Prevent double NAN or Infinity conditions

Question

Certain operations in .Net can give a double variable the values "NAN" or "1.#INF".  That's great flexibility but I would rather an exception be thrown any time a double is given one of these values.  In my program these values should not exist, and when it does it inevitably leads to an error down the road.  I keep trying to put in checks using double.IsNAN or double.IsInfinity but certain things keep slipping by.  So there is any setting in .Net to just make it always invalid for a double to be set to these values?

(Edit:  also annoying that I think you have to use two checks:  IsNaN and IsInfinity to catch these two possibilities.  If anyone knows of one check that catches any double that is not a real numeric value it would be a help.)

Answer 1

You might want to check out this Microsoft article for ideas:

Test for NaN Correctly

Mike McIntyre's .Net Journal

getdotnetcode.com

---

Mike McIntyre Senior Developer / Partner aZ Software Developers

Answer 2

I have not tested it, but what about something like this.

Class myDouble
    Private _value As [Double]
    Public Sub New(ByVal value As [Double])
        If [Double].IsInfinity(value) OrElse [Double].IsNaN(value) Then
            Throw New NotImplementedException()
        End If
        
        _value = value
    End Sub
    
    Public Shared Widening Operator CType(ByVal value As [Double]) As myDouble
        Return New myDouble(value)
    End Operator
    Public Shared Widening Operator CType(ByVal value As myDouble) As [Double]
        Return value._value
    End Operator
End Class

Answer 3

jgalley - that is an interesting solution.  I'm still surprised there isn't a well-known solution (other than repeatedly using IsNAN and IsInfinity) since I imagine there are lots of routines out there that aren't designed to handle numbers of these special types. 

Do you know what the performance ramifications are of using CType?  If myDouble gets converted everytime it's used in an arithmetic operation (thinking about large loops with lots of calculations), will that have a big impact?  Since it is just returning the private _value I would hope not, but I don't know much about implementing CType. (Edit:  and I wonder if there would be a performance improvement if using a structure instead of class...)

Answer 4

It's hard to tell the performanceof using CType, but i think the answer would be yes.

Answer 5

From what I've read, CType can have a significant performance impact.  I don't know if there is any difference between existing CType's and user-created implementations like the above.  I will give galley the answer, but when numbers are concerned I think calculation performance could be a big issue (at least for what I'm doing).   I'll try to test it out if I ever find the time.