare, but actually implement IComparable<T>. It helps if you ever want to sort a list of your type.

Good points all around. I think I'm going to need sorting in the near future so I might as well implement IComparable<T>. Thanks guys!

My point was that you don't have to use { -1, 0, 1 }. But since I made my original claim that you can do it without conditionals, and my continued assertion that you can, I am left with no choice but to resort to witchcraft:

static int Compare(int a, int b)
{
   return (a - b) ^ (int)((0 - ((a ^ b ^ (a - b)) >> 31)));
}

You are the devil.

@evildictaitor: ok.  this is still absorbing.  would changing the >> 32 to << -1 make it a better candidate for genericizing (i.e. handling int64 as well)?

@davewill: No. Left shift and right shift are undefined when the value is negative. Your example might work on x86 (because it bitmasks the right operand) but might not work on ARM.

   return (a - b) ^ (int)((0 - ((a ^ b ^ (a - b)) >> ((sizeof(T) << 3) - 1))));

Would be more generic - assuming of course that you have 8-bit bytes