If vs重载vs反射

Question

我有很多类的属性，如:

class C1
{
    [PropName("Prop1")]
    public string A {get;set;}
    [PropName("Prop2")]
    public string B {get;set;}
    [PropName("Prop3")]
    public string C {get;set;}
} 
class C2
{
    [PropName("Prop1")]
    public string D {get;set;}
    [PropName("Prop2")]
    public string E {get;set;}
    [PropName("Prop3")]
    public string F {get;set;}
} 

属性告诉我们什么是实际的属性，但是c#属性的名称并不总是匹配的。对于C1和C2, C1。A与C2.D具有相同的属性。

这些类不是任何继承链的一部分，我无法控制它们，所以我不能更改它们。

"Prop1"，"Prop2"，…有一些常见的操作。"PropN"。写这些操作的最佳解决方案是什么，既不需要太多的代码重复，又能使其易于维护?

解决方案#1 (if语句-很多)

void OperationWithProp1(object o)
{
    string prop1;        
    C1 class1 = o as C1;
    if (class1 != null)
        prop1 = class1.A;
    C2 class2 = o as C2;
    if (class2 != null)
        prop1 = class2.D;
    // Do something with prop1
}

解决方案#2(重载-很多)

void OperationWithProp1(string prop1)
{
    // Do something with prop1
}
void RunOperationWithProp1(C1 class1)
{
    OperationWithProp1(class1.A);
}
void RunOperationWithProp1(C2 class2)
{
    OperationWithProp1(class2.D);
}

解决方案#3(反射)-我担心性能，因为这些操作中的每一个都将被调用数千次，并且有几百个操作

void OperationWithProp1(object o)
{
     // Pseudo code:
     // Get all properties from o that have the PropName attribute
     // Look if any attribute matches "Prop1"
     // Get the value of the property that matches
     // Do something with the value of the property
}

你会选择哪个解决方案，为什么?你有其他的款式吗?

---

编辑说明:

很多类意味着几十个类

很多属性意味着30-40个属性/class

Answer 1

您可以创建一个包装器类来公开您需要的属性，并包装实际的C1和C2类的实例。一种方法是通过委托:

interface WithProperties {
   string A {get;set;}
   string B {get;set;}
   string C {get;set;}
}
class WrappedCX<T> : WithProperties {
    private readonly T wrapped;
    private readonly Func<T,string> getA;
    private readonly Action<T,string> setA;
    private readonly Func<T,string> getB;
    private readonly Action<T,string> setB;
    private readonly Func<T,string> getC;
    private readonly Action<T,string> setC;
    public WrappedCX(T obj, Func<T,string> getA, Action<T,string> setA, Func<T,string> getB, Action<T,string> setB, Func<T,string> getC, Action<T,string> setC) {
        wrapped = obj;
        this.getA = getA;
        this.setA = setA;
        this.getB = getB;
        this.setB = setB;
        this.getC = getC;
        this.setC = setC;
    }
    public string A {
        get {return getA(wrapped);}
        set {setA(wrapped, value);}
    }
    public string B {
        get {return getB(wrapped);}
        set {setB(wrapped, value);}
    }
    public string C {
        get {return getC(wrapped);}
        set {setC(wrapped, value);}
    }
}

现在你可以这样做:

C1 c1 = new C1();
C2 c2 = new C2();
WithProperties w1 = new WrappedCX(c1, c => c.A, (c,v) => {c.A=v;}, c => c.B, (c,v) => {c.B=v;}, c => c.C, (c,v) => {c.C=v;});
WithProperties w2 = new WrappedCX(c2, c => c.D, (c,v) => {c.D=v;}, c => c.E, (c,v) => {c.E=v;}, c => c.F, (c,v) => {c.F=v;});

此时，w1和w2都实现了通用的WithProperties接口，因此您可以使用它们而无需检查它们的类型。

为了更有趣，用一个接受单个obj形参的构造函数替换有七个参数的构造函数，通过反射获得它的类，检查它的属性是否定义了自定义属性，并创建/编译与属性A、B和C的getter和setter相对应的LINQ表达式。这将允许您构造WrappedCX，而不会在调用中尾随丑陋的lambdas。这里的权衡是，现在lambdas将在运行时构造，因此缺少属性的潜在编译错误将成为运行时异常。

Answer 2

您可以动态生成代理类，使用带有属性的"PropName"名称访问正确的成员。您还希望在生成对属性的调用之前检测属性是否实际实现了get/set。也可能是一种更复杂的方法来保证生成的代理的唯一类型名称…

参见Main()的用法，Main下面是OperationWithProp1()

的实现

(这里有很多代码)

public interface IC
{
    string Prop1 { get; set; }
    string Prop2 { get; set; }
    string Prop3 { get; set; }
}
public class C1
{
    [PropName("Prop1")]
    public string A { get; set; }
    [PropName("Prop2")]
    public string B { get; set; }
    [PropName("Prop3")]
    public string C { get; set; }
}
public class C2
{
    [PropName("Prop1")]
    public string D { get; set; }
    [PropName("Prop2")]
    public string E { get; set; }
    [PropName("Prop3")]
    public string F { get; set; }
}
public class ProxyBuilder
{
    private static readonly Dictionary<Tuple<Type, Type>, Type> _proxyClasses = new Dictionary<Tuple<Type, Type>, Type>();
    private static readonly AssemblyName _assemblyName = new AssemblyName("ProxyBuilderClasses");
    private static readonly AssemblyBuilder _assemblyBuilder = AppDomain.CurrentDomain.DefineDynamicAssembly(_assemblyName, AssemblyBuilderAccess.RunAndSave);
    private static readonly ModuleBuilder _moduleBuilder = _assemblyBuilder.DefineDynamicModule(_assemblyName.Name, _assemblyName.Name + ".dll");
    public static void SaveProxyAssembly()
    {
        _assemblyBuilder.Save(_assemblyName.Name + ".dll");
    }
    public static Type GetProxyTypeForBackingType(Type proxyInterface, Type backingType)
    {
        var key = Tuple.Create(proxyInterface, backingType);
        Type returnType;
        if (_proxyClasses.TryGetValue(key, out returnType))
            return returnType;
        var typeBuilder = _moduleBuilder.DefineType(
            "ProxyClassProxies." + "Proxy_" + proxyInterface.Name + "_To_" + backingType.Name,
            TypeAttributes.Public | TypeAttributes.Sealed,
            typeof (Object),
            new[]
            {
                proxyInterface
            });
        //build backing object field
        var backingObjectField = typeBuilder.DefineField("_backingObject", backingType, FieldAttributes.Private);
        //build constructor
        var ctor = typeBuilder.DefineConstructor(MethodAttributes.Public, CallingConventions.Standard, new[] {backingType});
        var ctorIL = ctor.GetILGenerator();
        ctorIL.Emit(OpCodes.Ldarg_0);
        var ctorInfo = typeof (Object).GetConstructor(types: Type.EmptyTypes);
        ctorIL.Emit(OpCodes.Call, ctorInfo);
        ctorIL.Emit(OpCodes.Ldarg_0);
        ctorIL.Emit(OpCodes.Ldarg_1);
        ctorIL.Emit(OpCodes.Stfld, backingObjectField);
        ctorIL.Emit(OpCodes.Ret);
        foreach (var targetPropertyInfo in backingType.GetProperties(BindingFlags.Public | BindingFlags.Instance))
        {
            var propertyName = targetPropertyInfo.Name;
            var attributes = targetPropertyInfo.GetCustomAttributes(typeof (PropName), true);
            if (attributes.Length > 0 && attributes[0] != null)
                propertyName = (attributes[0] as PropName).Name;
            var propBuilder = typeBuilder.DefineProperty(propertyName, PropertyAttributes.HasDefault, targetPropertyInfo.PropertyType, null);
            const MethodAttributes getSetAttrs =
                MethodAttributes.Public | MethodAttributes.SpecialName | MethodAttributes.HideBySig | MethodAttributes.Final | MethodAttributes.Virtual;
            //build get method
            var getBuilder = typeBuilder.DefineMethod(
                "get_" + propertyName,
                getSetAttrs,
                targetPropertyInfo.PropertyType,
                Type.EmptyTypes);
            var getIL = getBuilder.GetILGenerator();
            getIL.Emit(OpCodes.Ldarg_0);
            getIL.Emit(OpCodes.Ldfld, backingObjectField);
            getIL.EmitCall(OpCodes.Callvirt, targetPropertyInfo.GetGetMethod(), Type.EmptyTypes);
            getIL.Emit(OpCodes.Ret);
            propBuilder.SetGetMethod(getBuilder);
            //build set method
            var setBuilder = typeBuilder.DefineMethod(
                "set_" + propertyName,
                getSetAttrs,
                null,
                new[] {targetPropertyInfo.PropertyType});
            var setIL = setBuilder.GetILGenerator();
            setIL.Emit(OpCodes.Ldarg_0);
            setIL.Emit(OpCodes.Ldfld, backingObjectField);
            setIL.Emit(OpCodes.Ldarg_1);
            setIL.EmitCall(OpCodes.Callvirt, targetPropertyInfo.GetSetMethod(), new[] {targetPropertyInfo.PropertyType});
            setIL.Emit(OpCodes.Ret);
            propBuilder.SetSetMethod(setBuilder);
        }
        returnType = typeBuilder.CreateType();
        _proxyClasses.Add(key, returnType);
        return returnType;
    }
    public static TIProxy CreateProxyObject<TIProxy>(object backingObject, out TIProxy outProxy) where TIProxy : class
    {
        var t = GetProxyTypeForBackingType(typeof (TIProxy), backingObject.GetType());
        outProxy = Activator.CreateInstance(t, backingObject) as TIProxy;
        return outProxy;
    }

    private static void Main(string[] args)
    {
        var c1 = new C1();
        IC c1Proxy;
        CreateProxyObject(c1, out c1Proxy);
        var c2 = new C2();
        IC c2Proxy;
        CreateProxyObject(c2, out c2Proxy);
        c1Proxy.Prop1 = "c1Prop1Value";
        Debug.Assert(c1.A.Equals("c1Prop1Value"));
        c2Proxy.Prop1 = "c2Prop1Value";
        Debug.Assert(c2.D.Equals("c2Prop1Value"));
        //so you can check it out in reflector
        SaveProxyAssembly();
    }
    private static void OperationWithProp1(object o)
    {
        IC proxy;
        CreateProxyObject(o, out proxy);
        string prop1 = proxy.Prop1;
        // Do something with prop1
    }

Answer 3

在我看来，使用重载是为了清晰和可维护性。如果有很多重叠的代码，把它分解成一个单独的方法。

也就是说，我假设您首先关心的是可维护性，因为您没有提到速度。

Answer 4

为了获得最佳性能，您应该为每个属性编写一对静态方法，格式为:

[PropName("Prop1")]
static string Prop1Getter(thisType it) { return it.WhateverProperty; }
[PropName("Prop1")]
static string Prop1Setter(thisType it, string st) { it.WhateverProperty = st; }

我建议你使用反射来生成委托，并使用静态泛型类来缓存它们。实际上，您将拥有一个私有静态类PropertyAccessors<T>，其委托声明如下:

const int numProperties = 3;
public Func<T, string>[] Getters;
public Action<T, string>[] Setters;

静态构造函数会做如下操作:

Getters = new Func<T, string>[numProperties];
Setters = new Action<T, string>[numProperties];
for (int i = 0; i< numProperties; i++)
{
  int ii = i;  // Important--ensure closure is inside loop
  Getters[ii] = (T it) => FindSetAndRunGetter(ii, it);
  Setters[ii] = (T it, string st) => FindSetAndRunSetter(ii, it, st);
}

FindSetAndRunGetter(ii,it)方法应该搜索一个合适的属性getter，如果找到了，设置Getters[ii]指向合适的属性getter，运行一次，并返回结果。FindSetAndRunSetter(ii, it, st)也应该使用属性设置器，使用st作为参数运行一次。

使用这种方法将结合使用反射的多功能性和"自动升级"(意味着能够在未来的类中自动找到方法)，其速度与硬编码方法相当(如果不是更好的话)。一个麻烦的地方是需要像上面描述的那样定义静态方法。也许可以使用Reflection.Emit来自动生成包含这些方法的静态类，但这超出了我的专业水平。