Using reflection to dynamically query an assembly - c#

I am having difficulty with using reflections dynamically eg. query a .exe file without requiring a reference to be added for every assembly which I wish to query against.
So for instance, the code below is the regular way to get a hold of a class to then be checked.
AssemblyName assembly_name = new AssemblyName( "Name" );
The issue is not adding the argument in to the code but the code requirng direct reference to the new assembly to check against.
Any suggestions are welcome.

It sounds like you're really just trying to load an assembly at execution time. Look at Assembly.Load and Assembly.ReflectionOnlyLoad.

Maybe you're looking for something like Cecil. It's a library (available on Windows and other platforms) that allows to query metadata without the need to resolve all references.

I'm not really sure what you mean by "query". If you want to know how to create an instance from an assembly using reflection, here is an example:
// From within the current assembly
public CartesianType CreateInstance(string fullyQualifiedClassName)
{
Assembly assembly = Assembly.GetExecutingAssembly();
Type target = assembly.GetType(fullyQualifiedClassName, true, true);
return (CartesianType)Activator.CreateInstance(target);
}
// From an external assembly already referenced in your project
public SomeClass CreateInstance(string fullyQualifiedClassName)
{
Assembly assembly = Assembly.GetAssembly(typeof(SomeClass));
Type target = assembly.GetType(fullyQualifiedClassName, true, true);
return (SomeClass)Activator.CreateInstance(target);
}
All other methods must use Load or LoadFile, LoadFrom etc.

Related

How to unload assembly created by CSharpCodeProvider?

Problem
CSharpCodeProvider can be used to compile source .cs files into an assembly.
However, the assembly is automatically loaded into the AppDomain.CurrentDomain by default. In my case, this is a problem because I need to be able to re-compile the assembly again during runtime, and since it's already loaded in the CurrentDomain, I can't unload that, so I'm stuck.
I have looked through the docs and there seems to be no way to set the target app domain. I have also tried searching it on Google and only found answers where Assembly.Load was used, which I don't think I can use because I need to compile from raw source code, not a .dll
How would one go about doing this? Are there any alternatives or workarounds?
Main program
using (var provider = new CSharpCodeProvider())
{
param.OutputAssembly = "myCompiledMod"
var classFileNames = new DirectoryInfo("C:/sourceCode").GetFiles("*.cs", SearchOption.AllDirectories).Select(fi => fi.FullName).ToArray();
CompilerResults result = provider.CompileAssemblyFromFile(param, classFileNames);
Assembly newAssembly = result.CompiledAssembly // The assembly is already in AppDomain.CurrentDomain!
// If you try compile again, you'll get an error; that class Test already exists
}
C:/sourceCode/test.cs
public class Test {}
What I tried already
I already tried creating a new AppDomain and loading it in there. What happens is the assembly ends up being loaded in both domains.
// <snip>compile code</snip>
Evidence ev = AppDomain.CurrentDomain.Evidence;
AppDomain domain = AppDomain.CreateDomain("NewDomain", ev);
domain.Load(newAssembly);
The answer was to use CSharpCodeProvider().CreateCompiler() instead of just CSharpCodeProvider, and to set param.GenerateInMemory to false. Now I'm able to see line numbers and no visible assembly .dll files are being created, and especially not being locked. This allows for keeping an assembly in memory and reloading it when needed.

Activator.CreateInstance throws Exception "The system cannot find the file specified."

I've copied code for an assembly that is used in a solution to create a similar assembly. The GUTS was different, but the shell stayed the same.
These assemblies are used in a project at a client that are add-on's and not part of our core code. Now that I'm finished the assembly does not want to load like it is supposed to.
The code that loads the assembly is
var assemblyName = ((XmlElement)xmlDoc.GetElementsByTagName("AssemblyName")[0]).InnerText;
var qualifiedClass = ((XmlElement)xmlDoc.GetElementsByTagName("QualifiedClass")[0]).InnerText;
IExternalAddOn addOn = (IExternalAddOn)Activator.CreateInstance(assemblyName, qualifiedClass).Unwrap();
var properties = new Dictionary<Type, object>();
properties[typeof(DevExpress.XtraBars.Ribbon.RibbonControl)] = mainForm.ribbon;
var form = addOn.ShowForm(properties);
if (form != null)
{
form.MdiParent = mainForm;
form.Text = pListRow.NAME;
form.Show();
I get the exception on the CreateInstance part.
The interesting thing is that when I use
Assembly ass = Assembly.LoadFrom(assemblyName); // this is test code
Type at = ass.GetType(qualifiedClass);
IExternalAddOn addOn = (IExternalAddOn)Activator.CreateInstance(at);
to load the assembly and get the type, and CreateInstance it works.
Why do you need the Unwrap? Is there a difference in the two different ways of loading? And WHY does the first one not work?
Thanks
J
As soon as Assembly.LoadFrom works, I guess you are passing a file name as an assemblyName parameter, which is wrong in case of Activator.CreateInstance(assemblyName, qualifiedClass).
According to Activator.CreateInstance documentation:
assemblyName can be either of the following:
The simple name of an assembly, without its path or file extension. For example, you would specify TypeExtensions for an assembly whose path and name are .\bin\TypeExtensions.dll.
The full name of a signed assembly, which consists of its simple name, version, culture, and public key token; for example, "TypeExtensions, Version=1.0.0.0, Culture=neutral, PublicKeyToken=181869f2f7435b51".

How can I dynamically reference an assembly that looks for another assembly?

Apologies for the dodgy question - happy to rephrase if someone has a better suggestion.
I'm trying to create an object by dynamically invoking an assembly belonging to another application.
The following PowerShell code is working nicely for me:
[Reflection.Assembly]::LoadFrom("C:\Program Files\Vendor\Product\ProductAPI.dll")
$bobject = new-object ProductAPI.BasicObject
$bobject.AddName("Some Name")
I'm struggling to do the same thing in C#. Based on other posts on StackOverflow I currently have this:
System.Reflection.Assembly myDllAssembly =
System.Reflection.Assembly.LoadFile("C:\\Program Files\\Vendor\\Product\\ProductAPI.dll");
System.Type BasicObjectType = myDllAssembly.GetType("ProductAPI.BasicObject");
var basicObjectInstance = Activator.CreateInstance(BasicObjectType);
The final line results in a TargetInvocationException.
{"Could not load file or assembly 'AnotherObject, Version=1.2.345.0, Culture=neutral, PublicKeyToken=null' or one of its dependencies. The system cannot find the file specified."
It appears that the BasicObject constructor is trying to invoke AnotherObject (from AnotherObject.dll in the same folder) but can't find it.
Any tips on how to get around this?
If it can't find a dependent assembly in the usual places, you'll need to manually specify how to find them.
The two easiest ways I'm aware of for doing this:
manually load the dependent assemblies in advance with
Assembly.Load.
handle the AssemblyResolve event for the domain which is loading the
assembly with additional assembly dependencies.
Both essentially require you to know the dependencies for the assembly you're trying to load in advance but I don't think that's such a big ask.
If you go with the first option, it would also be worthwhile looking into the difference between a full Load and a reflection-only Load.
If you would rather go with 2 (which I'd recommend), you can try something like this which has the added benefit of working with nested dependency chains (eg MyLib.dll references LocalStorage.dll references Raven.Client.dll references NewtonSoft.Json.dll) and will additionally give you information about what dependencies it can't find:
AppDomain.CurrentDomain.AssemblyResolve += (sender,args) => {
// Change this to wherever the additional dependencies are located
var dllPath = #"C:\Program Files\Vendor\Product\lib";
var assemblyPath = Path.Combine(dllPath,args.Name.Split(',').First() + ".dll");
if(!File.Exists(assemblyPath))
throw new ReflectionTypeLoadException(new[] {args.GetType()},
new[] {new FileNotFoundException(assemblyPath) });
return Assembly.LoadFrom(assemblyPath);
};

Loading an Assembly if a certain Attribute is present

Ok, here is the deal:
I want to load a user defined Assembly into my AppDomain, but I only want to do so if the specified Assembly matches some requirements. In my case, it must have, among other requirements, an Assembly level Attribute we could call MandatoryAssemblyAttribute.
There are two paths I can go as far as I see:
Load the Assembly into my current AppDomain and check if the Attribute is present. Easy but inconvenient as I'm stuck with the loaded Assembly even if it doesnt have the MandatoryAssemblyAttribute. Not good.
I can create a new AppDomain and load the Assembly from there and check if my good old MandatoryAddemblyAttribute is present. If it is, dump the created AppDomain and go ahead and load the Assembly into my CurrentAppDomain, if not, just dump the new AppDomain, tell the user, and have him try again.
Easier said than done. Searching the web, I've found a couple of examples on how to go about this, including this previous question posted in SO:Loading DLLs into a separate AppDomain
The problem I see with this solution is that you actually have to know a type (full name) in the Assembly you want to load to begin with. That is not a solution I like. The main point is trying to plug in an arbitrary Assembly that matches some requirements and through attributes discover what types to use. There is no knowledge beforehand of what types the Assembly will have. Of course I could make it a requirement that any Assembly meant to be used this way should implement some dummy class in order to offer an "entry point" for CreateInstanceFromAndUnwrap. I'd rather not.
Also, if I go ahead and do something along this line:
using (var frm = new OpenFileDialog())
{
frm.DefaultExt = "dll";
frm.Title = "Select dll...";
frm.Filter = "Model files (*.dll)|*.dll";
answer = frm.ShowDialog(this);
if (answer == DialogResult.OK)
{
domain = AppDomain.CreateDomain("Model", new Evidence(AppDomain.CurrentDomain.Evidence));
try
{
domain.CreateInstanceFrom(frm.FileName, "DummyNamespace.DummyObject");
modelIsValid = true;
}
catch (TypeLoadException)
{
...
}
finally
{
if (domain != null)
AppDomain.Unload(domain);
}
}
}
This will work fine, but if then I go ahead and do the following:
foreach (var ass in domain.GetAssemblies()) //Do not fret, I would run this before unloading the AppDomain
Console.WriteLine(ass.FullName);
I get a FileNotFoundException. Why?
Another path I could take is this one: How load DLL in separate AppDomain But I'm not getting any luck either. I'm getting a FileNotFoundException whenever I choose some random .NET Assembly and besides it defeats the purpose as I need to know the Assembly's name (not file name) in order to load it up which doesn't match my requirements.
Is there another way to do this barring MEF (I am not targeting .NET 3.5)? Or am I stuck with creating some dummy object in order to load the Assembly through CreateInstanceFromAndUnwrap? And if so, why can't I iterate through the loaded assemblies without getting a FileNotFoundException? What am I doing wrong?
Many thanks for any advice.
The problem I see with this solution is that you actually have to know
a type (full name) in the Assembly
That is not quite accurate. What you do need to know is a type name is some assembly, not necessarily the assembly you are trying to examine. You should create a MarshalByRef class in your assembly and then use CreateInstanceAndUnwrap to create an instance of it from your own assembly (not the one you are trying to examine). That class would then do the load (since it lives in the new appdomain) and examine and return a boolean result to the original appdomain.
Here is some code to get you going. These classes go in your own assembly (not the one you are trying to examine):
This first class is used to create the examination AppDomain and to create an instance of your MarshalByRefObject class (see bottom):
using System;
using System.Security.Policy;
internal static class AttributeValidationUtility
{
internal static bool ValidateAssembly(string pathToAssembly)
{
AppDomain appDomain = null;
try
{
appDomain = AppDomain.CreateDomain("ExaminationAppDomain", new Evidence(AppDomain.CurrentDomain.Evidence));
AttributeValidationMbro attributeValidationMbro = appDomain.CreateInstanceAndUnwrap(
typeof(AttributeValidationMbro).Assembly.FullName,
typeof(AttributeValidationMbro).FullName) as AttributeValidationMbro;
return attributeValidationMbro.ValidateAssembly(pathToAssembly);
}
finally
{
if (appDomain != null)
{
AppDomain.Unload(appDomain);
}
}
}
}
This is the MarshalByRefObject that will actually live in the new AppDomain and will do the actual examination of the assembly:
using System;
using System.Reflection;
public class AttributeValidationMbro : MarshalByRefObject
{
public override object InitializeLifetimeService()
{
// infinite lifetime
return null;
}
public bool ValidateAssembly(string pathToAssembly)
{
Assembly assemblyToExamine = Assembly.LoadFrom(pathToAssembly);
bool hasAttribute = false;
// TODO: examine the assemblyToExamine to see if it has the attribute
return hasAttribute;
}
}
This can be easily done by using a managed assembly reader, such as Mono.Cecil. You'd check whether the assembly is decorated with an attribute, without loading the assembly in the AppDomain, and actually, without messing with AppDomains at all. For instance, with Cecil:
bool HasMandatoryAttribute (string fileName)
{
return AssemblyDefinition.ReadAssembly (fileName)
.CustomAttributes
.Any (attribute => attribute.AttributType.Name == "MandatoryAttribute");
}
That's basically what are doing most of he plugins systems, as creating an AppDomain and tearing it down is quite expensive for the runtime.

Assembly.GetType() and typeof() return different types?

Suppose you are given a Class.dll assembly compiled from the following simple code:
namespace ClassLibrary
{
public class Class
{
}
}
And consider a different project with the above Class.dll as a project reference and with the following code:
Assembly assembly = Assembly.LoadFrom(#"Class.dll");
Type reflectedType = assembly.GetType("ClassLibrary.Class");
Type knownType = typeof(ClassLibrary.Class);
Debug.Assert(reflectedType == knownType);
The assertion fails, and I don't understand why.
The assertion succeeds if I replace the ClassLibrary.Class with, say, the System.Text.RegularExpressions.Regex class and Class.dll with System.dll, so I'm guessing it has something to do with the project properties ? some Compilation switch perhaps?
Thanks in advance
The problem is load contexts: assemblies loaded via .LoadFrom are kept in a different "pile" than those loaded by Fusion (.Load). The types are actually different to the CLR. Check this link for more detail from a CLR architect.
You're probably loading two different copies of the same assembly.
Compare knownType.Assembly to reflectedType.Assembly in the debugger, and look at the paths and versions.
I would imagine that you are not referencing the same assembly that you are loading from disk.
This example (when compiled as Test.exe) works just fine:
using System;
using System.Reflection;
class Example
{
static void Main()
{
Assembly assembly = Assembly.LoadFrom(#"Test.exe");
Type reflectedType = assembly.GetType("Example");
Type knownType = typeof(Example);
Console.WriteLine(reflectedType == knownType);
}
}

Categories