Ensure early loading of (some) referenced assemblies

Question

I have a solution containing several projects.

There is one project that references all other projects, there may be references between these referenced projects as well.

Let's assume project A as the main one. Project B has a MyType class. Project C has an extension method for MyType.

Each assembly has an attribute. It allows me to determine if the assembly is mine or an external library (maybe it's not the best way to do it, but it works).

At some point in A I have a mechanism that uses reflection to find all extension methods of MyType. To do that I handle event AppDomain.AssemblyLoad to create a list of my loaded assemblies (marked with an attribute). Then I can scan the assemblies for proper extension methods.

Unfortunately, at this moment C is not loaded, because nothing from C has been called. Thus, the extension is not found.

I've tried using Assembly.GetExecutingAssembly().GetReferencedAssemblies(), but I think it returns only loaded assemblies, no luck here.

My final solution in A is:

public static void Main(string[] args)
{
    TypeFromProjectC temp;
    //further code...
}

It works because the assembly gets loaded on temp declaration, but...

I could try to load assemblies by path/name, but let's assume it's not an option. I want to create a generic solution that would not need to be altered each time I add another reference. In fact that is why I have added this attribute for my assemblies.

Something like GetReferencedAssemblies() would be great. It would be the best if I could check the assembly for my attribute before calling Load(), but it's not that necessary. (Does it even makes sense? When I think of it now, I'd guess that it must be loaded before the attributes are available, but maybe some meta information could do?)

It's the first time I'm playing with multiple assemblies with quite a complicated structure, so it's possible that I don't really have a clue how does it work and my question is plainly stupid.

I hope it's not ;]

Answer 1

You are looking for your ext methods so the referenced assemblies can be found int the following locations:

• the sub directories of the executing asm - simple to implement
• sub directories of called asms - simple to implement using the assembly loaded event
• mainfest - you have to figure this one out but I dont think it is that hard. Just parse the xml and read the elements
• GAC - if you register your asms in the GAC. The solution can be found here:Extract assembly from GAC

Answer 2

Some notes and comments before my answer:

1) You have a reference to ClassLibrary1 and ClassLibrary2, and instantiate Class1 from ClassLibrary1. In that case compiler throws away reference to ClassLibrary2 and Assembly.GetReferencedAssemblies() returns to you actual optimized list of references. When you use any type from ClassLibrary2, complier will respect that reference also. The best way for making strong refs is:

static class StrongReferencesLoader
{
    public static void Load()
    {
        var ref1 = typeof(Class1); // ClassLibrary1.dll
        var ref2 = typeof(Class2); // ClassLibrary2.dll
    }
}

This allows to reference static classes also and this allows to force loading of every strong-referenced library by calling this method.

2) You do use an AppDomain.AssemblyLoad event. Consider subscribing here from your entry point. Be careful, because if your entry point make something else this can lead to loading assembly before subscribing occurs.

class Program
{
    static void Main()
    {
        AppDomain.CurrentDomain.AssemblyLoad += CurrentDomain_AssemblyLoad;
        new Class1();
    }

    static void CurrentDomain_AssemblyLoad(object sender, AssemblyLoadEventArgs args)
    {
        Console.WriteLine("! " + args.LoadedAssembly.FullName);
    }
}

in this sample, you will not get an output about loading ClassLibrary1, because loading of library was performed before executing main method. Runtime loads all necessary libraries for executing specified method. If we extract this call to another method - everything become fine.

class Program
{
    static void Main()
    {
        AppDomain.CurrentDomain.AssemblyLoad += CurrentDomain_AssemblyLoad;
        RealMain();
    }

    static void RealMain()
    {
        new Class1();
    }

    static void CurrentDomain_AssemblyLoad(object sender, AssemblyLoadEventArgs args)
    {
        Console.WriteLine("! " + args.LoadedAssembly.FullName);
    }
}

In that case, Main is executed without problem, and subscribes immediately. After that, assembly loading performed due to call RealMain that contain a byte-code referenced to ClassLibrary1, and your event handler works.

3) Note that in any time you can list all loaded assemblies by AppDomain.CurrentDomain.GetAssemblies()

I'm not fully understand your problem. Is that only solution-wide analysis, or generic idea for future plugins, created by any one else? Solution-wide analysis can be solved in compile-time or run-time, but open plugin system is only runtime.

Run time solution

First of all, you should understand that it is impossible to detect every plugins. Let say, Class1 have extensions, written by some body for another application and installed in another directory:

some pseudo-structure:

\GAC\ClassLibrary1.dll
\ProgramFiles\MyApp\ConsoleApplication1.exe // our entry point
\ProgramFiles\SomeApp\ClassLibrary3.dll // some library from another solution, that references shared ClassLibrary1

You can detect plugins in your GAC, AppDomainBase path, RelativeSearchPath (all of that done by built-in assembly resolver), any custom hardcoded path, and paths of already loaded assemblies AppDomain.CurrentDomain.GetAssemblies().Select(x => Path.GetDirectoryName(x.Location)).Distinct(); And you can analyse that paths recursively. But you'l never find path like \ProgramFiles\SomeApp\ClassLibrary3.dll unless you scan entire computer.

I hope, that searching in deep to already loaded assemblies location can solve your problem.

Note that you can use Mono.Cecil, but it is convenient to use Assembly.ReflectionOnlyLoad for that purpose. You can load by file name using Assembly.ReflectionOnlyLoadFrom, and you can help resolve assemblies in another location by AppDomain.CurrentDomain.ReflectionOnlyAssemblyResolve event. But you can not use that for custom attributes analysis. In that case some hack is required, like special class name (static class MySuperMark{}), which existance can be checked with easy. It is less hacky to use special interface that every plugin required to implement (but it is not so close to your case).

Compile time solution

Only for solution-wide analysis. Since you have noted, that

I want to create a generic solution that would not need to be altered each time I add another reference.

Probably you are looking for solution-wide answer. I can suggest some metaprogramming here. Usually I'm solving such task by metacreator http://code.google.com/p/metacreator/. You can run some piece of code before compilation of ConsoleApplication1. For example, you can create a class holding strong references for every referenced dlls automatically by metacode and you can run such class for ensuring early loading of referenced assemblies. This class will actualized on every compilation automatically.

Please, describe more details, which case I can describe more deeply.

Answer 3

In my opinion, assembly analysis with Mono.Cecil is useful here. For example you can enumerate references (of course without ones deleted from project by compiler) without loading given assembly and any of its references by:

var assembly = Mono.Cecil.AssemblyDefinition.ReadAssembly("protobuf-net.dll");
foreach(var module in assembly.Modules)
{
    foreach(var reference in module.AssemblyReferences)
    {
    Console.WriteLine(reference.FullName);
    }
}

Such code prints:

mscorlib, Version=2.0.0.0, Culture=neutral, PublicKeyToken=b77a5c561934e089
System, Version=2.0.0.0, Culture=neutral, PublicKeyToken=b77a5c561934e089
System.Xml, Version=2.0.0.0, Culture=neutral, PublicKeyToken=b77a5c561934e089
System.ServiceModel, Version=3.0.0.0, Culture=neutral, PublicKeyToken=b77a5c561934e089
System.Runtime.Serialization, Version=3.0.0.0, Culture=neutral, PublicKeyToken=b77a5c561934e089

However IMO better idea is to scan through directory, possibly containing plugins and check them for your attribute using Cecil and therefore without loading (because you do not know at that point whether you want to load them at all):

var assembly = Mono.Cecil.AssemblyDefinition.ReadAssembly("protobuf-net.dll");
foreach(var attribute in assembly.CustomAttributes)
{
    Console.WriteLine(attribute.AttributeType.FullName);
}

There are couple options for comparing your attribute (which is a class from loaded assembly) to Cecil's attribute description, full type name is one of them.

Answer 4

Assembly.GetReferencedAssemblies() gives you a very reliable list of assemblies that your code actually uses. You will have to recurse through the list so you will find all of them, not just the ones that you main .exe assembly uses.

What you'll get does not necessarily match the assemblies you added with Project + Add Reference. The compiler makes sure to only include referenced assemblies in the metadata that your code actually uses. Which is perhaps why TypeFromProjectC temp; works.

If there are other assemblies lying around that ought to be loaded but are not found by recursing through GetReferencedAssemblies() then you have to load them yourself explicitly with Assembly.Load/From(). Actually using types from such assemblies has to be done manually too, using Activator.CreateInstance() to create an object and Reflection to call methods etcetera.

There's no evidence in your question that you actually intend to use Reflection, especially because you mention extension methods. An extension method that actually gets used in your code will always end up producing an assembly that's discovered through ReferencedAssemblies(). It is also fairly unclear why you want to do this in the first place. The just-in-time compiler is already very good at loading assemblies on demand, doing so when it compiles code just in time. Which is almost always desirable since it turns a long startup delay you'd get by forcing the CLR to find all assemblies into a hundred needle pricks that the user won't notice. Lazy assembly loading is very much a feature, not a bug.

Answer 5

If you don't have a static, compile-time reference to an assembly, directly or indirectly, such an assembly is not going to get loaded and thus is invisible to your enumeration logic. It is possible, however, to dynamically load all assemblies in, say, a particular directory at run-time.

On the other hand, I don't believe there is a place for a system-global dictionary of assemblies that might have a particular custom attribute, which seems to be what you're asking for.