MathLink error while calling a custom Mathematica function - MLGet out of sequence

Question

I am trying to call Mathematica 10 from F#. Following the C# sample code found in the Mathematica directory, I have written the following F# code:

type TestClass() =
    member this.AddNumsWithWM (m : int) (n : int) : int =
        let ml = MathLinkFactory.CreateKernelLink()
        ml.WaitAndDiscardAnswer()
        ml.PutFunction("EvaluatePacket", 1)
        ml.PutFunction("testFunc", 2) // IF BUILT-IN FUNCTION LIKE Plus, THEN NO ERROR
        ml.Put(m)
        ml.Put(n)
        ml.EndPacket()
        ml.WaitForAnswer() |> ignore
        let res = ml.GetInteger() // THE ERROR OCCURS HERE FOR CUSTOM FUNCTION
        ml.Close()
        res

The Mathematica function testFunc is simply:

testFunc[m_Integer, n_Integer] := m + 2 * n;

But when I call the function AddNumsWithWM 2 3, I get the following error at the line indicated above:

Wolfram.NETLink.MathLinkException was unhandled
Message: An unhandled exception of type 'Wolfram.NETLink.MathLinkException' occurred in Wolfram.NETLink.dll
Additional information: Error code: 3.

I don't understand, because the sample C#, from which this has been translated literally, works. Any idea why I would get the above error?

Thanks in advance for your help.

EDIT

I have verified something else. If I use a built-in function like Plus instead of my custom function testFunc, the code works.

So, then, I guess my question would be, Why does the code not work for a custom function? I would be very grateful for pointers on what I need to do differently in order to call a custom function.

I have edited the question accordingly.

Answer 1

testFunc[] needs to be defined prior to being called. I don't use .NETLink, so I can't guarantee if the following will work, but try ml.Evaluate("testFunc[m_Integer, n_Integer] := m + 2 * n;") to do so. I'm not sure if this returns any packets and needs to be followed up with an ml.WaitAndDiscardAnswer(). In any case, if you have many custom functions, I suggest putting them in a package to be loaded via an ml.Evaluate("Get[package];") to reduce overhead from multiple .NETLink calls.