The asserts can go.  I left them in figuring a crashing interpreter on a
debug build in someones sandbox or on a buildbox would get more
attention than a SystemError being raised.  I doubt that is a worthy
assumption on my part.

Both a crash and a SystemError are notable events.

shall I get rid of the asserts?

As for why i dislike signed size types... tons of reasons:
 * It wastes half the range of the integer.
 * It leads to security bugs.
 * on return values -1 and < 0 tests may be convenient to type but they 
   could just as easily compare to a known value defined as a constant; 
   all the things alexander belopolsky suggested.
 * sizes being passed -in- to a function never need to be negative
   meaning safe code requires extra checks like these.
 * sign extension of values going between registers of different sizes
   is needlessly expensive on some cpu architectures.  use unsigned
   types whenever possible for the best code.

anyways, I figure the Python C API is already set in stone using the
signed types so its too late to "fix" it without causing headaches
around the world.