Slight nitpick: > Here’s a somewhat non-exhaustive list of all the undefined behavior of signed integer arithmetic in LLVM which applies to all languages which

Slight nitpick: > Here’s a somewhat non-exhaustive list of all the undefined behavior of signed integer arithmetic in LLVM which applies to all languages which use LLVM: > [... division examples ...] > INT_MAX - INT_MIN That's not true: 'sub i32 INT_MAX, INT_MIN' is defined perfectly fine in LLVM in the (modular arithmetic) way you'd expect as long as the 'nsw' (no signed wrap) flag isn't set on the instruction. For the most part, LLVM IR by default doesn't care about the signedness of values, integer types are just treated as a bunch of bits with modular twos complement arithmetic. Exceptions to this rule are explicit, e.g. the mentioned nsw flag (and there's also nuw for "no unsigned wrap"). The division examples are correct because if you want a signed divide, you have to explicitly encode it as a signed divide instruction in LLVM IR, which has the listed undefined behaviors baked in.