I have written a program in vc++14 that runs perfectly on my pc, both as a standalone executable, and debugging in visual studio 2015.
However, when I attempt to run it on a different pc, it stops responding immedietly. Using debug symbols, I have found any instantiation of classes via make_unique, or make_shared fails, causing null pointer exceptions.
I have checked and made sure that the target machine has the necessary vc++ runtime environments, and included any other dlls I can think of.
I have also tried statically linking the necessary libraries during builds, with no improvement.
I have even tried an installer project, which detected no additional dependancies.
I'm now at a loss. Is there anything else I should be looking for?
Verify following.
Ensure you’re building a release build. It’s relatively hard to make debug builds run on a PC that has no visual studio installed.
Ensure you have the exact version of vc++14 runtime, including the service pack number. If you’re using vs2015 update 3, download and install this on the target PC: https://www.microsoft.com/en-us/download/details.aspx?id=53587 (the correct platform, 32 or 64).
Ensure you don’t have anything funny in your preprocessor: that you don’t have _DEBUG, no _SECURE_SCL, etc.
P.S. A dynamic link to C runtime is not necessary a good idea for a release build. One problem, on some Windows versions, you can’t update CRT to the version you need even using an offline installer, because recent CRT depends on some Windows Updates.
Related
We're currently beta testing a Windows application which is built with the latest Visual Studio in C++ and runs on Windows 10. The application dynamically links the VC++ libraries (static linkage is not an option for us).
On 75% of our testers' machines (including all our dev machines), the application works out of the box after installing, but with some others it does not start and fails (presumably) during the process of loading dynamic system libraries (since it does not trigger any kind of exception that would write a minidump as with runtime errors).
Some of these users have had errors about missing runtime dlls which were solved after installing the latest VC++ 2017 redistributable, however the application still would not run.
One user has also checked the library dependencies with the Dependencies tool (https://github.com/lucasg/Dependencies), but his results show nothing strange - there is no obvious difference between the output on a working machine and on his own. There are a few question marks (see screenshot: missing modules as shown in Dependencies) next to some UCRT subdependencies but they are there on working machines as well so I presume they are false positives.
I've also tried to deploy the relevant 40 something UCRT and VC++ dlls as an app local deploy next to the executable but it still wouldn't open on the affected machines (I might have missed some relevant ones, or they were still referenced from the System32 folder)
How would you debug such a problem, providing we cannot reproduce it locally (it works out of the box on two completely new devices with a fresh Windows 10 install and without a build environment) and there is a very little information on what might be going wrong with the library calls?
c000001d is illegal instruction exception code.
Either you are targeting instruction sets like AVX2 or SSE4.1 which the customer CPU doesn't support, or the executable is corrupted (e.g. downloaded in text mode instead of binary mode).
For best possible portability do not specify /arch:AVX or /arch:AVX2 when compiling with VC++. The compiler will then target the base instruction set available on the given architecture (x86 or x86_64 with SSE2).
I've run into an issue in VS2013 this morning, its one of those things that I'm sure to encounter in 1-3 years (either myself or a customer), and I'm going to say "Gee, I've seen that before, but I don't know what causes it".
We build for Windows Embedded platforms and are migrating customers from VS6.0 to VS2013. Since the target hardware is still WindowsEmbedded we use the v120_xp platform toolset. I've built ?hundreds? of projects and never encountered the error I saw this morning. A customer project with lots of DLL's, I successfully rebuilt all those with v120_xp, and turned on some OPTIMIZATION for performance. Not sure this is relevant. SSE2/Qpar and optimized for FAST execution. All the DLL's build fine. The customer application builds.
But at runtime I get this warning when the ntdll.dll is loaded:
ntdll.dll
%hs is either not designed to run on Windows or it contains an error.
Try installing the program again using the original installation media
or contact your system administrator or the software vendor for support.
Error status 0x.
From this StackOverFlowLink it appears the runtime is choosing the X64 version of ntdll.dll and possibly finding the x32 bit version. I'm pretty sure its something like that, but I'm trying to understand.
1) What could trigger this behavior? I haven't seen this before with other v120_xp projects in VS2013.
2) How can I tell which ntdll.dll it is seeing?
I tried dependancy walker x64 and x32 bit versions, and they seem to default different behavior.
I also played with setting compatibility mode on the exe to force Windows7 to run it in WinXP SP3 mode. That didn't work. We usually run our EXE's inside a GUI which might be forcing that...
I'm researching this today and will post my findings if I ever get to the bottom of what triggered this.
Two things that are unique about this project.
1) I turned lots of Performance CPU optimizations on in the C++ module.
2) There are a lot of customer DLL modules (also built optimized).
So one working theory I have is that the optimization or a DLL is tricking the runtime to thinking this is a X64 bit module.
I will be checking to see if a errent project doesn't have toolset v120_xp something like that.
I’m developing some Windows software that's sometimes used in embedded-like scenarios. That’s not uncommon for my users to have a dedicated Win7 or Win8 PC they never update, not even connect to the Internet. Users plug those PC to a specific industrial hardware, and use that system for one job only.
My software includes components I've written in Visual C++. I include CRT in my MSI packages using appropriate installer merge modules.
Unfortunately, for a PC that never updated, that doesn’t work: today I got a bug report, the app crashes on startup saying “can’t start because api-ms-win-crt-stdio-l1-1-0.dll is missing from your computer”
This answer suggests static link to CRT.
Did that. But some parts of my C++ code rely on OpenMP for parallelism.
Dependency walker shows me the DLL I’m building depends on vcomp140.dll even when compiled with Multi-threaded (/MT) runtime library settings.
Also it shows vcomp140.dll only depends on kernel32.dll and user32.dll.
Can I just place that single DLL, vcomp140.dll, in the installation folder? Will it work on offline Windows 7 PC?
Based on the VS2015 Redistribution List I would say that copying that file would be indeed what you need to do and would work fine (if you used VS2015 to build your app). Take care to copy the proper dll based on arm/x86/x64.
Just to make it clear, it is not possible to statically link openmp with Visual Studio. Only two things you can do:
Remove openmp (and compile with /MT /MTd)
deploy vcomp140.dll (or VC redistributable) with your application
Here's my configuration:
Computer A - Windows 7, MS Visual Studio 2005 patched for Win7 compatibility (8.0.50727.867)
Computer B - Windows XP SP2, MS Visual Studio 2005 installed (8.0.50727.42)
My project has some external dependencies (prebuilt DLLs - either build on A or downloaded from the Internet), a couple of DLLs built from sources and one executable. I am mostly developing on A and all is fine there. At some point I try to build my project on computer B, copying the prebuilt DLLs to the output folder. Everything builds fine, but trying to start my application I get
The application failed to initialize properly (0xc0150002)....
The event log contains two of those:
Dependent Assembly Microsoft.VC80.CRT could not be found and Last Error was The referenced assembly is not installed on your system.
plus the slightly more amusing
Generate Activation Context failed for
some.dll. Reference error message: The
operation completed successfully.
At this point I'm trying my Google-Fu, but in vain - virtually all hits are about running binaries on machines without Visual Studio installed. In my case, however, the executables fail to run on the computer they are built.
Next step was to try dependency walker and it baffled me even more - my DLLs built from sources on the same box cannot find MSVCR80.DLL and MSVCP80.DLL, however the executable seems to be alright in respect to those two DLLs i.e. when I open the executable with dependency walker it shows that the MSVC?80.DLLs can be found, but when I open one of my DLLs it says they cannot. That's where I am completely out of ideas what to do so I'm asking you, dear stackoverflow :)
I admit I'm a bit blurry on the whole side-by-side thing, so general reading on the topic will also be appreciated.
Your question has the answer to your problem: Computer A has VC runtime of version 8.0.50727.867, and Computer B has only version 8.0.50727.42.
You built your libraries on Computer A, and they depend on version 867 of VC runtime. (This can be found in manifest embedded in the libraries.) When you copy them to Computer B, these libraries still require version 867 of the runtime but you have only version 42.
To resolve the VC runtime assembly dependencies, you have to install VC runtime redistributables of version 867 on Computer B. However, I'd advise you to update Visual Studio on Computer B so that you have the same version on both computers. And even better, install Visual Studio 2005 SP1 on both computers and then install this security update to SP1. After installing the latter, your libraries will depend on version 8.0.50727.4053.
it's possible the problem is related with different versions of CRT runtime installed on both machines. is it possible to build all your modules to use statically linked CRT runtime to verify this?
first I'd check that prebuilt dlls by preparing dummy project to load them
I recently had the same type of error when building projects on one machine and then moving them to another machine. The biggest culprit here is likely a debug configuration for one of the binary components. That is, MSVC has the fairly rigid requirement of all DLLs/EXEs being built with the same runtime library, debug or release, otherwise they will not work together.
When I had this happen they also tend to compile just fine, but when you attempt to run them you get that extremely cryptic error message.
You need to ensure that every module you build together uses the same configuration, thus debug or release through the entire build chain. This error also likely comes up with mismatches in other libraries, so make sure your MSVC is the exact same version on the machines where you are building.
As you might be aware an update to visual studio 2005 was auto updated on most machines last week. This update included a new version of the visual c runtime library. As a result any binaries built after the update also require a new redistributable installed on client systems.
See http://support.microsoft.com/kb/971090/
And here is the installer for the new redistributable:
http://www.microsoft.com/downloads/details.aspx?displaylang=en&FamilyID=766a6af7-ec73-40ff-b072-9112bab119c2
This is fine for distributing new binaries to customers, I will ship the new redistributable with the installer and it will all work.
However I am really worried about my ability to hotfix existing customer sites if they discover a bug. In this case normally I would just send the dll or exe that was fixed.
However if I do this now, I will have to send these customers the new redistributable and now I will be using two different versions of the c runtime library in the same executable.
Is this a problem?
Can this cause my application to crash?
What happens if I allocate memory in one dll then deallocate it in another? Normally this works if the same release runtime library is used. I went through the our code about 3 years ago cleaning this up, but I cannot be sure that I have found and fixed all occurrences.
Is the allocate/deallocate in different dlls still a problem? Now that in the era of smart pointers etc it is very had to enforce this.
Can I control what runtime library version I depend on by changing manifests?
Any pointers or advice would be grateful.
Updated: I have just noticed this question
VC++: KB971090 and selecting Visual C Runtime DLL dependencies This is very similar, but my question is more concerned with using two different version of the runtime in one executable.
The version number specified in the application’s manifest file/resource only specifies the minimum version required to run the application. The default behavior of the loader is to first check the WINDOWS\WinSxS folder for the identical version or a superseding version of a dependency identified in an application manifest, and to use that version regardless of whether or not a private assembly containing the dependency has been provided with the application. (See http://msdn.microsoft.com/en-us/library/aa375674(VS.85).aspx).
So chances are your old binaries will also use the latest version of the Microsoft run time library anyway. Try to run the release build of your application (built before you update your Visual Studio) on a fully patched machine and use process explorer to see which DLLs it loads. The only problem is you will have to include the new run time redistributable file in your patch.
If you are still worried, you can try the method described here: http://tedwvc.wordpress.com/2009/08/10/avoiding-problems-with-vc2005-sp1-security-update-kb971090/
Yes you will not have to wait too much to see the problems using two runtime libraries.
If you allocate memory with a runtime and try to release it with another, your application will crash. It is and will continue being a problem.
Only the runtime what reserved the memory can track it. It is imposible for the other runtime to know how much memory you reserved.
You may want to read this, is a really good post about linking with msvcrt.dll.
I have heard (only by rumour) that if the manifest gives two versions of the CRT that only differ by minor revision number, then the application ends up only using the newer version at runtime. Ie, you do not run into problems with with multiple CRTs.
This is only rumour, and I'd love to hear a concrete answer on this.
See also: Visual Studio 2005 security updates and CRT DLL versions in manifest