Using Side-by-Side assemblies to load the x64 or x

Question

We have two versions of a managed C++ assembly, one for x86 and one for x64. This assembly is called by a .net application complied for AnyCPU. We are deploying our code via a file copy install, and would like to continue to do so.

Is it possible to use a Side-by-Side assembly manifest to loading a x86 or x64 assembly respectively when an application is dynamically selecting it's processor architecture? Or is there another way to get this done in a file copy deployment (e.g. not using the GAC)?

Answer 1

You can use the corflags utility to force an AnyCPU exe to load as an x86 or x64 executable, but that doesn't totally meet the file copy deployment requirement unless you choose which exe to copy based on the target.

Answer 2

Have a look at SetDllDirectory. I used it around the dynamically loading of an IBM spss assembly for both x64 and x86. It also solved paths for non assembly support dll's loaded by the assemblies in my case was the case with the spss dll's.

http://msdn.microsoft.com/en-us/library/ms686203%28VS.85%29.aspx

Answer 3

I created a simple solution that is able to load platform-specific assembly from an executable compiled as AnyCPU. The technique used can be summarized as follows:

1. Make sure default .NET assembly loading mechanism ("Fusion" engine) can't find either x86 or x64 version of the platform-specific assembly
2. Before the main application attempts loading the platform-specific assembly, install a custom assembly resolver in the current AppDomain
3. Now when the main application needs the platform-specific assembly, Fusion engine will give up (because of step 1) and call our custom resolver (because of step 2); in the custom resolver we determine current platform and use directory-based lookup to load appropriate DLL.

To demonstrate this technique, I am attaching a short, command-line based tutorial. I tested the resulting binaries on Windows XP x86 and then Vista SP1 x64 (by copying the binaries over, just like your deployment).

Note 1: "csc.exe" is a C-sharp compiler. This tutorial assumes it is in your path (my tests were using "C:\WINDOWS\Microsoft.NET\Framework\v3.5\csc.exe")

Note 2: I recommend you create a temporary folder for the tests and run command line (or powershell) whose current working directory is set to this location, e.g.

(cmd.exe)
C:
mkdir \TEMP\CrossPlatformTest
cd \TEMP\CrossPlatformTest

Step 1: The platform-specific assembly is represented by a simple C# class library:

// file 'library.cs' in C:\TEMP\CrossPlatformTest
namespace Cross.Platform.Library
{
    public static class Worker
    {
        public static void Run()
        {
            System.Console.WriteLine("Worker is running");
            System.Console.WriteLine("(Enter to continue)");
            System.Console.ReadLine();
        }
    }
}

Step 2: We compile platform-specific assemblies using simple command-line commands:

(cmd.exe from Note 2)
mkdir platform\x86
csc /out:platform\x86\library.dll /target:library /platform:x86 library.cs
mkdir platform\amd64
csc /out:platform\amd64\library.dll /target:library /platform:x64 library.cs

Step 3: Main program is split into two parts. "Bootstrapper" contains main entry point for the executable and it registers a custom assembly resolver in current appdomain:

// file 'bootstrapper.cs' in C:\TEMP\CrossPlatformTest
namespace Cross.Platform.Program
{
    public static class Bootstrapper
    {
        public static void Main()
        {
            System.AppDomain.CurrentDomain.AssemblyResolve += CustomResolve;
            App.Run();
        }

        private static System.Reflection.Assembly CustomResolve(
            object sender,
            System.ResolveEventArgs args)
        {
            if (args.Name.StartsWith("library"))
            {
                string fileName = System.IO.Path.GetFullPath(
                    "platform\\"
                    + System.Environment.GetEnvironmentVariable("PROCESSOR_ARCHITECTURE")
                    + "\\library.dll");
                System.Console.WriteLine(fileName);
                if (System.IO.File.Exists(fileName))
                {
                    return System.Reflection.Assembly.LoadFile(fileName);
                }
            }
            return null;
        }
    }
}

"Program" is the "real" implementation of the application (note that App.Run was invoked at the end of Bootstrapper.Main):

// file 'program.cs' in C:\TEMP\CrossPlatformTest
namespace Cross.Platform.Program
{
    public static class App
    {
        public static void Run()
        {
            Cross.Platform.Library.Worker.Run();
        }
    }
}

Step 4: Compile the main application on command line:

(cmd.exe from Note 2)
csc /reference:platform\x86\library.dll /out:program.exe program.cs bootstrapper.cs

Step 5: We're now finished. The structure of the directory we created should be as follows:

(C:\TEMP\CrossPlatformTest, root dir)
    platform (dir)
        amd64 (dir)
            library.dll
        x86 (dir)
            library.dll
    program.exe
    *.cs (source files)

If you now run program.exe on a 32bit platform, platform\x86\library.dll will be loaded; if you run program.exe on a 64bit platform, platform\amd64\library.dll will be loaded. Note that I added Console.ReadLine() at the end of the Worker.Run method so that you can use task manager/process explorer to investigate loaded DLLs, or you can use Visual Studio/Windows Debugger to attach to the process to see the call stack etc.

When program.exe is run, our custom assembly resolver is attached to current appdomain. As soon as .NET starts loading the Program class, it sees a dependency on 'library' assembly, so it tries loading it. However, no such assembly is found (because we've hidden it in platform/* subdirectories). Luckily, our custom resolver knows our trickery and based on the current platform it tries loading the assembly from appropriate platform/* subdirectory.

Answer 4

This solution can work for non managed assemblies as well. I have created a simple example similar to Milan Gardian's great example. The example I created dynamically loads a Managed C++ dll into a C# dll compiled for the Any CPU platform. The solution makes use of the InjectModuleInitializer nuget package to subscribe to the AssemblyResolve event before the dependencies of the assembly are loaded.

https://github.com/kevin-marshall/Managed.AnyCPU.git

Answer 5

My version, similar to @Milan, but with several important changes:

• Works for ALL DLLs that were not found
• Can be turned on and off

• AppDomain.CurrentDomain.SetupInformation.ApplicationBase is used instead of Path.GetFullPath() because the current directory might be different, e.g. in hosting scenarios, Excel might load your plugin but the current directory will not be set to your DLL.

• Environment.Is64BitProcess is used instead of PROCESSOR_ARCHITECTURE, as we should not depend on what the OS is, rather how this process was started - it could have been x86 process on a x64 OS. Before .NET 4, use IntPtr.Size == 8 instead.

Call this code in a static constructor of some main class that is loaded before all else.

public static class MultiplatformDllLoader
{
    private static bool _isEnabled;

    public static bool Enable
    {
        get { return _isEnabled; }
        set
        {
            lock (typeof (MultiplatformDllLoader))
            {
                if (_isEnabled != value)
                {
                    if (value)
                        AppDomain.CurrentDomain.AssemblyResolve += Resolver;
                    else
                        AppDomain.CurrentDomain.AssemblyResolve -= Resolver;
                    _isEnabled = value;
                }
            }
        }
    }

    /// Will attempt to load missing assembly from either x86 or x64 subdir
    private static Assembly Resolver(object sender, ResolveEventArgs args)
    {
        string assemblyName = args.Name.Split(new[] {','}, 2)[0] + ".dll";
        string archSpecificPath = Path.Combine(AppDomain.CurrentDomain.SetupInformation.ApplicationBase,
                                               Environment.Is64BitProcess ? "x64" : "x86",
                                               assemblyName);

        return File.Exists(archSpecificPath)
                   ? Assembly.LoadFile(archSpecificPath)
                   : null;
    }
}