Me too. I knew this was undefined in theory because C treats any kind of signed overflow as undefined behavior, but I'd never run into this specific issue. Trap

Me too. I knew this was undefined in theory because C treats any kind of signed overflow as undefined behavior, but I'd never run into this specific issue. Trapping it is a denial-of-service risk but at least it won't subtly break downstream invariants like most 2's complement signed overflow bugs. Some architectures have signed add/sub instructions that will optionally trap on overflow. If you're going to trap, it seems like you should do it across the board, rather than reserve it for one arithmetic instruction as with IDIV. The x86 IDIV/IMUL and DIV/MUL instructions have a nice left-inverse pairing (the multiplies take two n-bit operands to a 2n-bit result, the divides take a 2n-bit operand and an n-bit operand to an n-bit result) but that symmetry breaks down when you work with fixed-size operands and results as is usually the case.