Here's an example of both approaches. If I have missed something, or you want more information, just let me know.
static void Main()
{
// try to do "x + (3 * x)"
var single = BuildSingle<decimal>();
var composite = BuildComposite<decimal>();
Console.WriteLine("{0} vs {1}", single(13.2M), composite(13.2M));
}
// utility method to get the 3 as the correct type, since there is not always a "int x T"
static Expression ConvertConstant<TSource, TDestination>(TSource value)
{
return Expression.Convert(Expression.Constant(value, typeof(TSource)), typeof(TDestination));
}
// option 1: a single expression tree; this is the most efficient
static Func<T,T> BuildSingle<T>()
{
var param = Expression.Parameter(typeof(T), "x");
Expression body = Expression.Add(param, Expression.Multiply(
ConvertConstant<int, T>(3), param));
var lambda = Expression.Lambda<Func<T, T>>(body, param);
return lambda.Compile();
}
// option 2: nested expression trees:
static Func<T, T> BuildComposite<T>()
{
// step 1: do the multiply:
var paramInner = Expression.Parameter(typeof(T), "inner");
Expression bodyInner = Expression.Multiply(
ConvertConstant<int, T>(3), paramInner);
var lambdaInner = Expression.Lambda(bodyInner, paramInner);
// step 2: do the add, invoking the existing tree
var paramOuter = Expression.Parameter(typeof(T), "outer");
Expression bodyOuter = Expression.Add(paramOuter, Expression.Invoke(lambdaInner, paramOuter));
var lambdaOuter = Expression.Lambda<Func<T, T>>(bodyOuter, paramOuter);
return lambdaOuter.Compile();
}
Personally, I would aim towards the first method; it it both simpler and more efficient. This might involve passing the original parameter throughout a stack of nested code, but so be it. I have got some code somewhere that takes the "Invoke" approach (composite), and re-writes the tree as the first approach (single) - but it is quite complex and long. But very useful for Entity Framework (which doesn't support Expression.Invoke).