size_t in public API is a no-no

[Michi Henning]

I missed this when reviewing the branch because, of course, the bloody symbol mangling turns size_t into unsigned int or unsigned long.

We can't have size_t in the public API. That's not negotiable. size_t is an implementation-defined type that can differ across architectures. The correct fix is not to add arch-specific entries to the symbols file, but to not use size_t in the first place. (size_t is being abused here anyway because it doesn't express the concept of "non-negative".)

For a very cogent and detailed explanation, see John Lakos, "Large-Scale C++ Software Design".