Why use copy-on-write? I've heard that causes problems with multi-threaded code.

Some of the numbers that the library deals with can get huge, making copying an expensive proposition. I've added move semantics using Boost.Move, but they're limited in what they can do; the emulated version, for C++03 compilers, can't be used with the standard containers, for example. Copy-on-write bypasses those limitations.

It can cause problems when you try to use the library from multiple threads at the same time, which is why it's only used in the (single-threaded) internal functions. By default, when you get a number back from the library, it has its own unique storage, so it can be safely used across multiple threads; if you don't need to use integer objects outside of the thread that they're created in, you can disable that by setting the Threadsafe template parameter to false.

Note that even with the Threadsafe option set to true, you still have to ensure that only one thread accesses a particular integer at a time.