Plugin interface is unnecessarily complex

[Alan Griffiths]

The code to manage and validate plugin loading - like PluginClassHandler is complex beyond the needs of the problem it solves.

For instance: extern unsigned int pluginClassHandlerIndex; - is only ever referenced by PluginClassHandler (and a test harness) and could therefore be a private static member (with a const accessor to support the test.)

The class PluginClassHandler<class Tp, class Tb, int ABI = 0> is only instantiated on two Tb types - CompScreen and CompWindow and the use of ABI is restricted to one output string (which doesn't seem to justify instantiating separate types).

Also the function - PluginClassHandler<Tp,Tb,ABI>::initializeIndex() - is only ever instantiated for Tb = CompScreen.

---

VTableForScreenAndWindow<> seem to duplicate code that is in VTableForScreen<> - why not VTableForScreenAndWindow<...> : public VTableForScreen<...> as the name suggests?

And the duplication of code between functions like:

template <typename T, typename T2>
bool CompPlugin::VTableForScreenAndWindow<T,T2>::initScreen (CompScreen *s)
{
    T * ps = T::get (s);
    if (!ps)
 return false;

    return true;
}

and

template <typename T, typename T2>
bool CompPlugin::VTableForScreenAndWindow<T,T2>::initWindow (CompWindow *w)
{
    T2 * pw = T2::get (w);
    if (!pw)
 return false;

    return true;
}

Strongly suggests a missing abstraction.