#r "nuget: DiffSharp.Core, 1.0.7-preview1873603133"
#r "nuget: DiffSharp.Backends.Reference, 1.0.7-preview1873603133"
#r "nuget: DiffSharp.Backends.Torch, 1.0.7-preview1873603133"
open DiffSharp
open DiffSharp.Util
let t3 = dsharp.tensor [[1.1; 2.2]; [1.1; 2.2]; [1.1; 2.2]]
1   t3 //Does work!!

open System.Linq.Expressions
let addB = Expression.Parameter(typeof<int>, "b")
let addC = Expression.Parameter(typeof<Tensor>, "c")
Expression.Add(
            addC,
            addB
        ) //=> Throw exception...

(*
System.InvalidOperationException: The binary operator Add is not defined for the types 'DiffSharp.Tensor' and 'System.Int32'.
   at System.Linq.Expressions.Expression.GetUserDefinedBinaryOperatorOrThrow(ExpressionType binaryType, String name, Expression left, Expression right, Boolean liftToNull)
   at System.Linq.Expressions.Expression.Add(Expression left, Expression right, MethodInfo method)
   at System.Linq.Expressions.Expression.Add(Expression left, Expression right)
   at <StartupCode$FSI_0048>.$FSI_0048.main@()
Stopped due to error
*)

Why 1 t3 is legal but unable to parse into Linq expression? And I would like to know, if this RFC is applied, will it helpful to resolve this issue? F# RFC FS-1043 - Extension members become available to solve operator trait constraints

CodePudding user response：

@panagiotis

type TRIVIAL = {
    orz: int
}
    with
        static member ( ) (vl : TRIVIAL, vr : TRIVIAL) =
            {
                orz = vl.orz   vr.orz
            }
        static member ( ) (vl : TRIVIAL, vr : int) =
            {
                orz = vl.orz   vr
            }
        static member ( ) (vl : int, vr : TRIVIAL) =
            {
                orz = vl   vr.orz * 100
            }
let addA = Expression.Parameter(typeof<TRIVIAL>, "a")
let addB = Expression.Parameter(typeof<int>, "b")

Expression.Add(
            addB,
            addA
        )

But why this works as well?

CodePudding user response：

According to @NetMage's comments, the last solution is like the following code snippets:

The value static member part:

type Value =
    | Number of BigRational
    | Approximation of Approximation
    | ComplexInfinity
    | PositiveInfinity
    | NegativeInfinity
    | Undefined
    | RealVec of Vector<float>
    | ComplexVec of Vector<complex>
    | RealMat of Matrix<float>
    | ComplexMat of Matrix<complex>
    | DSTen of Tensor
    with 
        static member ( ) (vl : Value, vr : Value) =
            match vl with
            | Number vlv ->
                match vr with
                | Number vrv ->
                    Number (vlv * vrv)
            | Approximation (Real vlv) ->
                match vr with
                | Approximation (Real vrv) ->
                    Approximation (Real (vlv   vrv))
                | DSTen dt ->
                    DSTen (vlv   dt)
        static member (*) (vl : Value, vr : float) =
            match vl with
            | Approximation (Real vlv) ->
                Approximation (Real (vlv * vr))
        static member (*) (vl : float, vr : Value) =
            match vr with
            | Approximation (Real vrv) ->
                Approximation (Real (vl * vrv))
        static member ( ) (vl : Value, vr : float) =
            match vl with
            | Approximation (Real vlv) ->
                Approximation (Real (vlv   vr))
        static member ( ) (vl : float, vr : Value) =
            match vr with
            | Approximation vrv ->
                match vrv with
                | Approximation.Real vrvv ->
                    Approximation (Real (vl   vrvv))
            | DSTen dt ->
                DSTen (vl   dt)

The casting part:

    let exprObj2ValueToInject =
        ExprHelper.Quote<Func<obj, MathNet.Symbolics.Value>> (fun j ->
            match j with
            | :? Value -> (j :?> Value) 
            | _ when j.GetType() = typeof<float> ->
                Value.Approximation (Approximation.Real (j :?> float))
            | :? Vector<float> ->
                Value.RealVec (j :?> Vector<float>)
            | :? Matrix<float> ->
                Value.RealMat (j :?> Matrix<float>)
            | _ ->
                failwithf "orz010: %s, %A" (j.GetType().FullName) j
            )
            :> Expression :?> LambdaExpression

The invocation part:

                    let xsvv =
                        xsv
                        |> List.map (fun xsExp ->
                            let casted = Expression.Convert(xsExp, typeof<obj>) :> Expression
                            let ivk = Expression.Invoke(exprObj2ValueToInject, [|casted|])
                            ivk :> Expression
                        )


                    let vLambda = Expression.Invoke(exprBy2Lambda, xsvv)

No matter what is passed in, I convert the value to object and using match to determine how to wrap it.