How are the C and C++ Interfaces working?

I worked with both MATLAB C++ Interface (clibgen package) and Matlabs C Interface (callib, loadlibrary, ...). Now, I am interested in getting to know how these interfaces are implemented:

Regarding the C++ interface:

I noticed that Matlab does not "really" call the C++ functions but builds a C interface DLL instead. To do this, the clibgen package reads the header files, creates a temporary C++ file and builds a shared library from it. The c++ file supplies a C interface as a proxy to call C++ functions via a C binary interface.

When all the prerequisites are done, the C++ functions can be used by prefixing them with "clib.packagename", e.g. clib.package.DoStuff(). I am particularly interested in how Matlab resolves the expression "clib.package.DoStuff" into a string key that can be passed to the "cppGetPassThrough" function. How does Matlab know what functions and classes are available in a DLL? 

Regarding the C interface:

I suspect that loading a DLL also invokes a call to some compiler. I am not entirely sure about but it might have something to do with compatibility to 64bit?

Answer 1

[MathWorks Support Team]

Regarding the C++ interface:

C++ interface embeds an XML file into the shared library during the build step. This XML contains mappings between library constructs and the pass-throughs in the shared library. When you call “clib.packagename.foo”, MATLAB retrieves the XML from the shared library to find and call the pass-through in the shared library for construct 'foo'.  

 

Regarding the C interface:

Keep in mind that "loadlibrary" and "calllib" are legacy features for pure C libraries. Having said that, a thunk file is generated and built to adjust the MATLAB calling conventions to the native DLL calling conventions. Moving forward, it is highly recommended that new applications use the C++ Interface. The C++ Interface can generate and call a C library that has the appropriate extern C linkage for handling mangled names.