Issue using startMATLAB in C++ with "MatlabEngine.hpp" in r2021b

Hello,

This is an identical issue to this and this, but essentially i have followed the Cmake instructions to run the Matlab API in c++, ie found the package, linked the engine libraries etc. It seems to build and link just fine but on running it will crash with an access violation on 0x0 address when Matlab calls 'cpp_engine_create_session()'. Again pretty much identical to similarly raised issues but i can't find an answer anywhere as to what causes this and how to fix it?

[chen dong]

I got exactly the same problem, using R2022a. Any one help?

[Joe Elsom]

Hi,

Not a solution, but some info that may help us all with this. using R2022a.

I've been facing the same issue. I found my code runs fine if the dependency libMatlabEngine.dll is NOT in my build folder (i.e the folder where the exe is run). However, if the exe is run on a different machine, it fails to start since libMatlabEngine.dll is not found. Adding the .dll on the 2nd machine results in the access violation.

So, it appears this issue happens when the exe is using libMatlabEngine.dll directly. I'm not actually sure what the code is using when libMatlabEngine.dll is NOT in the build folder.

[Joe Elsom]

More Info :

After some digging i found this post refering to the matlab C++ engine requiring a valid installation of Matlab on the machine. I experiemented with this and found a few things.

1) To run the C++ build on the build machine, make sure <matlabroot>\<version>\extern\bin\win64 is on the PATH. This can be set via the environment variables. When the above is on the path, do not include libMatlabEngine.dll or libMatlabDataArray.dll in the build folder. The call to matlab::engine::startMatlab() should work, since it will find and use the .dlls inside the local Matlab install via the environment variable. Note the code will not work on a 2nd machine unless it also has a valid Matlab install + the PATH envrionment variable set properly.

2) if libMatlabEngine.dll is included in the build folder (i.e via a post-build event command xcopy) on when the call to matlab::engine::startMatlab() is made, it will try to use the local libMatlabEngine.dll in the build folder. In my case, this causes the access violation exception. I assume, based on the linked post above this is because libMatlabEngine.dll is looking for other Matlab dependencies usually relative to it in the local Matlab install folder (the environment variable set in #1 above).

3) Setting an incorrect environment variable (i.e pointing to the .\extern\bin\win64 folder of a Matlab Runtime install) will raise the exception that libMatlabEngine.dll is not found. This is becasue the Runtime doesn't include it. I copied it to the folder and got a different exception. I assume becasue there are Matlab install dependencies missing when you're only using the Runtime.

So, at the moment at least it looks like it is not possible to run the C++ Engine API on a 2nd machine wihtout a valid install of Matlab (the Runtime doesn't suffice). Additionally, it is possible to run the C++ Engine API locally on the build machine by setting a PATH environement variable that points to the correct folder in a valid Matlab install.

Hopefully this is helpful. I'm still after a proper solution for distributing C++ builds that rely on Matlab though. I guess the Matlab compiler will be required.

Answer 1

[Scott Dattalo]

Thanks @Joe Elsom. I was have the same issue with using locally referenced libMatlab*.dll's. Moving those out of the way fixed my problem. If those DLL's are locally referenced, then a problem occurs when the C++ app invokes startMATLAB():

  // Start MATLAB engine synchronously
  std::unique_ptr<MATLABEngine> matlabPtr = startMATLAB();

I'm using MATLAB2019a and VS2019.

1) System environment variables. I had to add this to the beginning of my Windows PATH

C:\Program Files\MATLAB\R2019a\extern\bin\win64

2) The VS2019 Configuration Platform needs to be set to "x64". You can use either the "Release" or "Debug" configurations.

3) In Configuration Properties -> Debugging, I added.

C:\Program Files\MATLAB\R2019a\extern\bin\win64\;%PATH%

*note* - (1) and (2) seem the same. I should go back and see if both are needed.

4) Configuration Properties -> C/C++ -> General

Additional Include Directories set to C:\Program Files\MATLAB\R2019a\extern\include

5) Configuration Properties -> Linker -> General

Additional Library Directories set to C:\Program Files\MATLAB\R2019a\extern\lib\win64\microsoft

6) Configuration Properties -> Linker -> Input

Additional Dependencies prepend with -> libMatlabEngine.lib;libMatlabDataArray.lib;

I used the MATLAB Square Root Example:

#include "MatlabDataArray.hpp"
#include "MatlabEngine.hpp"
#include <iostream>
void callSQRT() {
  using namespace matlab::engine;

  // Start MATLAB engine synchronously
  std::unique_ptr<MATLABEngine> matlabPtr = startMATLAB();

  //Create MATLAB data array factory
  matlab::data::ArrayFactory factory;

  // Define a four-element typed array
  matlab::data::TypedArray<double> const argArray =
    factory.createArray({ 1,4 }, { -2.0, 2.0, 6.0, 8.0 });

  // Call MATLAB sqrt function on the data array
  matlab::data::Array const results = matlabPtr->feval(u"sqrt", argArray);

  // Display results
  for (int i = 0; i < results.getNumberOfElements(); i++) {
    double a = argArray[i];
    std::complex<double> v = results[i];
    double realPart = v.real();
    double imgPart = v.imag();
    std::cout << "Square root of " << a << " is " <<
      realPart << " + " << imgPart << "i" << std::endl;
  }
}

int main() {
  std::cout << "Starting\n";
  callSQRT();
  return 0;
}

Scott

[Joe Elsom]

@Scott Dattalo Nice, this was the same solution for me.

I've also found, anything implementing the Matlab Engine API for C++ cannot be distributed on the MATLAB Runtime. The Matlab Engine API for C++ must have a local, fully licensed installation of Matlab to function. This is because the MCR does not include the Engine API.

If you wish to write distributable programs in C++ that depend on Matlab implementations, you must first compile your Matlab code as a C++ Shared Library using the Matlab Compiler. You will then need to refactor the C++ code to deprecate the Engine API and implement only the Data API and MatlabCppSharedLib.hpp include. With this distributable method, Matlab is started with the calls below (copied from the linked documentation page above).

std::shared_ptr<mc::MATLABApplication> setup() { auto mode =

mc::MATLABApplicationMode::IN_PROCESS; // Specify MATLAB startup options

std::vector<std::u16string> options = {}; std::shared_ptr<mc::MATLABApplication>

matlabApplication = mc::initMATLABApplication(mode, options);

return matlabApplication;

}

Thanks,

Joe.

[Gerrith]

@Joe Elsom do you know if this is still the case for MATLAB2024b? Does the MCR now include the MATLAB Engine API for C++?

@Scott Dattalo thanks for your solution. Worked for me, after setting the enviroment variables and the Visual Studio project settings, I've just needed to restart Visual Studio for the effects to take place. I'm using MATLAB 2024b and VS2019

[Joe Elsom]

@Gerrith not sure honestly. I've since given up on the C++ API. There isn't any mention of the runtime in the docs from what i can see. The 2024b runtime bin/win64 folder does not contain libMatlabEngine.dll so i assume it is still not supported.

Answer 2

[Xiaodong Liu]

The solution is very simple. After setting include and library folders, DONOT copy ANY dll files to your own build folder. Then, it will work.

Answer 3

[gen]

Please "cd" to ${matlabroot}\extern\bin\win64, then invoke your testVal.exe using absolute path. That will do.