How to structure a large MATLAB Project with custom MEX Files?

I work on a large MATLAB Codebase and for performance reasons we had to rewrite some functions in c++, compile them into a MEX File and then run them instead of the MATLAB Functions.

Now I store the MATLAB Code, the c++ Code and the MEX Files together in the same repository. This approach works, but is cumbersome. If I change something in the c++ Code I have to compile it by hand and then copy the new MEX File into the right spot. This approach makes debugging a nightmare and is also a bad practice as you shouldn't store artifacts together with the corresponding code.

Ideally I would like to change something in the c++ code, test it without manual compiling and then commit just the changed code.

Is there a best practice for this kind of scenario? Currently I use cmake for the mex files because of dependency management. And I don't know of any just in time mechanism of compiling the MEX Files via cmake from MATLAB.

Answer 1

[Yongjian Feng]

Some advices:

1. Might not be a good idea to store the MEX files in the repo. Use a build machine dedicated to compile. Then your MEX files can always be regenerated when needed
2. Use unittest to test your C++ code. It is just C++ code, create C++ unite tests to test them.
3. Store those unit tests in your repo as well.

[Lukas Post]

The Problem is, that the MEX Files have to be somewhere for them to be called from MATLAB. Is it really a better Idea to setup a whole CI/CD Pipeline and Artifactory than just to compile the code and store the mex files until the code changes? Seems a little bit Overkill for what I assume can be fixed with a little bit of reorganisation.

[Yongjian Feng]

Put this in your build machine script. The build script shall

1. Compile the C++ code and generate MEX files
2. Move the MEX files to the correct locations