问题
For properties there are GetGetMethod and GetSetMethod so that I can do:
Getter = (Func<S, T>)Delegate.CreateDelegate(typeof(Func<S, T>),
propertyInfo.GetGetMethod());
and
Setter = (Action<S, T>)Delegate.CreateDelegate(typeof(Action<S, T>),
propertyInfo.GetSetMethod());
But how do I go about FieldInfos?
I am not looking for delegates to GetValue and SetValue (which means I will be invoking reflection each time)
Getter = s => (T)fieldInfo.GetValue(s);
Setter = (s, t) => (T)fieldInfo.SetValue(s, t);
but if there is a CreateDelegate approach here? I mean since assignments return a value, can I treat assignments like a method? If so is there a MethodInfo handle for it? In other words how do I pass the right MethodInfo of setting and getting a value from a member field to CreateDelegate method so that I get a delegate back with which I can read and write to fields directly?
Getter = (Func<S, T>)Delegate.CreateDelegate(typeof(Func<S, T>), fieldInfo.??);
Setter = (Action<S, T>)Delegate.CreateDelegate(typeof(Action<S, T>), fieldInfo.??);
I can build expression and compile it, but I am looking for something simpler. In the end I don't mind going the expression route if there is no answer for the asked question, as shown below:
var instExp = Expression.Parameter(typeof(S));
var fieldExp = Expression.Field(instExp, fieldInfo);
Getter = Expression.Lambda<Func<S, T>>(fieldExp, instExp).Compile();
if (!fieldInfo.IsInitOnly)
{
var valueExp = Expression.Parameter(typeof(T));
Setter = Expression.Lambda<Action<S, T>>(Expression.Assign(fieldExp, valueExp), instExp, valueExp).Compile();
}
Or am I after the nonexistent (since I have nowhere seen something like that yet) ?
回答1:
Field access isn't performed via a method (like getters and setters)--it's performed with an IL instruction--so there's nothing you can assign to a delegate. you'll have to use the expression route to create a "block" of code (effectively IL) that can be assigned to a delegate.
回答2:
As Peter Ritchie suggested, you can compile your own code at runtime. The method will be compiled as soon as you invoke the delegate for the first time. So the first call will be slow, but any subsequent call will be as fast as you can get in .NET without unmanaged pointers/unions. Except for the first call, the delegate is around 500 times faster than FieldInfo directly.
class DemoProgram
{
class Target
{
private int value;
}
static void Main(string[] args)
{
FieldInfo valueField = typeof(Target).GetFields(BindingFlags.NonPublic| BindingFlags.Instance).First();
var getValue = CreateGetter<Target, int>(valueField);
var setValue = CreateSetter<Target, int>(valueField);
Target target = new Target();
setValue(target, 42);
Console.WriteLine(getValue(target));
}
static Func<S, T> CreateGetter<S, T>(FieldInfo field)
{
string methodName = field.ReflectedType.FullName + ".get_" + field.Name;
DynamicMethod setterMethod = new DynamicMethod(methodName, typeof(T), new Type[1] { typeof(S) }, true);
ILGenerator gen = setterMethod.GetILGenerator();
if (field.IsStatic)
{
gen.Emit(OpCodes.Ldsfld, field);
}
else
{
gen.Emit(OpCodes.Ldarg_0);
gen.Emit(OpCodes.Ldfld, field);
}
gen.Emit(OpCodes.Ret);
return (Func<S, T>)setterMethod.CreateDelegate(typeof(Func<S, T>));
}
static Action<S, T> CreateSetter<S,T>(FieldInfo field)
{
string methodName = field.ReflectedType.FullName+".set_"+field.Name;
DynamicMethod setterMethod = new DynamicMethod(methodName, null, new Type[2]{typeof(S),typeof(T)},true);
ILGenerator gen = setterMethod.GetILGenerator();
if (field.IsStatic)
{
gen.Emit(OpCodes.Ldarg_1);
gen.Emit(OpCodes.Stsfld, field);
}
else
{
gen.Emit(OpCodes.Ldarg_0);
gen.Emit(OpCodes.Ldarg_1);
gen.Emit(OpCodes.Stfld, field);
}
gen.Emit(OpCodes.Ret);
return (Action<S, T>)setterMethod.CreateDelegate(typeof(Action<S, T>));
}
}
Keep in mind that structs are passed by value. That means an Action<S, T> can not be used to change members of a struct if it is passed by value as the first argument.
回答3:
[2019 edit: Since this post has always been one of my favorites, it is bittersweet to note that the approach I show here has been wholly superseded, in my own projects, by a newer, entirely different, and much sleeker technique, which I detail at this answer].
Using the new “ref return” feature in C# 7.0 can make the process of creating and using runtime dynamically-generated get/set accessors much simpler and syntactically transparent. Instead of having to use DynamicMethod to emit separate getter and setter functions for accessing the field, you can now have a single method that returns a managed pointer-type reference to the field, essentially a single accessor that (in turn) enables convenient, ad-hoc get a̲n̲d̲ set access. Below, I provide a helper utility function which simplifies generating a ByRef getter function for any arbitrary (i.e. private) instance field in any class.
➜ For “just the code,” skip to note below.
As a running example, let's say we want to access a private instance field m_iPrivate, an int defined in the class OfInterestClass:
public class OfInterestClass
{
private int m_iPrivate;
};
Next let's assume we have a static field “reference-getter” function that takes a OfInterestClass instance and returns the desired field value by reference using the new C# 7 “ref return” capability (below, I'll provide code to generate such functions at runtime, via DynamicMethod):
public static ref int __refget_m_iPrivate(this OfInterestClass obj)
{
/// ...
}
Such a function (“ref-getter,” let's say) is all we need in order to to have full read/write access to the private field. In the following examples, note especially the setter-invoking operation—and the demonstrations of using the (i.e.) ++ and += operators—since applying those operators directly to a method call may look a little unusual if you're not up-to-speed on C#7.
void MyFunction(OfInterestClass oic)
{
int the_value = oic.__refget_m_iPrivate(); // 'get'
oic.__refget_m_iPrivate() = the_value + 100; // 'set'
/// or simply...
oic.__refget_m_iPrivate() += 100; // <-- yes, you can
oic.__refget_m_iPrivate()++; // <-- this too, no problem
ref int prv = ref oic.__refget_m_iPrivate(); // via "ref-local" in C#7
prv++;
foo(ref prv); // all of these directly affect…
prv = 999; // …field m_iPrivate 'in-situ'
}
As is the point, every operation shown in these examples manipulates m_iPrivate in situ (i.e., directly within its containing instance oic) such that any and all changes are publicly visible there immediately. It's important to realize that this means that prv, despite being int-typed and locally declared, does not behave like your typical “local” variable. This is especially significant for concurrent code; not only are changes visible b̲e̲f̲o̲r̲e̲ MyFunction has exited, but now with C# 7, callers have the ability to retain a ref return managed pointer (as a ref local) and thus continue modifying the target for an arbitrarily long time a̲f̲t̲e̲r̲wards (albeit necessarily remaining below the ref-obtaining stack frame, that is).
Of course a main and obvious advantage of using a managed pointer here—and elsewhere in general—is that it continues to remain valid (again, within its stack frame's lifetime), even as oic—itself a reference-type instance allocated in the GC heap—may be moved around during garbage collection. This is a gigantic difference versus native pointers.
As sketched above, the ref-getter is a static extension method that can be declared and/or used from anywhere. But if you're able to create your own class that's derived from OfInterestClass (that is, if OfInterestClass isn't sealed), you can make this even nicer. In a derived class, you can expose C# syntax for using the base class's private field as if it were a public field of your derived class. To do this, just add a C# read-only ref return property to your class which binds the static ref-getter method to the current instance this:
public ref int m_iPrivate => ref __refget_m_iPrivate(this);
Here, the property is made public so anybody can access the field (via a reference to our derived class). We've essentially publicly published the private field from the base class. Now, in the derived class (or elsewhere, as appropriate) you can do any or all of the following:
int v = m_iPrivate; // get the value
m_iPrivate = 1234; // set the value
m_iPrivate++; // increment it
ref int pi = ref m_iPrivate; // reference as C# 7 ref local
v = Interlocked.Exchange(ref m_iPrivate, 9999); // even do in-situ atomic operations on it!
As you can see, because the property, like the earlier method, also has a by reference return value, it behaves almost exactly like a field does.
So now for the details. How do you create the static ref-getter function that I showed above? Using DynamicMethod, this should be trivial. For example, here is the IL code for a traditional (by-value) static getter function:
// static int get_iPrivate(OfInterestClass oic) => oic.m_iPrivate;
IL_0000: ldarg.0
IL_0001: ldfld Int32 m_iPrivate/OfInterestClass
IL_0006: ret
And here is the IL code that we want instead (ref-return):
// static ref int refget_iPrivate(OfInterestClass oic) => ref oic.m_iPrivate;
IL_0000: ldarg.0
IL_0001: ldfld̲a Int32 m_iPrivate/OfInterestClass
IL_0006: ret
The only difference from the by-value getter is that we are using the ldflda (load field address) opcode instead of ldfld (load field). So if you're well-practiced with DynamicMethod it should be no problem, right?
Wrong!...
unfortunatelyDynamicMethoddoes not allow a by-ref return value!
If you try to call the DynamicMethod constructor specifying a ByRef type as the return value...
var dm = new DynamicMethod(
"", // method name
typeof(int).MakeByRefType(), // by-ref return type <-- ERROR
new[] { typeof(OfInterestClass) }, // argument type(s)
typeof(OfInterestClass), // owner type
true); // private access
...the function throws NotSupportedException with the following message:
The return Type contains some invalid type (i.e. null, ByRef)
Apparently, this function did not get the memo on C#7 and ref-return. Fortunately, I found a simple workaround that gets it working. If you pass a non-ref type into the constructor as a temporary "dummy," but then immediately afterwards use reflection on the newly-created DynamicMethod instance to change its m_returnType private field to be the ByRef-type type (sic.) that you actually want, then everything seems to work just fine.
To speed things up, I'll cut to the completed generic method which automates the whole process of by creating/returning a static ref-getter function for the private instance field of type U, having the supplied name, and defined in class T.
If you just want the complete working code, copy from below this point to the end
First we have to define a delegate that represents the ref-getter, since a Func<T,TResult> delegate with ByRef usage can't be declared. Fortunately, the older delegate syntax does work for doing so (phew!).
public delegate ref U RefGetter<T, U>(T obj);
Place the delegate, along with the following static function in a centralized utility class where both can be accessed throughout your project. Here's the final ref-getter creation function which can be used to create a static ref-getter for the so-named instance field in any class.
public static RefGetter<T, U> create_refgetter<T, U>(String s_field)
{
const BindingFlags bf = BindingFlags.NonPublic |
BindingFlags.Instance |
BindingFlags.DeclaredOnly;
var fi = typeof(T).GetField(s_field, bf);
if (fi == null)
throw new MissingFieldException(typeof(T).Name, s_field);
var s_name = "__refget_" + typeof(T).Name + "_fi_" + fi.Name;
// workaround for using ref-return with DynamicMethod:
// a.) initialize with dummy return value
var dm = new DynamicMethod(s_name, typeof(U), new[] { typeof(T) }, typeof(T), true);
// b.) replace with desired 'ByRef' return value
dm.GetType().GetField("m_returnType", bf).SetValue(dm, typeof(U).MakeByRefType());
var il = dm.GetILGenerator();
il.Emit(OpCodes.Ldarg_0);
il.Emit(OpCodes.Ldflda, fi);
il.Emit(OpCodes.Ret);
return (RefGetter<T, U>)dm.CreateDelegate(typeof(RefGetter<T, U>));
}
Returning now to outset of this article, we can easily provide the __refget_m_iPrivate function that got everything started. Instead of a static function written directly in C#, we will use the static ref-getter creation function to create the function body at runtime and store it in a static delegate-typed field (with the same signature). The syntax for calling it in the instance property (as shown above, and repeated below) or elsewhere is the same as if the compiler had been able to write the function.
Finally, to cache the dynamically-created ref-getter delegate, place the following line in any static class of your choice. Replace OfInterestClass with the type of the base class, int with the field type of the private field, and change the string argument to match the name of the private field. If you aren't able to create your own class derived from OfInterestClass (or don't want to), you're done; just make this field public and you can call it like a function, passing any OfInterestClass instance to get a reference which lets you read, write, or monitor its int-valued private field "m_iPrivate."
// Static delegate instance of ref-getter method, statically initialized.
// Requires an 'OfInterestClass' instance argument to be provided by caller.
static RefGetter<OfInterestClass, int> __refget_m_iPrivate =
create_refgetter<OfInterestClass, int>("m_iPrivate");
Optionally, if you want to publish the hidden field with a cleaner or more natural syntax, you can define a (non-static) proxy class of your own which either contains an instance of—or perhaps even better (if possible), derives from—the field hiding class OfInterestClass. Instead of deploying the line of code previously shown globally in a static class, place it in your proxy class instead, and then also add the following line:
// optional: ref-getter as an instance property (no 'this' argument required)
public ref int m_iPrivate => ref __refget_m_iPrivate(this);
回答4:
No there is no easy way to create a delegate to get/set a field.
You will have to make your own code to provide that functionallity. I would suggest two functions in a shared library to provide this.
Using your code (in this example I only show the creation of the get-delegate):
static public class FieldInfoExtensions
{
static public Func<S, T> CreateGetFieldDelegate<S, T>(this FieldInfo fieldInfo)
{
var instExp = Expression.Parameter(typeof(S));
var fieldExp = Expression.Field(instExp, fieldInfo);
return Expression.Lambda<Func<S, T>>(fieldExp, instExp).Compile();
}
}
This makes it easy to create a get-delegate from a FieldInfo (assuming the field is of type int):
Func<MyClass, int> getter = typeof(MyClass).GetField("MyField").CreateGetFieldDelegate<MyClass, int>();
Or if we change your code a little bit:
static public class TypeExtensions
{
static public Func<S, T> CreateGetFieldDelegate<S, T>(this Type type, string fieldName)
{
var instExp = Expression.Parameter(type);
var fieldExp = Expression.Field(instExp, fieldName);
return Expression.Lambda<Func<S, T>>(fieldExp, instExp).Compile();
}
}
This makes it even more easy:
Func<MyClass, int> getter = typeof(MyClass).CreateGetFieldDelegate<MyClass, int>("MyField");
It is also possible to create these delegates using IL, but that code would be more complex and does not have much more performance, if any.
回答5:
I'm not aware if you would use Expression, then why avoiding reflection? Most operations of Expression rely on reflection.
GetValue and SetValue themself are the get method and set method, for the fields, but they are not for any specific field.
Fields are not like properties, they are fields, and there's no reason to generate get/set methods for each one. However, the type may vary with different field, and thus GetValue and SetValue are defined the parameter/return value as object for variance. GetValue is even a abstract method, that is, for every class(still reflection) overriding it, must be within the identical signature.
If you don't type them, then the following code should do:
public static void SomeMethod(FieldInfo fieldInfo) {
var Getter=(Func<object, object>)fieldInfo.GetValue;
var Setter=(Action<object, object>)fieldInfo.SetValue;
}
but if you want, there's a constrained way:
public static void SomeMethod<S, T>(FieldInfo fieldInfo)
where S: class
where T: class {
var Getter=(Func<S, object>)fieldInfo.GetValue;
var Setter=(Action<S, T>)fieldInfo.SetValue;
}
For the reason that the Getter still be Func<S, object>, you might want to have a look of:
Covariance and Contravariance in C#, Part Three: Method Group Conversion Variance on Mr. Lippert's blog.
回答6:
Here is an another option for creating a delegate when you are working with objects (don't know specific type of a field). Though it is slower if field is a structure (because of boxing).
public static class ReflectionUtility
{
public static Func<object, object> CompileGetter(this FieldInfo field)
{
string methodName = field.ReflectedType.FullName + ".get_" + field.Name;
DynamicMethod setterMethod = new DynamicMethod(methodName, typeof(object), new[] { typeof(object) }, true);
ILGenerator gen = setterMethod.GetILGenerator();
if (field.IsStatic)
{
gen.Emit(OpCodes.Ldsfld, field);
gen.Emit(field.FieldType.IsClass ? OpCodes.Castclass : OpCodes.Box, field.FieldType);
}
else
{
gen.Emit(OpCodes.Ldarg_0);
gen.Emit(OpCodes.Castclass, field.DeclaringType);
gen.Emit(OpCodes.Ldfld, field);
gen.Emit(field.FieldType.IsClass ? OpCodes.Castclass : OpCodes.Box, field.FieldType);
}
gen.Emit(OpCodes.Ret);
return (Func<object, object>)setterMethod.CreateDelegate(typeof(Func<object, object>));
}
public static Action<object, object> CompileSetter(this FieldInfo field)
{
string methodName = field.ReflectedType.FullName + ".set_" + field.Name;
DynamicMethod setterMethod = new DynamicMethod(methodName, null, new[] { typeof(object), typeof(object) }, true);
ILGenerator gen = setterMethod.GetILGenerator();
if (field.IsStatic)
{
gen.Emit(OpCodes.Ldarg_1);
gen.Emit(field.FieldType.IsClass ? OpCodes.Castclass : OpCodes.Unbox_Any, field.FieldType);
gen.Emit(OpCodes.Stsfld, field);
}
else
{
gen.Emit(OpCodes.Ldarg_0);
gen.Emit(OpCodes.Castclass, field.DeclaringType);
gen.Emit(OpCodes.Ldarg_1);
gen.Emit(field.FieldType.IsClass ? OpCodes.Castclass : OpCodes.Unbox_Any, field.FieldType);
gen.Emit(OpCodes.Stfld, field);
}
gen.Emit(OpCodes.Ret);
return (Action<object, object>)setterMethod.CreateDelegate(typeof(Action<object, object>));
}
}
回答7:
ref return restriction in DynamicMethod seems to be gone at least on v4.0.30319.
Modified this answer by Glenn Slayden with info from microsoft docs:
using System;
using System.Reflection;
using System.Reflection.Emit;
namespace ConsoleApp {
public class MyClass {
private int privateInt = 6;
}
internal static class Program {
private delegate ref TReturn OneParameter<TReturn, in TParameter0>(TParameter0 p0);
private static void Main() {
var myClass = new MyClass();
var method = new DynamicMethod(
"methodName",
typeof(int).MakeByRefType(), // <- MakeByRefType() here
new[] {typeof(MyClass)},
typeof(MyClass),
true); // skip visibility
const BindingFlags bindings = BindingFlags.Instance | BindingFlags.Public | BindingFlags.NonPublic;
var il = method.GetILGenerator();
il.Emit(OpCodes.Ldarg_0);
il.Emit(OpCodes.Ldflda, typeof(MyClass).GetField("privateInt", bindings));
il.Emit(OpCodes.Ret);
var getPrivateInt = (OneParameter<int, MyClass>) method.CreateDelegate(typeof(OneParameter<int, MyClass>));
Console.WriteLine(typeof(string).Assembly.ImageRuntimeVersion);
Console.WriteLine(getPrivateInt(myClass));
}
}
}
Outputs:
6
来源：https://stackoverflow.com/questions/16073091/is-there-a-way-to-create-a-delegate-to-get-and-set-values-for-a-fieldinfo