Currently, the implementation:

• Works with pthreads only, but API is intentionally opaque to allow other implementations (Windows for instance).
• Starts all threads on creation of the thread pool.
• Reserves one task for signaling the queue is full.
• Stops and joins all worker threads on destroy.

The API contains addtional unused 'flags' parameters that would allow some additional options:

• Lazy creation of threads (easy)
• Reduce number of threads automatically (hard)
• Unlimited queue size (medium)
• Kill worker threads on destroy (hard, dangerous)
• Support Windows API (medium)
• Reduce locking contention (medium/hard)