C++ is a great language for writing applications that will run on a number of platforms. There are compilers for many different platforms, and some powerful libraries you can use. Still, in the end your code needs to talk to the operating system, and that means that most cross platform applications have at least a few little corners where platform-specific code lives. The challenge is how to ensure that your Windows code runs on Windows, your Linux code runs on Linux, etc, without maintaining a number of different branches or hand-merging and splitting every time you deploy.

Being C++, an unspoken requirement in solving this problem is "be as fast as possible". Putting everything through a library and paying for extra indirection, looking up something that cannot change once the program has started executing, is not as fast as possible. You also want developer convenience and productivity. If you support five platforms, and something is the same on four and different on one, copying that code around for the four that are the same is not a productive way to behave. You would like a default behaviour, and then special code for special cases.

Michael Tedder has an intriguing approach using templates. As he says:

Instead of declaring a base interface class with virtual functions then deriving each platform with a different implementation, we declare a class with one template parameter — a platform ID — then specialize it to provide a different implementation for each platform.  The template class is then typedef‘d to expose the specialization for the platform ID being compiled to the application, allowing the implementation to be used without any virtual functions and also allow for inlining of functions as well.

He has some pretty convincing armwaving about using this not for just Windows/Linux/Android but for any hardware differences even on the same operating system - like what kind of graphics architecture you have or anything else that can't change at runtime. It's a good example of how the power of templates makes things possible that would always incur a runtime cost in any other language, or a significant burden on a developer to move code around building custom versions of an application. Worth a read!

Kate