Improving methods with variable parameters using .NET Generics - c#

I have a lot of functions which are currently overloaded to operate on int and string:
bool foo(int);
bool foo(string);
bool bar(int);
bool bar(string);
void baz(int p);
void baz(string p);
I then have a lot of functions taking 1, 2, 3, or 4 arguments of either int or string, which call the aforementioned functions:
void g(int p1) { if(foo(p1)) baz(p1); }
void g(string p1) { if(foo(p1)) baz(p1); }
void g(int p2, int p2) { if(foo(p1)) baz(p1); if(bar(p2)) baz(p2); }
void g(int p2, string p2) { if(foo(p1)) baz(p1); if(bar(p2)) baz(p2); }
void g(string p2, int p2) { if(foo(p1)) baz(p1); if(bar(p2)) baz(p2); }
void g(string p2, string p2) { if(foo(p1)) baz(p1); if(bar(p2)) baz(p2); }
// etc.
Note: The implementation of the g() family is just an example
More types than the current int or string might be introduced at any time. The same goes for functions with more arguments than 4. The current number of identical functions is barely manageable. Add one more variant in either dimension and the combinatoric explosion will be so huge, it might blow away the application.
In C++, I'd templatize g() and be done.
I understand that .NET generics are different. I have been fighting them for two hours now trying to come up with a solution that doesn't involve copy and pasting code, to no avail.
C# generics won't require me to type out identical code for a family of functions taking five arguments of either of three types?
What am I missing?
Edit: These functions are used to parse a bunch of arguments (currently either int or string) from some source. Imagine bar() and baz() being able to read both int or string, and the g() family specifying the type and number of arguments to parse (implicitly, by their arguments' types).

Consider using inheritance for this case. I am assuming that foo, bar and baz are inherent to the type (int or string in your case). If this is not true please correct or comment this answer.
using System;
namespace ConsoleApplication3
{
abstract class Param
{
public abstract bool Foo();
public abstract bool Bar();
public abstract void Baz();
public static IntParam Create(int value)
{
return new IntParam(value);
}
public static StringParam Create(string value)
{
return new StringParam(value);
}
}
abstract class Param<T> : Param {
private T value;
protected Param() { }
protected Param(T value) { this.value = value; }
public T Value {
get { return this.value; }
set { this.value = value; }
}
}
class IntParam : Param<int>
{
public IntParam() { }
public IntParam(int value) : base(value) { }
public override bool Foo() { return true; }
public override bool Bar() { return true; }
public override void Baz()
{
Console.WriteLine("int param value is " + this.Value);
}
}
class StringParam : Param<string>
{
public StringParam() { }
public StringParam(string value) : base(value) { }
public override bool Foo() { return true; }
public override bool Bar() { return true; }
public override void Baz()
{
Console.WriteLine("String param value is " + this.Value);
}
}
class Program
{
static void g(Param p1)
{
if (p1.Foo()) { p1.Baz(); }
}
static void g(Param p1, Param p2)
{
if (p1.Foo()) { p1.Baz(); }
if (p2.Bar()) { p2.Baz(); }
}
static void Main(string[] args)
{
Param p1 = Param.Create(12);
Param p2 = Param.Create("viva");
g(p1);
g(p2);
g(p1, p1);
g(p1, p2);
g(p2, p1);
g(p2, p2);
Console.ReadKey();
}
}
}
This would output:
int param value is 12
String param value is viva
int param value is 12
int param value is 12
int param value is 12
String param value is viva
String param value is viva
int param value is 12
String param value is viva
String param value is viva
For a new supported type you:
create a new class that supports the type and extends Param<T>;
implement Foo, Bar and Baz for that new type;
Create a new g method (just one) that has another parameter.
Specially for 3) this would greatly reduce explosion of methods. Now you write a single g method for any given number of parameters. With previous design you had to write, for n parameters, 2^n methods (n = 1 -> 2 methods, n = 2 -> 4 methods, n = 3 -> 8 methods, ..).

Your true problem here is most likely one of design, rather than something generics can be used for. Generics should be used for things that are actually type-agnostic, not as a catch-all to make life a bit easier. Perhaps try posting some actual example code you're using, and someone may have an idea about how to redesign your solution in a way that will allow you to extend it without so much headache.
As a teaser, consider something like this:
public void DoSomethingConditionally<T>(T key, Func<T, bool> BooleanCheck, Action<T> WhatToDo)
{
if (BooleanCheck(key)) WhatToDo(key);
}
And you could call it like this:
DoSomethingConditionally<String>("input", v => v == "hello", s => Console.WriteLine(s));
I've used lambda expressions here, but you could just as easily predefine a few Func<>s that perform some common expressions. This would be a much better pattern than method overloading, and would force you to handle new input types at design time.

Not as optimal as I would like... but what if foo, bar, and baz had generic versions as well?
static bool foo(int input)
{
return input > 5;
}
static bool foo(string input)
{
return input.Length > 5;
}
static void baz(int input)
{
Console.WriteLine(input);
}
static void baz(string input)
{
Console.WriteLine(input);
}
static bool foo<T>(T input)
{
if (input is int) return foo((int)(object)input);
if (input is string) return foo((string)(object)input);
return false;
}
static void baz<T>(T input)
{
if (input is int) baz((int)(object)input);
else if (input is string) baz((string)(object)input);
else throw new NotImplementedException();
}
static void g<T>(T input)
{
if (foo(input))
baz(input);
}
static void g<T, U>(T input, U inputU)
{
g(input);
g(inputU);
}

Use a list of objects.
In the case the number of parameters are unknown at planning time, just use a list of objects. Something like:
void g(params object[] args) {
foreach (object arg in args) {
if ((arg is int) && (foo((int)arg))) baz((int)arg) else
if ((arg is string) && (foo((string)arg))) baz((string)arg)
}
}
(Assuming you have bool foo(int), bool foo(string)...)
So you can call:
g(p1, p2);
g(p1);
g(p1, p2, p3)
with any combination of the types, since every reference derives from object (which it could be many more types than required, int and string, but could be handy in future to support more other types).
This is possible since you could use Reflection to recognize the type at runtime.
Another way to execute a sequence of operation is the use of interfaces, defining the action to execute at certain conditions, on certain objects.
interface IUpdatable {
void Update(object[] data);
}
class object1 : IUpdatable { public void Update(object data) { baz(data); } }
class object2 : IUpdatable { public void Update(object data) { baz(data); } }
void g(params IUpdatable[] args) {
foreach (IUpdatable arg in args) {
arg.Update(args);
}
}
But this way you have to model p1 and p2 (but also p3, as objects implementing an interface, which could be not possible.

I would have done this as a comment to #smink, but I don't have enough rep...
If you extend the Param base class to have implicit operators you are back to not having to wrap the contents in code (though the runtime still incurs the wrapping overhead)...
abstract class Param
{
...
public static implicit operator Param(int value)
{ return new IntParam(value); }
}

If you are using C# / .NET 4.0, you can achieve multiple dispatch using the dynamic feature so you only have to implement a single overload of g based on the number of arguments and the proper foo/bar/baz overloads by type inside each g implementation will be resolved at runtime.
void g(dynamic p1) { if (foo(p1)) baz(p1); }
void g(dynamic p1, dynamic p2) { if (foo(p1)) baz(p1); if (bar(p2)) baz(p2); }
Edit:
Even though you are unable to use C# / .NET 4.0, you can still use this approach using reflection. I've added another foo/bar/baz overload for double showing how well this generalizes and allows you to eliminate duplicate g implementations.
bool foo(int p) {Console.WriteLine("foo(int)=" + p); return p == 0;}
bool foo(string p) {Console.WriteLine("foo(string)=" + p); return p == "";}
bool foo(double p) { Console.WriteLine("foo(double)=" + p); return p == 0.0; }
bool bar(int p) {Console.WriteLine("bar(int)=" + p); return p == 1;}
bool bar(string p) { Console.WriteLine("bar(string)=" + p); return p == ""; }
bool bar(double p) { Console.WriteLine("bar(double)=" + p); return p == 1.1; }
void baz(int p) {Console.WriteLine("baz(int)=" + p);}
void baz(string p) { Console.WriteLine("baz(string)=" + p); }
void baz(double p) { Console.WriteLine("baz(double)=" + p); }
//these object overloads of foo/bar/baz allow runtime overload resolution
bool foo(object p)
{
if(p == null) //we need the type info from an instance
throw new ArgumentNullException();
//may memoize MethodInfo by type of p
MethodInfo mi = typeof(Program).GetMethod(
"foo",
BindingFlags.Instance | BindingFlags.InvokeMethod | BindingFlags.NonPublic,
null,
new Type[] { p.GetType() },
null
);
if (mi.GetParameters()[0].ParameterType == typeof(object))
throw new ArgumentException("No non-object overload found");
return (bool)mi.Invoke(this, new object[] { p });
}
bool bar(object p)
{
if (p == null)
throw new ArgumentNullException();
MethodInfo mi = typeof(Program).GetMethod(
"bar",
BindingFlags.Instance | BindingFlags.InvokeMethod | BindingFlags.NonPublic,
null,
new Type[] { p.GetType() },
null
);
if (mi.GetParameters()[0].ParameterType == typeof(object))
throw new ArgumentException("No non-object overload found");
return (bool)mi.Invoke(this, new object[] { p });
}
void baz(object p)
{
if (p == null)
throw new ArgumentNullException();
MethodInfo mi = typeof(Program).GetMethod(
"baz",
BindingFlags.Instance | BindingFlags.InvokeMethod | BindingFlags.NonPublic,
null,
new Type[] { p.GetType() },
null
);
if (mi.GetParameters()[0].ParameterType == typeof(object))
throw new ArgumentException("No non-object overload found");
mi.Invoke(this, new object[] { p });
}
//now you don't need to enumerate your identical implementations of g by type
void g(object p1) { if (foo(p1)) baz(p1); }
void g(object p1, object p2) { if (foo(p1)) baz(p1); if (bar(p2)) baz(p2); }

Unfortunately, generics cannot handle this situation. At least, not well. If you make your methods generic, then just about any type can be passed into them. There won't be an adequate where clause on the generics to limit it to just string and int. If your methods are going to do specific int/string related operations inside them, then generics wont work at all.
Generics in C# are not nearly as powerful as templates in C++, and yes they can cause some major headaches at time. It just takes time to get used to them and get a feel for what they can and cannot do.

This might be a bit heavy handed, but would encapsulating the different parameters types as classes work?:
public abstract class BaseStuff
{
public abstract bool Foo();
public abstract bool Bar();
public abstract void Baz();
public void FooBaz()
{
if(Foo()) Baz();
}
public void BarBaz()
{
if(Bar()) Baz();
}
}
public class IntStuff : BaseStuff
{
private int input;
public IntStuff(int input)
{
this.input = input;
}
public bool Foo()
{
//logic using input for example
return input > 0;
}
//implement Bar and Baz using input
}
public class StringStuff : BaseStuff
{
private string input;
public IntStuff(string input)
{
this.input = input;
}
//Implement Foo, Bar and Baz
}
And then have some G methods somewhere:
public void G(BaseStuff stuff1)
{
stuff1.FooBaz();
}
public void G(BaseStuff stuff1, BaseStuff stuff2)
{
stuff1.FooBaz();
stuff2.BarBaz();
}
And you can then call with:
G(new IntStuff(10), new StringStuff("hello"));
G(new StringStuff("hello"), new StringStuff("world"));

You can use code generation to solve this.
Look at Reflection.Emit. You can also generate code with T4 in Visual Studio.
The types are really getting in the way here. You could also try to solve this with a dynamic language or with the C# 4 dynamic keyword.

If you are using c# 4.0 you can do this with the option parameter
Or you can use object
Foo(object o)
{
if (o is int){ }
else if (o is string){ }
}
Or you can use the generic method Foo<T>(T o){ }

Related

C# automatic casting when calling overloaded function

Let's imagine that I have the following overloaded function:
void DoSomething(int x) { ... }
void DoSomething(float x) { ... }
void DoSomething(decimal x) { ... }
In the following method, I need to call the correct overload. This is how a simple implementation would look like:
void HaveToDoSomething(object data)
{
if (data is int) DoSomething((int)data);
else if (data is float) DoSomething((float)data);
else if (data is decimal) DoSomething((decimal)data);
}
This is tedious when there are ~20 types to check. Is there a better way of doing all this casting automatically?
Something I forgot to mention: DoSomething wouldn't work with generics, as each type needs to be handled differently, and I only know the type at runtime.
One possible approach would be to use dynamic:
void HaveToDoSomething(dynamic data)
{
DoSomething(data);
}
You can use Reflection but it can have a performance impact:
public class Example
{
void DoSomething(int i)
{
}
void DoSomething(float i)
{
}
}
public static class ExampleExtensions
{
public static void DoSomethingGeneric(this Example example, object objectParam)
{
var t = objectParam.GetType();
var methods = typeof(example).GetMethods().Where(_ => _.Name == "DoSomething");
var methodInfo = methods.Single(_ => _.GetParameters().First().ParameterType == t);
methodInfo.Invoke(example, new[] { objectParam });
}
}

Operator overloading with generic function parameters

I have a problem with operator resolution on generic methods.
From my understanding of section 7.3.4 of the spec within the function EqualOperatorGeneric (sample code below) the correct overload of the == operator on the type A should be found, but instead it seems to get the candidate for (object, object).
Am I doing something very obvious wrong? Is there a method to get the expected behaviour and if not can I turn the given case into a compile time or runtime error?
public class A
{
public A(int num)
{
this.Value = num;
}
public int Value { get; private set; }
public override bool Equals(object obj)
{
var other = obj as A;
if (Object.ReferenceEquals(other, null))
return false;
return Object.Equals(this.Value, other.Value);
}
public override int GetHashCode()
{
return this.Value.GetHashCode();
}
public static bool operator ==(A l, A r)
{
if (Object.ReferenceEquals(l, null))
{
return !Object.ReferenceEquals(r, null);
}
return l.Equals(r);
}
public static bool operator !=(A l, A r)
{
return !(l == r);
}
}
class Program
{
static void Main(string[] args)
{
Console.WriteLine(EqualOperatorGeneric(new A(1), new A(1)));
}
public static bool EqualOperatorGeneric<L, R>(L l, R r)
where L : class
where R : class
{
return l == r;
}
}
Output:
False
When EqualOperatorGeneric is compiled the == operator needs to be bound statically, when the method is compiled, to a single implementation. It is not bound separately for each separate usage of the generic method.
This is what differentiates generics from, say, C++ templates. The generic method is compiled once and then applied to every usage with every set of type arguments, whereas templates are compiled separately for each set of generic arguments.
After scrounging the spec I realized you can use the dynamic keyword to defer binding of the operator from compile time to runtime. This fixes the issues I have been having:
public static bool EqualOperatorGeneric<L, R>(L l, R r)
{
dynamic dl = l, dr = r;
return dl == dr;
}

C# - Passing all the method arguments to another method?

I've got a class, which has three overloaded methods. Let's say, there is:
class MyChildClass
{
public void myMethod(int i)
{ /* do something with i */ }
public void myMethod(int a, string b)
{ /* get i from a and b and call: */ myMethod(i); }
public void myMethod(string c, string d)
{ /* get i from c and d and call: */ myMethod(i); }
}
Now I'd like this class to be a private field in other (parent) class, but I need those three methods to be accessible. For now, I just did:
class MyBaseClass
{
private MyChildClass myObject = new myChildClass(); // or similar
public void myMethod(int i)
{ myObject.myMethod(i); }
public void myMethod(int a, string b)
{ myObject.myMethod(a, b); }
public void myMethod(string c, string s)
{ myObject.myMethod(c, d); }
}
Is there a way to implement it as one short method? Something which would look like:
public void myMethod(unknownListOfArgumentsOfDifferentTypes args)
{ myObject.myMethod(args); }
I tried to use public void myMethod(params object[] something) but it didn't work. Is it possible, or do I have to "project" every method into another?
EDIT: Child class has various methods and fields, which I want to be accessible for parent class only. That's why I don't want parent to derive after it. I didn't explain that, sorry if it looked like child class contains only those three methods. Those are the methods I want to be accessible as public methods of parent class.
Why don't you do
class MyChildClass : MyBaseClass
{
}
same effect, less code, and this way MyChildClass is a MyBaseClass
If you implement some sort of generic facade using reflection you'll just be reducing performance, bypassing the benefits of type safety and delaying the discovery of problems.
You'll also have have a "has a" relationship instead "is a" relationship which is incongruent with your class names.
If you want to give up this simplicty with its associated benefits you could make use the GetMethodBySig extension accepted in this post.
Something like this,
class SemiGenericFacade<T> where T : new()
{
private readonly t = new T();
public void CallVoidOnT(string name, params object[] parameters)
{
var paramTypes = parameters.Select(p => typeof(p))
var match = typeof(T).GetMethodBySig(typeof(void), paramTypes)
.Single(mi => mi.Name == name);
match.Invoke(this.t, parameters);
}
}
Following on from Piotr Justyna's comment, implementing and using this method results in the cat turning into a tiger and eating her kittens.
If you were to do this it would make sense to add to the linked extension
public static class Extensions
{
public static MethodInfo GetMethodByNameThenSig(
this Type type,
string name,
Type returnType,
params Type[] parameterTypes)
{
return type.GetMethods().Where((m) =>
{
if (m.Name != name)
{
return false;
}
if (m.ReturnType != returnType)
{
return false;
}
var parameters = m.GetParameters();
if ((parameterTypes == null || parameterTypes.Length == 0))
{
return parameters.Length == 0;
}
if (parameters.Length != parameterTypes.Length)
{
return false;
}
for (int i = 0; i < parameterTypes.Length; i++)
{
if (parameters[i].ParameterType != parameterTypes[i])
{
return false;
}
}
return true;
}).Single();
}
}
Which you could use like this,
class GenericFacade<T> where T : new()
{
private readonly t = new T();
public void CallOnInternal(string name, params object[] parameters)
{
var paramTypes = parameters.Select(p => typeof(p))
var match = typeof(T).GetMethodByNameThenSig(
name,
typeof(void),
paramTypes);
match.Invoke(this.t, parameters);
}
public TResult CallOnInternal<TResult>(string name, params object[] parameters)
{
var paramTypes = parameters.Select(p => typeof(p))
var match = typeof(T).GetMethodByNameThenSig(
name,
typeof(TResult),
paramTypes);
return (TResult)match.Invoke(this.t, parameters);
}
}
FINAL EDIT
Looking at the code involved to use reflection and considering the cost associated with the loss of type safety. I'd suggest its better to establish the "has-a" relationship explicitly in the traditional manner.
You can use public void myMethod(params object[] something) as in:
public static void Main()
{
UnknownArgumentsMethod1(1, 2, 3, "foo");
}
public static void UnknownArgumentsMethod1(params object[] list)
{
UnknownArgumentsMethod2(list);
}
public static void UnknownArgumentsMethod2(params object[] list)
{
foreach (object o in list)
{
if (o.GetType() == typeof(int))
{
Console.WriteLine("This is an integer: " + (int)o);
}
else if (o.GetType() == typeof(string))
{
Console.WriteLine("This is a string: " + (string)o);
}
}
}
The obvious answer would be to have inheritance.
In your case (even though the names of the classes suggest otherwise) the way to do it is by inheriting the ChildClass in the BaseClass and that way you would have the methods from the ChildClass exposed through the BaseClass.
ex:
class MyBaseClass: MyChildClass
{
}
If the classes are not related and you just want to have an instance of MyChildClass in MyBaseClass but only expose a certain set of methods but by not making the others private what you could do is expose the MyChildClass instance through an interface that only exposes the necessary fields like so:
public class BaseClass
{
public IChildClass ChildClassInstance = new ChildClass();
}
public class ChildClass : IChildClass
{
public void myMethod(int i)
{ /* do something with i */ }
public void myMethod(int a, string b)
{ /* get i from a and b and call: */ myMethod(i); }
public void myMethod(string c, string d)
{ /* get i from c and d and call: */ myMethod(i); }
}
public interface IChildClass
{
void myMethod(int i);
void myMethod(int a, string b);
}
and then you could access only the methods that you allow to be exposed through an instance of the base class:
BaseClass test = new BaseClass();
test.ChildClassInstance.myMethod(1);
test.ChildClassInstance.myMethod(1,"test");

Double Dispatch [duplicate]

I have heard/read the term but don't quite understand what it means.
When should I use this technique and how would I use it? Can anyone provide a good code sample?
The visitor pattern is a way of doing double-dispatch in an object-oriented way.
It's useful for when you want to choose which method to use for a given argument based on its type at runtime rather than compile time.
Double dispatch is a special case of multiple dispatch.
When you call a virtual method on an object, that's considered single-dispatch because which actual method is called depends on the type of the single object.
For double dispatch, both the object's type and the method sole argument's type is taken into account. This is like method overload resolution, except that the argument type is determined at runtime in double-dispatch instead of statically at compile-time.
In multiple-dispatch, a method can have multiple arguments passed to it and which implementation is used depends on each argument's type. The order that the types are evaluated depends on the language. In LISP, it checks each type from first to last.
Languages with multiple dispatch make use of generic functions, which are just function delcarations and aren't like generic methods, which use type parameters.
To do double-dispatch in C#, you can declare a method with a sole object argument and then specific methods with specific types:
using System.Linq;
class DoubleDispatch
{
public T Foo<T>(object arg)
{
var method = from m in GetType().GetMethods()
where m.Name == "Foo"
&& m.GetParameters().Length==1
&& arg.GetType().IsAssignableFrom
(m.GetParameters()[0].GetType())
&& m.ReturnType == typeof(T)
select m;
return (T) method.Single().Invoke(this,new object[]{arg});
}
public int Foo(int arg) { /* ... */ }
static void Test()
{
object x = 5;
Foo<int>(x); //should call Foo(int) via Foo<T>(object).
}
}
The code posted by Mark isn't complete and what ever is there isn't working.
So tweaked and complete.
class DoubleDispatch
{
public T Foo<T>(object arg)
{
var method = from m in GetType().GetMethods(System.Reflection.BindingFlags.Instance | System.Reflection.BindingFlags.Public | System.Reflection.BindingFlags.NonPublic)
where m.Name == "Foo"
&& m.GetParameters().Length == 1
//&& arg.GetType().IsAssignableFrom
// (m.GetParameters()[0].GetType())
&&Type.GetType(m.GetParameters()[0].ParameterType.FullName).IsAssignableFrom(arg.GetType())
&& m.ReturnType == typeof(T)
select m;
return (T)method.Single().Invoke(this, new object[] { arg });
}
public int Foo(int arg)
{
return 10;
}
public string Foo(string arg)
{
return 5.ToString();
}
public static void Main(string[] args)
{
object x = 5;
DoubleDispatch dispatch = new DoubleDispatch();
Console.WriteLine(dispatch.Foo<int>(x));
Console.WriteLine(dispatch.Foo<string>(x.ToString()));
Console.ReadLine();
}
}
Thanks Mark and others for nice explanation on Double Dispatcher pattern.
C# 4 introduces the pseudo type dynamic which resolves the function call at runtime (instead of compile time). (That is, the runtime type of the expression is used). Double- (or multi-dispatch) can be simplified to:
class C { }
static void Foo(C x) => Console.WriteLine(nameof(Foo));
static void Foo(object x) => Console.WriteLine(nameof(Object));
public static void Main(string[] args)
{
object x = new C();
Foo((dynamic)x); // prints: "Foo"
Foo(x); // prints: "Object"
}
Note also by using dynamic you prevent the static analyzer of the compiler to examine this part of the code. You should therefore carefully consider the use of dynamic.
The other answers use generics and the runtime type system. But to be clear the use of generics and runtime type system doesn't have anything to do with double dispatch. They can be used to implement it but double dispatch is just dependent on using the concrete type at runtime to dispatch calls. It's illustrated more clearly I think in the wikipedia page. I'll include the translated C++ code below. The key to this is the virtual CollideWith on SpaceShip and that it's overridden on ApolloSpacecraft. This is where the "double" dispatch takes place and the correct asteroid method is called for the given spaceship type.
class SpaceShip
{
public virtual void CollideWith(Asteroid asteroid)
{
asteroid.CollideWith(this);
}
}
class ApolloSpacecraft : SpaceShip
{
public override void CollideWith(Asteroid asteroid)
{
asteroid.CollideWith(this);
}
}
class Asteroid
{
public virtual void CollideWith(SpaceShip target)
{
Console.WriteLine("Asteroid hit a SpaceShip");
}
public virtual void CollideWith(ApolloSpacecraft target)
{
Console.WriteLine("Asteroid hit ApolloSpacecraft");
}
}
class ExplodingAsteroid : Asteroid
{
public override void CollideWith(SpaceShip target)
{
Console.WriteLine("ExplodingAsteroid hit a SpaceShip");
}
public override void CollideWith(ApolloSpacecraft target)
{
Console.WriteLine("ExplodingAsteroid hit ApolloSpacecraft");
}
}
class Program
{
static void Main(string[] args)
{
Asteroid[] asteroids = new Asteroid[] { new Asteroid(), new ExplodingAsteroid() };
ApolloSpacecraft spacecraft = new ApolloSpacecraft();
spacecraft.CollideWith(asteroids[0]);
spacecraft.CollideWith(asteroids[1]);
SpaceShip spaceShip = new SpaceShip();
spaceShip.CollideWith(asteroids[0]);
spaceShip.CollideWith(asteroids[1]);
}
}
Full listing of working code
using System;
using System.Linq;
namespace TestConsoleApp
{
internal class Program
{
public static void Main(string[] args)
{
const int x = 5;
var dispatch = new DoubleDispatch();
Console.WriteLine(dispatch.Foo<int>(x));
Console.WriteLine(dispatch.Foo<string>(x.ToString()));
Console.ReadLine();
}
}
public class DoubleDispatch
{
public T Foo<T>(T arg)
{
var method = GetType()
.GetMethods()
.Single(m =>
m.Name == "Foo" &&
m.GetParameters().Length == 1 &&
arg.GetType().IsAssignableFrom(m.GetParameters()[0].ParameterType) &&
m.ReturnType == typeof(T));
return (T) method.Invoke(this, new object[] {arg});
}
public int Foo(int arg)
{
return arg;
}
public string Foo(string arg)
{
return arg;
}
}
}

Simulate variadic templates in C#

Is there a well-known way for simulating the variadic template feature in C#?
For instance, I'd like to write a method that takes a lambda with an arbitrary set of parameters. Here is in pseudo code what I'd like to have:
void MyMethod<T1,T2,...,TReturn>(Fun<T1,T2, ..., TReturn> f)
{
}
C# generics are not the same as C++ templates. C++ templates are expanded compiletime and can be used recursively with variadic template arguments. The C++ template expansion is actually Turing Complete, so there is no theoretically limit to what can be done in templates.
C# generics are compiled directly, with an empty "placeholder" for the type that will be used at runtime.
To accept a lambda taking any number of arguments you would either have to generate a lot of overloads (through a code generator) or accept a LambdaExpression.
There is no varadic support for generic type arguments (on either methods or types). You will have to add lots of overloads.
varadic support is only available for arrays, via params, i.e.
void Foo(string key, params int[] values) {...}
Improtantly - how would you even refer to those various T* to write a generic method? Perhaps your best option is to take a Type[] or similar (depending on the context).
I know this is an old question, but if all you want to do is something simple like print those types out, you can do this very easily without Tuple or anything extra using 'dynamic':
private static void PrintTypes(params dynamic[] args)
{
foreach (var arg in args)
{
Console.WriteLine(arg.GetType());
}
}
static void Main(string[] args)
{
PrintTypes(1,1.0,"hello");
Console.ReadKey();
}
Will print "System.Int32" , "System.Double", "System.String"
If you want to perform some action on these things, as far as I know you have two choices. One is to trust the programmer that these types can do a compatible action, for example if you wanted to make a method to Sum any number of parameters. You could write a method like the following saying how you want to receive the result and the only prerequisite I guess would be that the + operation works between these types:
private static void AddToFirst<T>(ref T first, params dynamic[] args)
{
foreach (var arg in args)
{
first += arg;
}
}
static void Main(string[] args)
{
int x = 0;
AddToFirst(ref x,1,1.5,2.0,3.5,2);
Console.WriteLine(x);
double y = 0;
AddToFirst(ref y, 1, 1.5, 2.0, 3.5, 2);
Console.WriteLine(y);
Console.ReadKey();
}
With this, the output for the first line would be "9" because adding to an int, and the second line would be "10" because the .5s didn't get rounded, adding as a double. The problem with this code is if you pass some incompatible type in the list, it will have an error because the types can't get added together, and you won't see that error at compile time, only at runtime.
So, depending on your use case there might be another option which is why I said there were two choices at first. Assuming you know the choices for the possible types, you could make an interface or abstract class and make all of those types implement the interface. For example, the following. Sorry this is a bit crazy. And it can probably be simplfied.
public interface Applyable<T>
{
void Apply(T input);
T GetValue();
}
public abstract class Convertable<T>
{
public dynamic value { get; set; }
public Convertable(dynamic value)
{
this.value = value;
}
public abstract T GetConvertedValue();
}
public class IntableInt : Convertable<int>, Applyable<int>
{
public IntableInt(int value) : base(value) {}
public override int GetConvertedValue()
{
return value;
}
public void Apply(int input)
{
value += input;
}
public int GetValue()
{
return value;
}
}
public class IntableDouble : Convertable<int>
{
public IntableDouble(double value) : base(value) {}
public override int GetConvertedValue()
{
return (int) value;
}
}
public class IntableString : Convertable<int>
{
public IntableString(string value) : base(value) {}
public override int GetConvertedValue()
{
// If it can't be parsed return zero
int result;
return int.TryParse(value, out result) ? result : 0;
}
}
private static void ApplyToFirst<TResult>(ref Applyable<TResult> first, params Convertable<TResult>[] args)
{
foreach (var arg in args)
{
first.Apply(arg.GetConvertedValue());
}
}
static void Main(string[] args)
{
Applyable<int> result = new IntableInt(0);
IntableInt myInt = new IntableInt(1);
IntableDouble myDouble1 = new IntableDouble(1.5);
IntableDouble myDouble2 = new IntableDouble(2.0);
IntableDouble myDouble3 = new IntableDouble(3.5);
IntableString myString = new IntableString("2");
ApplyToFirst(ref result, myInt, myDouble1, myDouble2, myDouble3, myString);
Console.WriteLine(result.GetValue());
Console.ReadKey();
}
Will output "9" the same as the original Int code, except the only values you can actually pass in as parameters are things that you actually have defined and you know will work and not cause any errors. Of course, you would have to make new classes i.e. DoubleableInt , DoubleableString, etc.. in order to re-create the 2nd result of 10. But this is just an example, so you wouldn't even be trying to add things at all depending on what code you are writing and you would just start out with the implementation that served you the best.
Hopefully someone can improve on what I wrote here or use it to see how this can be done in C#.
Another alternative besides those mentioned above is to use Tuple<,> and reflection, for example:
class PrintVariadic<T>
{
public T Value { get; set; }
public void Print()
{
InnerPrint(Value);
}
static void InnerPrint<Tn>(Tn t)
{
var type = t.GetType();
if (type.IsGenericType && type.GetGenericTypeDefinition() == typeof(Tuple<,>))
{
var i1 = type.GetProperty("Item1").GetValue(t, new object[]{});
var i2 = type.GetProperty("Item2").GetValue(t, new object[]{ });
InnerPrint(i1);
InnerPrint(i2);
return;
}
Console.WriteLine(t.GetType());
}
}
class Program
{
static void Main(string[] args)
{
var v = new PrintVariadic<Tuple<
int, Tuple<
string, Tuple<
double,
long>>>>();
v.Value = Tuple.Create(
1, Tuple.Create(
"s", Tuple.Create(
4.0,
4L)));
v.Print();
Console.ReadKey();
}
}
I don't necessarily know if there's a name for this pattern, but I arrived at the following formulation for a recursive generic interface that allows an unlimited amount of values to be passed in, with the returned type retaining type information for all passed values.
public interface ITraversalRoot<TRoot>
{
ITraversalSpecification<TRoot> Specify();
}
public interface ITraverser<TRoot, TCurrent>: ITraversalRoot<TRoot>
{
IDerivedTraverser<TRoot, TInclude, TCurrent, ITraverser<TRoot, TCurrent>> AndInclude<TInclude>(Expression<Func<TCurrent, TInclude>> path);
}
public interface IDerivedTraverser<TRoot, TDerived, TParent, out TParentTraverser> : ITraverser<TRoot, TParent>
{
IDerivedTraverser<TRoot, TInclude, TDerived, IDerivedTraverser<TRoot, TDerived, TParent, TParentTraverser>> FromWhichInclude<TInclude>(Expression<Func<TDerived, TInclude>> path);
TParentTraverser ThenBackToParent();
}
There's no casting or "cheating" of the type system involved here: you can keep stacking on more values and the inferred return type keeps storing more and more information. Here is what the usage looks like:
var spec = Traversal
.StartFrom<VirtualMachine>() // ITraverser<VirtualMachine, VirtualMachine>
.AndInclude(vm => vm.EnvironmentBrowser) // IDerivedTraverser<VirtualMachine, EnvironmentBrowser, VirtualMachine, ITraverser<VirtualMachine, VirtualMachine>>
.AndInclude(vm => vm.Datastore) // IDerivedTraverser<VirtualMachine, Datastore, VirtualMachine, ITraverser<VirtualMachine, VirtualMachine>>
.FromWhichInclude(ds => ds.Browser) // IDerivedTraverser<VirtualMachine, HostDatastoreBrowser, Datastore, IDerivedTraverser<VirtualMachine, Datastore, VirtualMachine, ITraverser<VirtualMachine, VirtualMachine>>>
.FromWhichInclude(br => br.Mountpoints) // IDerivedTraverser<VirtualMachine, Mountpoint, HostDatastoreBrowser, IDerivedTraverser<VirtualMachine, HostDatastoreBrowser, Datastore, IDerivedTraverser<VirtualMachine, Datastore, VirtualMachine, ITraverser<VirtualMachine, VirtualMachine>>>>
.Specify(); // ITraversalSpecification<VirtualMachine>
As you can see the type signature becomes basically unreadable near after a few chained calls, but this is fine so long as type inference works and suggests the right type to the user.
In my example I am dealing with Funcs arguments, but you could presumably adapt this code to deal with arguments of arbitrary type.
For a simulation you can say:
void MyMethod<TSource, TResult>(Func<TSource, TResult> f) where TSource : Tparams {
where Tparams to be a variadic arguments implementation class. However, the framework does not provide an out-of-box stuff to do that, Action, Func, Tuple, etc., are all have limited length of their signatures. The only thing I can think of is to apply the CRTP .. in a way I've not find somebody blogged. Here's my implementation:
*: Thank #SLaks for mentioning Tuple<T1, ..., T7, TRest> also works in a recursive way. I noticed it's recursive on the constructor and the factory method instead of its class definition; and do a runtime type checking of the last argument of type TRest is required to be a ITupleInternal; and this works a bit differently.
Code
using System;
namespace VariadicGenerics {
public interface INode {
INode Next {
get;
}
}
public interface INode<R>:INode {
R Value {
get; set;
}
}
public abstract class Tparams {
public static C<TValue> V<TValue>(TValue x) {
return new T<TValue>(x);
}
}
public class T<P>:C<P> {
public T(P x) : base(x) {
}
}
public abstract class C<R>:Tparams, INode<R> {
public class T<P>:C<T<P>>, INode<P> {
public T(C<R> node, P x) {
if(node is R) {
Next=(R)(node as object);
}
else {
Next=(node as INode<R>).Value;
}
Value=x;
}
public T() {
if(Extensions.TypeIs(typeof(R), typeof(C<>.T<>))) {
Next=(R)Activator.CreateInstance(typeof(R));
}
}
public R Next {
private set;
get;
}
public P Value {
get; set;
}
INode INode.Next {
get {
return this.Next as INode;
}
}
}
public new T<TValue> V<TValue>(TValue x) {
return new T<TValue>(this, x);
}
public int GetLength() {
return m_expandedArguments.Length;
}
public C(R x) {
(this as INode<R>).Value=x;
}
C() {
}
static C() {
m_expandedArguments=Extensions.GetExpandedGenericArguments(typeof(R));
}
// demonstration of non-recursive traversal
public INode this[int index] {
get {
var count = m_expandedArguments.Length;
for(INode node = this; null!=node; node=node.Next) {
if(--count==index) {
return node;
}
}
throw new ArgumentOutOfRangeException("index");
}
}
R INode<R>.Value {
get; set;
}
INode INode.Next {
get {
return null;
}
}
static readonly Type[] m_expandedArguments;
}
}
Note the type parameter for the inherited class C<> in the declaration of
public class T<P>:C<T<P>>, INode<P> {
is T<P>, and the class T<P> is nested so that you can do some crazy things such as:
Test
[Microsoft.VisualStudio.TestTools.UnitTesting.TestClass]
public class TestClass {
void MyMethod<TSource, TResult>(Func<TSource, TResult> f) where TSource : Tparams {
T<byte>.T<char>.T<uint>.T<long>.
T<byte>.T<char>.T<long>.T<uint>.
T<byte>.T<long>.T<char>.T<uint>.
T<long>.T<byte>.T<char>.T<uint>.
T<long>.T<byte>.T<uint>.T<char>.
T<byte>.T<long>.T<uint>.T<char>.
T<byte>.T<uint>.T<long>.T<char>.
T<byte>.T<uint>.T<char>.T<long>.
T<uint>.T<byte>.T<char>.T<long>.
T<uint>.T<byte>.T<long>.T<char>.
T<uint>.T<long>.T<byte>.T<char>.
T<long>.T<uint>.T<byte>.T<char>.
T<long>.T<uint>.T<char>.T<byte>.
T<uint>.T<long>.T<char>.T<byte>.
T<uint>.T<char>.T<long>.T<byte>.
T<uint>.T<char>.T<byte>.T<long>.
T<char>.T<uint>.T<byte>.T<long>.
T<char>.T<uint>.T<long>.T<byte>.
T<char>.T<long>.T<uint>.T<byte>.
T<long>.T<char>.T<uint>.T<byte>.
T<long>.T<char>.T<byte>.T<uint>.
T<char>.T<long>.T<byte>.T<uint>.
T<char>.T<byte>.T<long>.T<uint>.
T<char>.T<byte>.T<uint>.T<long>
crazy = Tparams
// trying to change any value to not match the
// declaring type makes the compilation fail
.V((byte)1).V('2').V(4u).V(8L)
.V((byte)1).V('2').V(8L).V(4u)
.V((byte)1).V(8L).V('2').V(4u)
.V(8L).V((byte)1).V('2').V(4u)
.V(8L).V((byte)1).V(4u).V('2')
.V((byte)1).V(8L).V(4u).V('2')
.V((byte)1).V(4u).V(8L).V('2')
.V((byte)1).V(4u).V('2').V(8L)
.V(4u).V((byte)1).V('2').V(8L)
.V(4u).V((byte)1).V(8L).V('2')
.V(4u).V(8L).V((byte)1).V('2')
.V(8L).V(4u).V((byte)1).V('2')
.V(8L).V(4u).V('9').V((byte)1)
.V(4u).V(8L).V('2').V((byte)1)
.V(4u).V('2').V(8L).V((byte)1)
.V(4u).V('2').V((byte)1).V(8L)
.V('2').V(4u).V((byte)1).V(8L)
.V('2').V(4u).V(8L).V((byte)1)
.V('2').V(8L).V(4u).V((byte)1)
.V(8L).V('2').V(4u).V((byte)1)
.V(8L).V('2').V((byte)1).V(4u)
.V('2').V(8L).V((byte)1).V(4u)
.V('2').V((byte)1).V(8L).V(4u)
.V('7').V((byte)1).V(4u).V(8L);
var args = crazy as TSource;
if(null!=args) {
f(args);
}
}
[TestMethod]
public void TestMethod() {
Func<
T<byte>.T<char>.T<uint>.T<long>.
T<byte>.T<char>.T<long>.T<uint>.
T<byte>.T<long>.T<char>.T<uint>.
T<long>.T<byte>.T<char>.T<uint>.
T<long>.T<byte>.T<uint>.T<char>.
T<byte>.T<long>.T<uint>.T<char>.
T<byte>.T<uint>.T<long>.T<char>.
T<byte>.T<uint>.T<char>.T<long>.
T<uint>.T<byte>.T<char>.T<long>.
T<uint>.T<byte>.T<long>.T<char>.
T<uint>.T<long>.T<byte>.T<char>.
T<long>.T<uint>.T<byte>.T<char>.
T<long>.T<uint>.T<char>.T<byte>.
T<uint>.T<long>.T<char>.T<byte>.
T<uint>.T<char>.T<long>.T<byte>.
T<uint>.T<char>.T<byte>.T<long>.
T<char>.T<uint>.T<byte>.T<long>.
T<char>.T<uint>.T<long>.T<byte>.
T<char>.T<long>.T<uint>.T<byte>.
T<long>.T<char>.T<uint>.T<byte>.
T<long>.T<char>.T<byte>.T<uint>.
T<char>.T<long>.T<byte>.T<uint>.
T<char>.T<byte>.T<long>.T<uint>.
T<char>.T<byte>.T<uint>.T<long>, String>
f = args => {
Debug.WriteLine(String.Format("Length={0}", args.GetLength()));
// print fourth value from the last
Debug.WriteLine(String.Format("value={0}", args.Next.Next.Next.Value));
args.Next.Next.Next.Value='x';
Debug.WriteLine(String.Format("value={0}", args.Next.Next.Next.Value));
return "test";
};
MyMethod(f);
}
}
Another thing to note is we have two classes named T, the non-nested T:
public class T<P>:C<P> {
is just for the consistency of usage, and I made class C abstract to not directly being newed.
The Code part above needs to expand ther generic argument to calculate about their length, here are two extension methods it used:
Code(extensions)
using System.Diagnostics;
using System;
namespace VariadicGenerics {
[DebuggerStepThrough]
public static class Extensions {
public static readonly Type VariadicType = typeof(C<>.T<>);
public static bool TypeIs(this Type x, Type d) {
if(null==d) {
return false;
}
for(var c = x; null!=c; c=c.BaseType) {
var a = c.GetInterfaces();
for(var i = a.Length; i-->=0;) {
var t = i<0 ? c : a[i];
if(t==d||t.IsGenericType&&t.GetGenericTypeDefinition()==d) {
return true;
}
}
}
return false;
}
public static Type[] GetExpandedGenericArguments(this Type t) {
var expanded = new Type[] { };
for(var skip = 1; t.TypeIs(VariadicType) ? true : skip-->0;) {
var args = skip>0 ? t.GetGenericArguments() : new[] { t };
if(args.Length>0) {
var length = args.Length-skip;
var temp = new Type[length+expanded.Length];
Array.Copy(args, skip, temp, 0, length);
Array.Copy(expanded, 0, temp, length, expanded.Length);
expanded=temp;
t=args[0];
}
}
return expanded;
}
}
}
For this implementation, I choosed not to break the compile-time type checking, so we do not have a constructor or a factory with the signature like params object[] to provide values; instead, use a fluent pattern of method V for mass object instantiation to keep type can be statically type checked as much as possible.

Categories