A.8 Summary

Threads are absent from standard C++. However, synchronization issues in multithreaded programs pervade application and library design. The trouble is, the threading models supported by various operating systems are very different. Therefore, Loki defines a high-level synchronization mechanism having a minimal interaction with a threading model provided from the outside.

The ThreadingModel policy and the three class templates that implement ThreadingModel define a platform for building generic components that support different threading models. At compile time, you can select support for object-level locking, class-level locking, or no locking at all.

The object-level locking strategy allocates one synchronization object per application object. The class-level locking strategy allocates one synchronization object per class. The former strategy is faster; the second uses a smaller amount of resources.

All implementations of ThreadingModel support a unique syntactic interface. This makes it easy for the library and for client code to use a uniform syntax. You can adjust locking support for a class without incurring changes to its implementation. For the same purpose, Loki defines a do-nothing implementation that supports a single-threaded model.