The attached patch is intended as a 1st step towards separation of the "locking policy" parts from the "synchronization" parts of shared_mutex. This separation is achieved by moving state_data member checks and manipulation into new methods in state_data. The separation is currently only done for exclusive and shared locking, but I will complete the separation to also cover upgrade locking if there is consensus that this is a good strategy. state_data can then be encapsulated with all data members private.
Advantages of the separation:
* Simplify implementation of different locking prioritization policies: Only requires reimplementation of state_data, with (hopefully) no changes to the outer shared_mutex class.
* Simplify implementation of "lock-free" synchronization (as in win32 counterpart): Can be done in shared_mutex without touching the "locking policy" logic in state_data.