How do I break down a chain of member access expressions?

问题

The Short Version (TL;DR):

Suppose I have an expression that's just a chain of member access operators:

Expression<Func<Tx, Tbaz>> e = x => x.foo.bar.baz;

You can think of this expression as a composition of sub-expressions, each comprising one member-access operation:

Expression<Func<Tx, Tfoo>>   e1 = (Tx x) => x.foo;
Expression<Func<Tfoo, Tbar>> e2 = (Tfoo foo) => foo.bar;
Expression<Func<Tbar, Tbaz>> e3 = (Tbar bar) => bar.baz;

What I want to do is break e down into these component sub-expressions so I can work with them individually.

The Even Shorter Version:

If I have the expression x => x.foo.bar, I already know how to break off x => x.foo. How can I pull out the other sub-expression, foo => foo.bar?

Why I'm Doing This:

I'm trying to simulate "lifting" the member access operator in C#, like CoffeeScript's existential access operator ?.. Eric Lippert has stated that a similar operator was considered for C#, but there was no budget to implement it.

If such an operator existed in C#, you could do something like this:

value = target?.foo?.bar?.baz;

If any part of the target.foo.bar.baz chain turned out to be null, then this whole thing would evaluate to null, thus avoiding a NullReferenceException.

I want a Lift extension method that can simulate this sort of thing:

value = target.Lift(x => x.foo.bar.baz); //returns target.foo.bar.baz or null

What I've Tried:

I've got something that compiles, and it sort of works. However, it's incomplete because I only know how to keep the left side of a member access expression. I can turn x => x.foo.bar.baz into x => x.foo.bar, but I don't know how to keep bar => bar.baz.

So it ends up doing something like this (pseudocode):

return (x => x)(target) == null ? null
       : (x => x.foo)(target) == null ? null
       : (x => x.foo.bar)(target) == null ? null
       : (x => x.foo.bar.baz)(target);

This means that the leftmost steps in the expression get evaluated over and over again. Maybe not a big deal if they're just properties on POCO objects, but turn them into method calls and the inefficiency (and potential side effects) become a lot more obvious:

//still pseudocode
return (x => x())(target) == null ? null
       : (x => x().foo())(target) == null ? null
       : (x => x().foo().bar())(target) == null ? null
       : (x => x().foo().bar().baz())(target);

The Code:

static TResult Lift<T, TResult>(this T target, Expression<Func<T, TResult>> exp)
    where TResult : class
{
    //omitted: if target can be null && target == null, just return null

    var memberExpression = exp.Body as MemberExpression;
    if (memberExpression != null)
    {
        //if memberExpression is {x.foo.bar}, then innerExpression is {x.foo}
        var innerExpression = memberExpression.Expression;
        var innerLambda = Expression.Lambda<Func<T, object>>(
                              innerExpression, 
                              exp.Parameters
                          );  

        if (target.Lift(innerLambda) == null)
        {
            return null;
        }
        else
        {
            ////This is the part I'm stuck on. Possible pseudocode:
            //var member = memberExpression.Member;              
            //return GetValueOfMember(target.Lift(innerLambda), member);
        }
    }

    //For now, I'm stuck with this:
    return exp.Compile()(target);
}

This was loosely inspired by this answer.

---

Alternatives to a Lift Method, and Why I Can't Use Them:

The Maybe monad

value = x.ToMaybe()
         .Bind(y => y.foo)
         .Bind(f => f.bar)
         .Bind(b => b.baz)
         .Value;

Pros:

1. Uses an existing pattern that's popular in functional programming
2. Has other uses besides lifted member access

Cons:

1. It's too verbose. I don't want a massive chain of function calls every time I want to drill a few members down. Even if I implement SelectMany and use the query syntax, IMHO that will look more messy, not less.
2. I have to manually rewrite x.foo.bar.baz as its individual components, which means I have to know what they are at compile time. I can't just use an expression from a variable like result = Lift(expr, obj);.
3. Not really designed for what I'm trying to do, and doesn't feel like a perfect fit.

ExpressionVisitor

I modified Ian Griffith's LiftMemberAccessToNull method into a generic extension method that can be used as I've described. The code is too long to include here, but I'll post a Gist if anyone's interested.

Pros:

1. Follows the result = target.Lift(x => x.foo.bar.baz) syntax
2. Works great if every step in the chain returns a reference type or a non-nullable value type

Cons:

1. It chokes if any member in the chain is a nullable value type, which really limits its usefulness to me. I need it to work for Nullable<DateTime> members.

Try/catch

try 
{ 
    value = x.foo.bar.baz; 
}
catch (NullReferenceException ex) 
{ 
    value = null; 
}

This is the most obvious way, and it's what I'll use if I can't find a more elegant way.

Pros:

1. It's simple.
2. It's obvious what the code is for.
3. I don't have to worry about edge cases.

Cons:

1. It's ugly and verbose
2. The try/catch block is a nontrivial* performance hit
3. It's a statement block, so I can't make it emit an expression tree for LINQ
4. It feels like admitting defeat

I'm not going to lie; "not admitting defeat" is the main reason I'm being so stubborn. My instincts say there must be an elegant way to do this, but finding it has been a challenge. I can't believe it's so easy to access the left side of an expression, yet the right side is nigh-unreachable.

I really have two problems here, so I'll accept anything that solves either one:

• Expression decomposition that preserves both sides, has reasonable performance, and works on any type
• Null-propagating member access

---

Update:

Null-propagating member access is planned for included in C# 6.0. I'd still like a solution to expression decomposition, though.

回答1:

If it's just a simple chain of member access expressions, there is an easy solution:

public static TResult Lift<T, TResult>(this T target, Expression<Func<T, TResult>> exp)
    where TResult : class
{
    return (TResult) GetValueOfExpression(target, exp.Body);
}

private static object GetValueOfExpression<T>(T target, Expression exp)
{
    if (exp.NodeType == ExpressionType.Parameter)
    {
        return target;
    }
    else if (exp.NodeType == ExpressionType.MemberAccess)
    {
        var memberExpression = (MemberExpression) exp;
        var parentValue = GetValueOfExpression(target, memberExpression.Expression);

        if (parentValue == null)
        {
            return null;
        }
        else
        {
            if (memberExpression.Member is PropertyInfo)
                return ((PropertyInfo) memberExpression.Member).GetValue(parentValue, null);
            else
                return ((FieldInfo) memberExpression.Member).GetValue(parentValue);
        }
    }
    else
    {
        throw new ArgumentException("The expression must contain only member access calls.", "exp");
    }
}

EDIT

If you want to add support for method calls, use this updated method:

private static object GetValueOfExpression<T>(T target, Expression exp)
{
    if (exp == null)
    {
        return null;
    }
    else if (exp.NodeType == ExpressionType.Parameter)
    {
        return target;
    }
    else if (exp.NodeType == ExpressionType.Constant)
    {
        return ((ConstantExpression) exp).Value;
    }
    else if (exp.NodeType == ExpressionType.Lambda)
    {
        return exp;
    }
    else if (exp.NodeType == ExpressionType.MemberAccess)
    {
        var memberExpression = (MemberExpression) exp;
        var parentValue = GetValueOfExpression(target, memberExpression.Expression);

        if (parentValue == null)
        {
            return null;
        }
        else
        {
            if (memberExpression.Member is PropertyInfo)
                return ((PropertyInfo) memberExpression.Member).GetValue(parentValue, null);
            else
                return ((FieldInfo) memberExpression.Member).GetValue(parentValue);
        }
    }
    else if (exp.NodeType == ExpressionType.Call)
    {
        var methodCallExpression = (MethodCallExpression) exp;
        var parentValue = GetValueOfExpression(target, methodCallExpression.Object);

        if (parentValue == null && !methodCallExpression.Method.IsStatic)
        {
            return null;
        }
        else
        {
            var arguments = methodCallExpression.Arguments.Select(a => GetValueOfExpression(target, a)).ToArray();

            // Required for comverting expression parameters to delegate calls
            var parameters = methodCallExpression.Method.GetParameters();
            for (int i = 0; i < parameters.Length; i++)
            {
                if (typeof(Delegate).IsAssignableFrom(parameters[i].ParameterType))
                {
                    arguments[i] = ((LambdaExpression) arguments[i]).Compile();
                }
            }

            if (arguments.Length > 0 && arguments[0] == null && methodCallExpression.Method.IsStatic &&
                methodCallExpression.Method.IsDefined(typeof(ExtensionAttribute), false)) // extension method
            {
                return null;
            }
            else
            {
                return methodCallExpression.Method.Invoke(parentValue, arguments);
            }
        }
    }
    else
    {
        throw new ArgumentException(
            string.Format("Expression type '{0}' is invalid for member invoking.", exp.NodeType));
    }
}

来源：https://stackoverflow.com/questions/11108254/how-do-i-break-down-a-chain-of-member-access-expressions