This is a header-only fixed point arithmetic library. It is used to build a fixed-point floating number based controller, and the library's primary goal is to simulate an ALU working with fixed-point floating numbers. Therefore, overflow is allowed and naturally expected in some cases. There are some manual test cases, but the reference implementation is spfpm library, and both libraries are tested with randomly generated test cases.
I couldn't exactly match the spfpm behaviour and our library's behaviour in bit clamps and especially overflow. It seems that will be future me's problem. But I expect the clamp mode to work perfectly fine.