In C, a symbol name matches the function name, if you have a function called 'void my_function()', the symbol will be 'my_function' and you will be able to load this function from a foreign function interface as libffi.load('my_function', ... types ...). The trouble with C++ bindings is that symbols are mangled, the function names are encoded, a function named 'Workaround1::get_callback()::callback' can be encoded (mangled) as _ZZN11Workaround112get_callbackEvE8callback - this encoding is specific to a given compiler ABI, in addition FFIs cannot load non-C types.
Tensorflow uses the hourglass design pattern, this library is implemented internally in C++, but does not exposes a C++ API, instead it exposes a C wrapper built with 'extern "C" and opaque pointers (void* pointers or incomplete types). This approach allows the library to be used in the same way as a C library and makes easier to call the library via FFI. Another benefit is the better binary compatibility as object-codes built with other compilers canlink against Tensorflow without any issues. Another library that uses this approach is the Squirrel embedded scripting language, that is implemented in C++, but only exposes a C API. So, this technique allows a C++ library to be used as C library and makes possible calling it through FFI.
I guess that in the case of LLVM it may also use this pattern of using C++ internally and only exposing a C API via extern "C". The disadvantage of this pattern is the need of lots cumbersome extern "C" wrappers for every member function of a class and for all exposed classes. LibClang LLVM may be helpful for automating the generation of this C wrapper code.
LLVM is a good example of C++ being able to interface with other languages using its C FFI, being completely written in C++ but having bindings for almost every programming language. I can count among them at least C, C++, D, Rust, OCaml and Python, for instance, and while it has an "LLVM-C" library available, you are not necessarily forced to use it; you can just export what you need directly from C++ using a file with a few functions defined as extern "C".
Rust is probably even trickier to interface with C but it's still pretty doable, given that GNOME is now rewriting several components in Rust (see librsvg).
I still think one of the main reasons C++ was discarded so quickly at the time, despite Qt and KDE being older than GNOME and in C++ already, was being born from a culture that had at the time already rejected the adoption of C++.
GNU has historically always discouraged its projects, especially those they cared about the most, from using C++. It's well documented how strongly RMS despised the language and how he actively pushed people towards using C for such tasks. I think it's still written in somewhere in the GNU website, even.
Heck the GCC developers had to ask RMS the permission to switch to C++ a few years ago, due to GCC becoming harder and harder to maintain. Ironically, it used internally its own private C implementation of dynamic vectors, while being one of the main implementers of the STL. I think it even had some sort of garbage collection inside, and that they largely reimplemented an object system too using C and macros (it's been a while since I read the GCC sources, so my memories could be incorrect). RMS wasn't really that happy about it, but it grumpily said they could.