templated opaque forward declaration

Question

I am trying to create a synchronized double ended queue that will have different implementations on different OSs (for example, std:: mutexes and condition variables on Android, Windows, Linux, etc - implementation in sync_deque_stl.cpp; grand central dispatch on iOS/MacOS - implementation in sync_deque_gcd.cpp). I am also using gyp to generate my projects.

1. I have a sync_deque.hpp file with the following definitions :

   template <class T> struct opaque_sync_deque;
   int insert(opaque_sync_deque<class T>&, T& item);
   ...
   

2. Then, I use gyp to create :

   • an Xcode project on MacOS/iOS to include sync_deque_gcd.cpp, but no sync_deque_stl.cpp
   • a Visual Studio project on Windows to include sync_deque_stl.cpp, but no sync_deque_gcd.cpp
   • (basically, each OS gets its own implementation .cpp of the sync_deque, but the header file is common between the OSs)

3. in the .cpp files, I use c++ type aliasing (example taken from sync_deque_stl.cpp) :

   template<typename T>
   using opaque_sync_deque = sync_deque_stl<T>;
   

which is valid C++11 syntax but it clashes with the forward declaration in sync_deque.hpp :

sync_deque_stl.cpp:29:1: error: redefinition of 'opaque_sync_deque' as different kind of symbol using opaque_sync_deque = sync_deque_stl;

Is it possible to still keep this setup (opaque template type and function signatures in the .hpp shared by all the OSs + implementation in separate .cpp files for different OSs) and still have it compile ? I believe I could have separate /gcd and /stl and separate sync _deque header files in each directory and then use gyp to select the right header file, but it would be easier to have only one sync_deque.hpp. Any help is appreciated.