Mismatch Detected for 'RuntimeLibrary'

Question

(This is already answered in comments, but since it lacks an actual answer, I'm writing this.)

This problem arises in newer versions of Visual C++ (the older versions usually just silently linked the program and it would crash and burn at run time.) It means that some of the libraries you are linking with your program (or even some of the source files inside your program itself) are using different versions of the CRT (the C RunTime library.)

To correct this error, you need to go into your Project Properties (and/or those of the libraries you are using,) then into C/C++, then Code Generation, and check the value of Runtime Library; this should be exactly the same for all the files and libraries you are linking together. (The rules are a little more relaxed for linking with DLLs, but I'm not going to go into the "why" and into more details here.)

There are currently four options for this setting:

1. Multithreaded Debug
2. Multithreaded Debug DLL
3. Multithreaded Release
4. Multithreaded Release DLL

Your particular problem seems to stem from you linking a library built with "Multithreaded Debug" (i.e. static multithreaded debug CRT) against a program that is being built using the "Multithreaded Debug DLL" setting (i.e. dynamic multithreaded debug CRT.) You should change this setting either in the library, or in your program. For now, I suggest changing this in your program.

Note that since Visual Studio projects use different sets of project settings for debug and release builds (and 32/64-bit builds) you should make sure the settings match in all of these project configurations.

For (some) more information, you can see these (linked from a comment above):

1. Linker Tools Warning LNK4098 on MSDN
2. /MD, /ML, /MT, /LD (Use Run-Time Library) on MSDN
3. Build errors with VC11 Beta - mixing MTd libs with MDd exes fail to link on Bugzilla@Mozilla

UPDATE: (This is in response to a comment that asks for the reason that this much care must be taken.)

If two pieces of code that we are linking together are themselves linking against and using the standard library, then the standard library must be the same for both of them, unless great care is taken about how our two code pieces interact and pass around data. Generally, I would say that for almost all situations just use the exact same version of the standard library runtime (regarding debug/release, threads, and obviously the version of Visual C++, among other things like iterator debugging, etc.)

The most important part of the problem is this: having the same idea about the size of objects on either side of a function call.

Consider for example that the above two pieces of code are called A and B. A is compiled against one version of the standard library, and B against another. In A's view, some random object that a standard function returns to it (e.g. a block of memory or an iterator or a FILE object or whatever) has some specific size and layout (remember that structure layout is determined and fixed at compile time in C/C++.) For any of several reasons, B's idea of the size/layout of the same objects is different (it can be because of additional debug information, natural evolution of data structures over time, etc.)

Now, if A calls the standard library and gets an object back, then passes that object to B, and B touches that object in any way, chances are that B will mess that object up (e.g. write the wrong field, or past the end of it, etc.)

The above isn't the only kind of problems that can happen. Internal global or static objects in the standard library can cause problems too. And there are more obscure classes of problems as well.

All this gets weirder in some aspects when using DLLs (dynamic runtime library) instead of libs (static runtime library.)

This situation can apply to any library used by two pieces of code that work together, but the standard library gets used by most (if not almost all) programs, and that increases the chances of clash.

What I've described is obviously a watered down and simplified version of the actual mess that awaits you if you mix library versions. I hope that it gives you an idea of why you shouldn't do it!

I had this problem along with mismatch in ITERATOR_DEBUG_LEVEL. As a sunday-evening problem after all seemed ok and good to go, I was put out for some time. Working in de VS2017 IDE (Solution Explorer) I had recently added/copied a sourcefile reference to my project (ctrl-drag) from another project. Looking into properties->C/C++/Preprocessor - at source file level, not project level - I noticed that in a Release configuration _DEBUG was specified instead of NDEBUG for this source file. Which was all the change needed to get rid of the problem.

I downloaded and extracted Crypto++ in C:\cryptopp. I used Visual Studio Express 2012 to build all the projects inside (as instructed in readme), and everything was built successfully. Then I made a test project in some other folder and added cryptolib as a dependency.

The conversion was probably not successful. The only thing that was successful was the running of VCUpgrade. The actual conversion itself failed but you don't know until you experience the errors you are seeing. For some of the details, see Visual Studio on the Crypto++ wiki.

---

Any ideas how to fix this?

To resolve your issues, you should download vs2010.zip if you want static C/C++ runtime linking (/MT or /MTd), or vs2010-dynamic.zip if you want dynamic C/C++ runtime linking (/MT or /MTd). Both fix the latent, silent failures produced by VCUpgrade.

---

vs2010.zip, vs2010-dynamic.zip and vs2005-dynamic.zip are built from the latest GitHub sources. As of this writing (JUN 1 2016), that's effectively pre-Crypto++ 5.6.4. If you are using the ZIP files with a down level Crypto++, like 5.6.2 or 5.6.3, then you will run into minor problems.

There are two minor problems I am aware. First is a rename of bench.cpp to bench1.cpp. Its error is either:

• C1083: Cannot open source file: 'bench1.cpp': No such file or directory
• LNK2001: unresolved external symbol "void __cdecl OutputResultOperations(char const *,char const *,bool,unsigned long,double)" (?OutputResultOperations@@YAXPBD0_NKN@Z)

The fix is to either (1) open cryptest.vcxproj in notepad, find bench1.cpp, and then rename it to bench.cpp. Or (2) rename bench.cpp to bench1.cpp on the filesystem. Please don't delete this file.

The second problem is a little trickier because its a moving target. Down level releases, like 5.6.2 or 5.6.3, are missing the latest classes available in GitHub. The missing class files include HKDF (5.6.3), RDRAND (5.6.3), RDSEED (5.6.3), ChaCha (5.6.4), BLAKE2 (5.6.4), Poly1305 (5.6.4), etc.

The fix is to remove the missing source files from the Visual Studio project files since they don't exist for the down level releases.

Another option is to add the missing class files from the latest sources, but there could be complications. For example, many of the sources subtly depend upon the latest config.h, cpu.h and cpu.cpp. The "subtlety" is you won't realize you are getting an under-performing class.

An example of under-performing class is BLAKE2. config.h adds compile time ARM-32 and ARM-64 detection. cpu.h and cpu.cpp adds runtime ARM instruction detection, which depends upon compile time detection. If you add BLAKE2 without the other files, then none of the detection occurs and you get a straight C/C++ implementation. You probably won't realize you are missing the NEON opportunity, which runs around 9 to 12 cycles-per-byte versus 40 cycles-per-byte or so for vanilla C/C++.

Issue can be solved by adding CRT of msvcrtd.lib in the linker library. Because cryptlib.lib used CRT version of debug.