BerkeleyDB is the same way except that it is often unclear how much of the underlying C API is supported. You've got to be clear in *every* case where your API differs from the C API you are wrapping.