Editor's Note: This is the fourth installment of a five-part series by Gigi Sayfan on creating cross-platform plugins in C++. Other installments are: Part 1, Part 2, Part 3, and Part 5.
This is the fourth article in a series about developing cross-platform plugins in C++. In the previous articles -- Part 1, Part 2, and Part 3 -- I examined the difficulties of working with C++ plugins in portable way due to the binary compatibility problem. I then introduced the plugin framework, its design and implementation, explored the life cycel of a plugin, covered cross-platform development and dived into designing object models for use in plugin-based systems (with special emphasis on dual C/C++ objects).
In this installment, I demonstrate how to create hybrid C/C++ plugins where the plugin communicates with the application through a C interface for absolute compatibility, but the developer programs against a C++ interface.
Finally, I introduce the RPG (Role Playing Game) that serves (faithfully) as a sample application that use the plugin framework and hosts plugins. I explain the concept of the game, how and why its interfaces were designed, and finally explore the application object model.
C++ Facade for Plugin Developers
As you recall, the plugin framework supports both C and C++ plugins. C plugins are very portable, but not so easy to work with. The good news for plugin developers is that it is possible to have a C++ programming model with C compatibility. This is going cost you though. The transition from C to C++ and back is not free. Whenever a plugin method is invoked by the application it arrives a C function call through the C interface. An elaborate set of C++ wrapper classes (provided by the application for use by plugin developers) will encapsulate the C plugin, wrap every C argument with non-primitive data type in a correponding C++ type and invoke the C++ method implementation provided by the plugin object. Then the return value (if any) must be converted back from C++ to C and sent through the C interface to the application. This sounds kinda familiar.
Isn't the dual C/C++ model with the automatically adapted C object exactly the same? In a word, "No." The situation here is totally different. The objects in question are always come from the application object model. The application instantiated them and it may convert a dual object between its C and C++ interfaces. On the plugin side, you get the C interface and you have no knowledge about the dual object. This knowledge is necessary to cast from one interface to the other. In addition, even if the plugin knew the type of the dual object it wouldn't be enough because the appliucation and the plugin might have been built using different compilers, different memory models, different calling conventions, etc. The physical layout in memory of the same object might be very different. If you can guarantee that the application and the plugins are vtable-compatible just use the direct C++ interface.