After some help and reading, I am sure the concept activation context can be used to solve my problem. In my application, I have to load an unmanaged dll c.dll by another managed dll b.dll , which is loaded by a.dll , which is loaded by app.exe . The sequence is that app.exe -> a.dll (managed) -> b.dll (managed) -> c.dll (unmanaged) I have the source of a.dll and b.dll , but app.exe and c.dll belongs to third party. Now, regardless of whether a.exe loads other versions of c.dll or not, I have to load my own c.dll (my c.dll resides in a different folder with that carried by a.exe installation). how does the concept activation context can be migrated to the .net environment to solve this problem?

It should be possible but I think you will have to manually manage activation contexts. For example, in my proof of concept test a C# Winform app uses a class from a referenced C# Class Library. That class references a C++/CLI class library that loads a native DLL. The test code in the C++/CLI class library uses LoadLibrary to first load version 2.0 of the native DLL from one file system location. It specifies a path in the call to LoadLibrary. An exported function is called that indicates that version 2.0 of the DLL was used. Then the C++/CLI code manually creates and activates an activation context for its preferred DLL, version 1.0. LoadLibrary is called again without specifying a path. This time, the exported function from the preferred version 1.0 is used to resolve the export. The default activation context is then restored and LoadLibrary is called a third time without specifying a path and the exported function from version 2.0 is called again. And here you can see that both DLLs are loaded in the process -

How to use the concept of activation context in .net to load unmanaged dlls?

Accepted answer

RLWA32 40,861 Reputation points

2022-05-26T16:39:11.913+00:00

It should be possible but I think you will have to manually manage activation contexts.

For example, in my proof of concept test a C# Winform app uses a class from a referenced C# Class Library. That class references a C++/CLI class library that loads a native DLL. The test code in the C++/CLI class library uses LoadLibrary to first load version 2.0 of the native DLL from one file system location. It specifies a path in the call to LoadLibrary. An exported function is called that indicates that version 2.0 of the DLL was used. Then the C++/CLI code manually creates and activates an activation context for its preferred DLL, version 1.0. LoadLibrary is called again without specifying a path. This time, the exported function from the preferred version 1.0 is used to resolve the export. The default activation context is then restored and LoadLibrary is called a third time without specifying a path and the exported function from version 2.0 is called again.

And here you can see that both DLLs are loaded in the process -
Please sign in to rate this answer.

3 people found this answer helpful.
燕鹏李 301 Reputation points

2022-05-28T11:00:39.887+00:00

I will be back to your idea in the future. Now, I decide to use the out-of-process model to solve this problem.

RLWA32 40,861 Reputation points

2022-05-29T12:17:40.05+00:00

FYI -- I expanded the proof of concept test to include C++/CLI class libraries that were linked with the native DLLs import libraries. At runtime they called the native functions directly without using LoadLibrary/GetProcAddress. In this scenario the system handled activation contexts, both DLLs were loaded and the correct functions were called.

燕鹏李 301 Reputation points

2022-05-30T12:46:46.76+00:00

It is very helpful. Sometimes what works and what doesn't work is equally important.

燕鹏李 301 Reputation points

2022-06-14T14:25:23.95+00:00

I tried it out today, but I cannot load different version dlls successfully. What I have done is:
(some background info: application has loaded its own version a_v1.dll, which loaded its dependencies b_v1.dll, c_v1.dll. These dependencies is not static dependecies of a_v1.dll.
I have the source of my.dll (C#), which is a managed dll. it loads a_v2.dll using LoadLibraryEx. a_v2.dll will load its own dependencies b_v2.dll, c_v2.dll. All a_v1.dll, b_v1.dll, c_v1.dll, a_v2.dll, b_v2.dll, c_v2.dll, are unmanaged dll (written in c++))

embed my.dll.manifest into my.dll using mt.exe <?xml version='1.0' encoding='UTF-8' standalone='yes'?>
<assembly xmlns='urn:schemas-microsoft-com:asm.v1' manifestVersion='1.0'>
<file name="a_v2.dll" />
<file name="b_v2.dll" />
<file name="c_v2.dll" />
</assembly>

Embedding command: mt.exe -manifest my.dll.manifest -outputresource:my.dll;#2

燕鹏李 301 Reputation points

2022-06-14T14:41:39.743+00:00

Second. Before loading a_v2.dll inside my.dll, I created an activation context using my.dll, then activate this activation context, then LoadLibraryEx("a_v2.dll" ....), then deactivate this activation context

// The ACTCTX setting ACTCTX actCtx = new ACTCTX(); actCtx.cbSize = Marshal.SizeOf(typeof(ACTCTX)); actCtx.lpSource = manifestPath; // path of my.dll actCtx.wLangId = 1033; actCtx.lpAssemblyDirectory = assemblyDirectory; actCtx.hModule = IntPtr.Zero; actCtx.lpResourceName = (IntPtr)2; actCtx.dwFlags = ACTCTX_FLAG_RESOURCE_NAME_VALID;

Third. I put my.dll, a_v2.dll, b_v2.dll, c_v2.dll in the same directory.

Even so, I still cannot load a_v2.dll and its dependencies using the concept activation context from a managed dll. I read some theory on this topic, but I cannot get successed in practice :(

RLWA32 40,861 Reputation points

2022-06-14T15:36:36.537+00:00

This is the proof of concept sample that I previously used. https://1drv.ms/u/s!AmnqrCFBv4nDgjlafpVAPgnmSP-T?e=VNubjb

燕鹏李 301 Reputation points

2022-06-16T10:01:11.73+00:00

I really appreciate your code samples. After reading your code since yesterday, I took a closer look at how the activation context works. Now as far as my own problem is concerned, I can load my own DLLs separately (private dependencies). After debugging with Process Monitor, I found that, my add-in cannot use the resources loaded by the main application, or it would crashed. Is it also possible loaded resources privately (such as v8_context_snapshot.bin this resource, I want to load my own copy separately for my add-in, no matter whether the application loads this resource or not).

RLWA32 40,861 Reputation points

2022-06-16T10:28:42.217+00:00

Now as far as my own problem is concerned, I can load my own DLLs separately (private dependencies).

Glad you got it working.

After debugging with Process Monitor, I found that, my add-in cannot use the resources loaded by the main application, or it would crashed. Is it also possible loaded resources privately (such as v8_context_snapshot.bin this resource, I want to load my own copy separately for my add-in, no matter whether the application loads this resource or not).

This seems to me to be a different issue that should be the topic of a new question. I suggest that you put together a demo project that reproduces your issue which you can share with the community.

燕鹏李 301 Reputation points

2022-06-16T10:39:57.507+00:00

ok, sir. I'll sort out my problem and upload a demo project.

燕鹏李 301 Reputation points

2022-06-17T14:36:47.53+00:00

I have difficulty reproducing the error in a simple demo project, so I choose to download the pdb file and the source codes for cef library to debug this error. Finally, I found where the error is

Although I can load two different versions libcef.dll, these two DLLs have the same code to register window class and create a window, so they register the window class with the same name:

// static LPCTSTR CefBrowserPlatformDelegateNativeWin::GetWndClass() { return L"CefBrowserWindow"; }

This will arise the error 1410. There is no way to pass the window class name from the outer code since it is a string constant in its code. The only way to solve this problem is to modify and recompile the cef library.

Though I won't to modify and recompile the cef library (it is huge, my computer doesn't have enough resources to do this task), I really appreciate your help in my understanding the concept of activation context.

RLWA32 40,861 Reputation points

2022-06-17T15:00:44.657+00:00

Maybe you need to make sure that the activation context for your private version of the library is active before any window creation functions are called. That is what happens when ISOLATION_AWARE_ENABLED is used. This causes various functions to be wrapped with code that activates the context, invokes the Windows API function and then deactivates the context.

燕鹏李 301 Reputation points

2022-06-17T15:22:41.453+00:00

Wow, it makes sense. I can't quite give up on this idea yet and continue to fight this problem tomorrow based on your hint.

The activation context is not activated when I create the browser window (so the RegisterClassEx and CreateWindowEx).
Sign in to comment

Share via

How to use the concept of activation context in .net to load unmanaged dlls?

0 additional answers