Suppose my code possesses the knowledge about the metadata of a
nonexistent class library "mytest.dll", such as the types in this library, the functions of the types, the parameters and return types of the functions, etc.
How does my code manufacture this DLL using techniques such as reflection?
I know my code can generate the "mytest.cs" text file, then execute the compiler to produce the DLL, then delete the "mytest.cs" file. Just want to know if there are "more advanced" or "cooler" ways to do it.
Thanks.
There are 4 main steps in the process to compile and execute dynamic .net scripts from your application, even really complex scenarios can be simplified in this way:
Generate the code
Compile the script
Load the assembly
Execute the code
Lets generate a simple Hello Generated C# World App right now!:
Create a method that will generate an assembly that has 1 class called HelloWorldApp, this class has 1 method called GenerateMessage it will have X input parameters that will be integers, it will return a CSV string of the arguments that were passed in to it.
This solution requires the following package to be installed:
PM> Install-Package 'Microsoft.CodeAnalysis.CSharp.Scripting'
And will require the following using statements:
using Microsoft.CodeAnalysis;
using Microsoft.CodeAnalysis.CSharp;
using Microsoft.CodeAnalysis.Emit;
using System;
using System.Collections.Generic;
using System.IO;
using System.Linq;
using System.Reflection;
using System.Text;
Orchestration
The following method encapsulates the above steps:
private static void GenerateAndExecuteApp(int numberOfParameters)
{
string nameSpace = "Dynamic.Example";
string className = "HelloWorldApp";
string methodName = "GenerateMessage";
// 1. Generate the code
string script = BuildScript(nameSpace, className, methodName, numberOfParameters);
// 2. Compile the script
// 3. Load the Assembly
Assembly dynamicAssembly = CompileScript(script);
// 4. Execute the code
int[] arguments = Enumerable.Range(1, numberOfParameters).ToArray();
string message = ExecuteScript(dynamicAssembly, nameSpace, className, methodName, arguments);
Console.Out.WriteLine(message);
}
Generate the code
You say you already have item 1 sorted out, you can use StringBuilder, T4 templates or other mechanisms to generate the code files.
generating the code itself is its own question if you need help with that.
However, for our demo app, the following would work:
private static string BuildScript(string nameSpace, string className, string methodName, int numberOfParameters)
{
StringBuilder code = new StringBuilder();
code.AppendLine("using System;");
code.AppendLine("using System.Linq;");
code.AppendLine();
code.AppendLine($"namespace {nameSpace}");
code.AppendLine("{");
code.AppendLine($" public class {className}");
code.AppendLine(" {");
var parameterNames = Enumerable.Range(0, numberOfParameters).Select(x => $"p{x}").ToList();
code.Append($" public string {methodName}(");
code.Append(String.Join(",", parameterNames.Select(x => $"int {x}")));
code.AppendLine(")");
code.AppendLine(" {");
code.Append(" return $\"");
code.Append(String.Join(",", parameterNames.Select(x => $"{x}={{{x}}}")));
code.AppendLine("\";");
code.AppendLine(" }");
code.AppendLine(" }");
code.AppendLine("}");
return code.ToString();
}
For an input value of 3, the following code is generated:
using System;
using System.Linq;
namespace Dynamic.Example
{
public class HelloWorldApp
{
public string GenerateMessage(int p0,int p1,int p2)
{
return $"p0={p0},p1={p1},p2={p2}";
}
}
}
Compile the script (and Load it)
These are two discrete steps, however it is easiest to code them together in the same method, for this example we will ignore the generated dll and load the assembly directly into memory, that is generally the more likely use case for this type of scripting scenario anyway.
The hardest element of this is usually the referencing of the relevant dlls. There are a number of ways to achieve this, including loading all the dlls that are in the current executing context, I find a simple way to do this is to access the Assembly reference from the Type reference for the types we want to use inside the dynamic script:
List<string> dlls = new List<string> {
typeof(object).Assembly.Location,
typeof(Enumerable).Assembly.Location
};
Cut a long story short, this method compiles and loads the assembly into memory. It includes some crude compilation error handling, just to demonstrate how to do it:
private static Assembly CompileScript(string script)
{
SyntaxTree syntaxTree = CSharpSyntaxTree.ParseText(script);
// use "mytest.dll" if you want, random works well enough
string assemblyName = System.IO.Path.GetRandomFileName();
List<string> dlls = new List<string> {
typeof(object).Assembly.Location,
typeof(Enumerable).Assembly.Location
};
MetadataReference[] references = dlls.Distinct().Select(x => MetadataReference.CreateFromFile(x)).ToArray();
CSharpCompilation compilation = CSharpCompilation.Create(
assemblyName,
syntaxTrees: new[] { syntaxTree },
references: references,
options: new CSharpCompilationOptions(OutputKind.DynamicallyLinkedLibrary));
// Now we actually compile the script, this includes some very crude error handling, just to show you can
using (var ms = new MemoryStream())
{
EmitResult result = compilation.Emit(ms);
if (!result.Success)
{
IEnumerable<Diagnostic> failures = result.Diagnostics.Where(diagnostic =>
diagnostic.IsWarningAsError ||
diagnostic.Severity == DiagnosticSeverity.Error);
List<string> errors = new List<string>();
foreach (Diagnostic diagnostic in failures)
{
//errors.AddDistinct(String.Format("{0} : {1}", diagnostic.Id, diagnostic.Location, diagnostic.GetMessage()));
errors.Add(diagnostic.ToString());
}
throw new ApplicationException("Compilation Errors: " + String.Join(Environment.NewLine, errors));
}
else
{
ms.Seek(0, SeekOrigin.Begin);
return Assembly.Load(ms.ToArray());
}
}
}
Execute the code
Finally, we can use reflection to instantiate an instance of the new app and then we can obtain a reference to the method and it. The name of the parameters is irrelevant, as long
we pass them through in the correct order:
for this demo the order is sort of irrelevant to, given they are all the same type ;)
private static string ExecuteScript(Assembly assembly, string nameSpace, string className, string methodName, int[] arguments)
{
var appType = assembly.GetType($"{nameSpace}.{className}");
object app = Activator.CreateInstance(appType);
MethodInfo method = appType.GetMethod(methodName);
object result = method.Invoke(app, arguments.Cast<object>().ToArray());
return result as string;
}
Output
The final output from all this for our method with 3 passed into it is:
p0=1,p1=2,p2=3
So that was super crude, you can bypass most of the indirect reflection aspects through the use of Interfaces. If your generated script inherits from types or interfaces that the calling code also has a strong reference to, then ExecuteScript in the above example might look like this:
private static string ExecuteScript(Assembly assembly, string nameSpace, string className)
{
var appType = assembly.GetType($"{nameSpace}.{className}");
object app = Activator.CreateInstance(appType);
if (app is KnownInterface known)
{
return known.GenerateMessage(1,2,3);
}
throw new NotSupportedException("Couldn't resolve known type");
}
The major benefit to using an interface or base class reference is that you can natively set properties or call other methods without having to reflect references to them all or to resort to using dynamic which would work, but becomes a bit harder to debug.
Of course the interface solution is hard to implement when we had a variable number of parameters, so that's not the best example, usually with dynamic scripts you would construct a known environment, say a known class and methods, but you might want to inject custom code into the body of the method.
It's a bit of fun in the end, but this simple example shows that C# can be used as a runtime scripting engine without too much trouble.
I have a problem with the protobuf-net and Android app built with IL2CPP.
Everything worked fine when I used MONO instead of IL2CPP for development. Now I need to use IL2CPP for x64 support. I didn't know System.Reflection.Emit is not supported with IL2CPP and protobuf-net is using it.
Is there a way to make the protobuf-net work with IL2CPP?
I got same problem on iOS. You have to compile ProtoModel before.
using Assembly = UnityEditor.Compilation.Assembly;
private static void BuildMyProtoModel()
{
RuntimeTypeModel typeModel = TypeModel.Create();
foreach (var t in GetTypes(CompilationPipeline.GetAssemblies(AssembliesType.Player)))
{
var contract = t.GetCustomAttributes(typeof(ProtoContractAttribute), false);
if (contract.Length > 0)
{
MetaType metaType = typeModel.Add(t, true);
// support ISerializationCallbackReceiver
if (typeof(ISerializationCallbackReceiver).IsAssignableFrom(t))
{
MethodInfo beforeSerializeMethod = t.GetMethod("OnBeforeSerialize");
MethodInfo afterDeserializeMethod = t.GetMethod("OnAfterDeserialize");
metaType.SetCallbacks(beforeSerializeMethod, null, null, afterDeserializeMethod);
}
//add unity types
typeModel.Add(typeof(Vector2), false).Add("x", "y");
typeModel.Add(typeof(Vector3), false).Add("x", "y", "z");
}
}
typeModel.Compile("MyProtoModel", "MyProtoModel.dll"); //build model
string protoSchema = typeModel.GetSchema(null);//content for .proto file, you can generate a proto file for a specific type by passing it instead of null
}
private static IEnumerable<Type> GetTypes(IEnumerable<Assembly> assemblies)
{
foreach (Assembly assembly in assemblies)
{
foreach (Type type in AppDomain.CurrentDomain.Load(assembly.name).GetTypes())
{
yield return type;
}
}
}
Copy MyProtoModel.dll from root to Plugin folder.
And use like this:
TypeModel typeModel = new MyProtoModel();
I create small project Protobuf-net & Unity:
https://github.com/koshelevpavel/UniBufExample
https://github.com/koshelevpavel/UniBuf
But it just experimental and it don't have any documents.
Small example MonoBehaviour:
https://gist.github.com/koshelevpavel/8e2d62053fc79b2bf9e2105d18b056bc
In the DotNetYaml sample code I'm looking at, there's a C# construct:
var deserializer = new Deserializer(namingConvention: new CamelCaseNamingConvention());
var order = deserializer.Deserialize<Order>(input);
What is the equivalent F# code? I've tried
let deserializer = new Deserializer(namingConvention=new CamelCaseNamingConvention())
deserializer.Deserialize<Meta>(input)
If you have a C# library that defines optional parameters, then you can use the syntax you are using in your question. To quickly show that's the case, I compiled the following C# code as a library:
using System;
namespace Demo {
public class MyClass {
public static void Foo(int first, string second = "foo", string third = "bar") { }
}
}
You can reference this and use it from F# as follows:
open Demo
MyClass.Foo(1, third="hi")
I tried to do this with YamlDotNet which, I guess, is the library that you were using, but I get an error that the Deserializer class does not have namingConvention as an argument, so my guess would be that you are probably using a different version of the library than you are thinking (or perhaps, my guess of what library you're using was wrong...).
I am trying to directly invoke IronPython's built-in modules from C#. It looks like I'm missing some important initialization, that I can't find anywhere in the code.
Here's what I do:
namespace py.consoleio
{
using IronPython.Runtime;
using Microsoft.Scripting.Hosting;
using Microsoft.Scripting.Hosting.Providers;
using Microsoft.Scripting.Runtime;
public static class consoleio
{
public static string name;
static void Main()
{
var setup = new ScriptRuntimeSetup();
setup.LanguageSetups.Add(
IronPython.Hosting.Python.CreateLanguageSetup(null));
var dlrRuntime = new ScriptRuntime(setup);
var scriptDomainManager = HostingHelpers.GetDomainManager(dlrRuntime);
var pythonContext = new PythonContext(scriptDomainManager, null);
var context = new CodeContext(new PythonDictionary(), new ModuleContext(new PythonDictionary(), DefaultContext.DefaultPythonContext));
name = IronPython.Modules.Builtin.input(context, "What is your name?\n");
IronPython.Modules.Builtin.print(context, "Hi, %s.", consoleio.name);
System.GC.KeepAlive(pythonContext);
}
}
}
That properly outputs "What is your name?", but then crashes trying to decode input: unknown encoding: cp437.
Now I've already found, that encodings are initialized in Src/StdLib/Lib/encodings/init.py
I can't find how it gets to loading this module in a normal IronPython run (e.g. a console host), so I can't reproduce it in C# program.
My goal here is to invoke IronPython functions without dynamic dispatch.
UPD. Now I also tried to do this:
var engine = Python.CreateEngine();
this.ScriptDomainManager = HostingHelpers.GetDomainManager(engine.Runtime);
to the same result
Figured that one out: encodings module is implemented in Python in IronPython (core modules are in C#). It always worked with IronPythonConsole project, because it implicitly adds IronPython source for standard libraries to Python path. I just had to explicitly specify path like this:
var options = new Dictionary<string, object> { ["SearchPaths"] = path };
var engine = Python.CreateEngine(options);
I have simple IScript interface. And I want enforce that all scripts implement it.
public interface IScript<T>
{
T Execute(object[] args);
}
I want to use Roslyn scripting API to achive this. Something like this is possible with CSScript (see Interface Alignment).
var code = #"
using System;
using My.Namespace.With.IScript;
public class Script : IScript<string>
{
public string Execute()
{
return ""Hello from script!"";
}
}
";
var script = CSharpScript.Create(code, ScriptOptions.Default); // + Load all assemblies and references
script.WithInterface(typeof(IScript<string>)); // I need something like this, to enforce interface
script.Compile();
string result = script.Execute(); // and then execute script
Console.WriteLine(result); // print "Hello from script!"
Type safety is a static thing enforced a compile time (of your application). Creating and running a CSharpScript is done at runtime. So you cannot enforce type safety at runtime.
Maybe CSharpScript is not the right way to go. By using this SO answer,
You can compile a piece of C# code into memory and generate assembly bytes with Roslyn.
You would then change the line
object obj = Activator.CreateInstance(type);
to
IScript<string> obj = Activator.CreateInstance(type) as IScript<string>;
if (obj != null) {
obj.Execute(args);
}