I have an x64 MFC application (VS2019, fresh project from the wizard and filled with lots of old project code). Debug build is working fine but the project has a problem to start up in the Release build. Should be noted that it is a big project with several different subprojects and libraries included. (EDIT: pugi xml library seems to be causing the trouble, see below)
In AfxMainWin() function the pThread and pApp pointers are created correctly, and I can for example look at the command line arguments pThread->m_lpCmdLine. Then AfxMainWin calls pThread->InitInstance(), but inside the InitInstace function the this-pointer is corrupt, which leads to a crash when the code tries to access m_lpCmdLine or any other member-variable.
EDIT1: I found out that the this-pointer was okay on the first line in InitInstance(). Then a small init function was run that destroyed the this-pointer in InitInstance. This function reads settings from an xml file using Pugi Xml library, and specifically this line is the one that destroys my this pointer:
status = ((pugi::xml_document*)get_params())->load_file(file).status;
Probable reason: The pugixml.dll is incomaptible with my release build and causes memory corruption. I have used dependency walker on the pugixml.dll, and the result was not good. This DLL seems to be built in debug mode since it adds dependencies to the debug versions of MFC libraries, such as VCRUNTIME140D.DLL - and that is not good for a release build.
I do not have access to any release version of pugixml.dll, so my solution will be to throw out the pugixml.dll and include the .cpp in my project instead, or maybe use another xml reader instead. That should do the trick!
But I have a final question for you folks, so that I learn as much as possible from this: what is the most likely cause of my memory corruption? Is it the debug DLL that is incompatible with my release build? Or is it more likely that the DLL is old (v140 toolset) and my project is toolset v142? Or is it, as suggested by #Iinspectable , the actual function call (with c-style casting of the returned pointer) that is messed up and trashing my stack?
I note that the get_params() function returns a void* pointer (void* the_params;), but that void* pointer has earlier been created using the_params = new pugi::xml_document; - so at least it is being casted to the correct type.
EDIT2: after replacing the faulty pugixml.dll file with pugixml.cpp I could debug the issue, turns out to be some pugixml-related problem. I had to post that as another issue, since it is now a very different problem.
Related
I have been using a graphics library from Smaller Animals Software called ImgSource. Unfortunately, Smaller Animals Software has closed and is no longer available to answer questions. Recently, I had a system failure that deleted my only up-to-date copy of the library (I thought I had a backup but was wrong). I did, however, have the source code. I recompiled the library, both release and debug. (Both are static .lib files) I am also using MSVS 2019 Community edition and the project is an MFC project. The problem, and why I'm posting here, is that when I link the new release library with a previous project, the project compiles properly. However, when I build the debug version, it will compile, but not link and produces the linker error discussed LNK2038: mismatch detected for 'RuntimeLibrary': value 'MT_StaticRelease' doesn't match value 'MD_DynamicRelease' in file.obj
A reasonable conclusion is that there is something wrong with the debug library. However, if I build a new project and link the debug library, everything works fine. So, the error does not appear to be in the library. It seems I could start from scratch and completely redo these projects, but they do reflect a tremendous amount of work.
I can't figure out how to modify the settings in the existing projects so that they will compile in debug mode. I have tried the suggestions offered in the article referenced above. Further suggestions would be greatly appreciated.
dllmain.cpp(16): warning C4447: 'main' signature found without threading mode
l. Consider using 'int main(Platform::Array<Platform::String^>^ args)'.
Above is a warning I got from building a Windows Store App DLL project. I didn't change anything in that default dllmain.cpp file except for including my own version of pch.
The documentation for this warning (along with many VS2012 errors/warnings) is nowhere to be found on MSDN and here is the only relevant link I can find:
http://social.msdn.microsoft.com/Forums/en-US/vssetup/thread/6daa9587-fe54-4e84-a8b9-0e5c52c2f6e8/
and the op there didn't get an answer.
If anyone knows what it means and how to fix it, it would be great!
As far as I can tell, you can safely ignore the warning. The compiler cribs when it sees a Win32-style DllMain being compiled using the /ZW flag (Consume Windows Runtime Extensions). However, the function gets called as you'd normally expect.
Alternatively, you can work around the warning by compiling dllmain.cpp without /ZW. You might need to adjust the PCH settings for this to properly work. This is the path taken by the DLL (Windows Store apps) C++/CX project template in Visual Studio.
Incidentally, the reason you do not get the warning when you're trying to build a Windows Runtime Component project (which builds everything using /ZW) is that a Windows Runtime Component doesn't declare a DllMain. This is not to say that it can't; it just picks up the dummy DllMain that the CRT defines (which basically turns off per-thread initialization and reports success).
I'm working on a project that creates a DLL in C for Windows CE 5.0 using STANDARDSDK_500. The project is relatively simple with just one C source file. What I've noticed is that when I clean and build the project using Visual Studio 2005, it fails to create my Import lib. When I tell Visual Studio to rebuild it, it correctly creates the import lib.
Turning on verbose linking, I discover that in the failure case (clean and build) CVTRES.exe is discarding my functions, like so:
Invoking CVTRES.exe:
/machine:thumb
/verbose
/out:"C:\...\Temp\lnkC7E.tmp"
/readonly
/windowsce
".\standardsdk_500 (armv4i)\release\WLTBApi.res"
Microsoft (R) Windows Resource To Object Converter Version 8.00.50727.42
...
adding resource. type:VERSION, name:1, language:0x0409, flags:0x30, size:892
Discarded '.idata$4' from coredll.lib(COREDLL.dll)
...
Discarded WLRegisterStartupApp from WLTBApi.obj // my functions! Oh noes!
...
This discards all of my functions, which causes the linker to not create an import lib because there are no exported functions.
I figured that the input was the .res file mentioned in the commandline - WLTBApi.res. This file is identical in the intermediate directory after trying to compile both the working way (rebuild) and the non-working way (clean. build).
So, my first question is: what are the other inputs to the CVTRES.exe program so that I can find discrepencies between the two builds?
Second: what other troubleshooting techniques would you recommend?
Additional Info - I thought perhaps the .obj files were additional inputs - they differ between the builds. I now believe that they are outputs of CVTRES.exe. If anyone knows whether or not I'm correct in this, let me know.
Okay, so I found a solution, but I'm terribly dissatisfied with it. I discovered that the project I was building had a Dependency, but one that wasn't actually a dependency. In other words, the project, WLTBApi was building a DLL, and had a dependency configured, WLTBApiLib, but it wasn't actually using any of the output of that project. By removing the dependency, the problem went away.
I'd still love to know more about how to find an actual answer to what was going on, but maybe this answer will help someone else in the future.
I am trying to build an old version of an application which consists of VC++ projects that were written in Visual Studio 2003.
My OS is Windows 7 Enterprise (64-bit).
When I try and build the solution I get the following errors:
error C4772: #import referenced a type from a missing type library; '__missing_type__' used as a placeholder
fatal error C1084: Cannot read type library file: 'Smegui.tlb': Error loading type library/DLL.
They both complain about the following import statement:
#import "Smegui.tlb" no_implementation
This is not a case of the file path being incorrect as renaming the Smegui.tlb file causes the compiler to throw another error saying it cannot find the library.
Smegui is from another application that this one depends on. I thought perhaps I was missing a dll but there is no such thing as Smegui.dll.
All I know about .tlb files is that they are a type library and you can create them from an assembly using tlbexp.exe or regasm.exe (the later also registers the assembly with COM)
There is also an Apache Ant build script which uses a custom task to invoke devenv.com to build the projects. This is the same script that the build server originally used to build the application. It gives me the same errors when I try and run it.
The strangest thing about this is that I knew it ought to work seeing as it is all freshly checked out from subversion. I tried many different combinations of admin vs user elevation, VS vs Ant build, cleaning, release.
I have got it to build successfully about 5 times but the build seems to be non-deterministic.
If anyone can shed some light on how this tlb stuff even works or what this error might mean I would greatly appreciate it.
I found a far more reliable solution: open the tlb with oleview.exe and then close it.
Not sure what this actually does but it works every time.
I think oleview is actually one of the samples included with Visual Studio but I haven't had the time to debug it and see what it is doing.
I ran into this error because one type library was trying to load a dependent type library, which it could not find. Even though the dependent type library was in the same directory, and even though that directory was in the searchable path, the compiler would error loading the first type library, but not mention the dependent type library in the error.
To find the pseudo-missing type library, I ran Process Monitor (procman64.exe) during the compile. This showed that after the reported type library had successfully loaded, a dependent type library could not be found. It even showed all of the places that it was looking for the dependent type library, none of which were where it should have been looking (e.g.: ).
The fix was to add a <PreBuildEvent> to the project to copy the dependent .tlb file to one of the directories that was actually being searched.
<PreBuildEvent>
<Command>copy /Y ..\Lib\Interop\CWSpeechRecLib.tlb .\</Command>
</PreBuildEvent>
http://msdn.microsoft.com/en-us/library/sce74ah7%28VS.71%29.aspx
smegui.tlb is referencing some other tlb that the compiler can't find. If you have the .idl for smegui you might be able to figure out what the other is. I suspect the missing tlb is something that original build machine had registered but that your machine doesn't have registered.
A type library is a binary description of a set of interfaces, coclasses and enums. They're usually generated for COM components, in the case of tlbexp and regasm the tlb is created from the assembly metadata. For native COM components they are usually generated from an idl (Interface Description Language) file by the midl tool.
Edit:
I just noticed you're on x64 Windows. Are you building the project with a new version of Visual Studio? If so, are you targeting x86 or x64? If the latter, it may simply be a 32bit component that the compiler can't find (or less likely, a x64 component the x86 compiler can't find if you are targeting x86), for WOW64 the registry is virtualized for x86 vs. x64 applications.
Well I finally found out why I managed to get it to build sometimes and not others... sort of.
So long as I ran the build script with elevated administrator permissions and let that get as far as it could until that error occurred, then run the build script again as a protected administrator succeeded. Those steps must be done in that exact order with no other steps in between. If I try build in Visual Studio it does not work (although I did get it to succeed once). Probably some kind of virtualisation issue although it still doesn't quite make sense.
Well I don't need help on this any more and I know it's probably impossible to fully answer this question without knowing exactly what the build is doing. However if anyone does have any more thoughts I would happily receive them.
Cheers,
Steiny
I'm having a weird problem when trying to register the .dll i created using regsvr32.
During development everything went fine, the debug version registers and works fine. Now i wanted to create a Release version, but that Version does not register anymore.
regsvr32 comes up with the following error:
The module "mpegsplitter.dll" failed to load.
Make sure the binary is stored at the specified path or
debug it to check for problems with the binary or
dependent .DLL files.
The specified procedure could not be found.
Some research brought me to the dependency walker, which does tell me this
Error: At least one module has an unresolved import due to a missing export function in an implicitly dependent module.
It also does show a dependency on "crtdll.dll" that the debug version does not have (The function view shows soem functions that normally should be in ole32.dll), which is colored red'ish.
So far so good, i guess its somehow related to what the dependency walker shows there.
But where do i go from here? How do i fix it?
Any help would be greatly appreciated, that has been keeping me busy for several hours already.
Thanks!
I have the same problem. When I compared the different between "Command Line" (in Project Properties -> Linker) of Release and Debug mode, I found out that the "Optimization" options (in Project Properties -> Linker) of Release mode was turned on while ion Debug not.
Turning of Optimization for linker in Release mode solved the problem
Is it possible that the debug version is compiled with _ATL_MIN_CRT but the release version isn't? You can set this with the Minimize CRT Use in ATL project property as well.
I fixed it. It was actually being caused by the order of some mingw libraries i included to link against ffmpeg. Oh well, how weird.
In my case, the difference was in Module Definition File entry between DEBUG and RELEASE. The DEBUG version was pointing to the .DEF file where as the RELEASE had it empty.