No integer overflow warning reported in the call to constructor

Question

I have the following snippet:

#include <iostream>

void func(uint64_t i) { std::cout<<i<<std::endl; }

class A{ 
public:
    A(uint64_t i){ std::cout<<i<<std::endl; } 
};

int main(int /*argc*/, char* /*argv*/[]) {
    func(1024 * 1024 * 1024 * 1024 * 1024ull); // clang [-Winteger-overflow]: Overflow in expression; result is 0 with type 'int'   
    A a(1024 * 1024 * 1024 * 1024 * 1024ull); // no error
    return 0; 
}

I know I should have written 1024ull * 1024 * ..., but I am wondering why my compiler (clang) only throws the overflow error at the line with func, and is not aware of the problem in the constructor

Answer 1

This is slightly simplified your code, where there are two similar signed integer overflows (already in 1024 * 1024 * 1024 * 1024):

#include <cstdint>

void func(uint64_t) {}

struct A { 
    A(uint64_t){} 
};

int main() {
    func(1024 * 1024 * 1024 * 1024 * 1024ull); // warning, ok
    A a(1024 * 1024 * 1024 * 1024 * 1024ull); // no warning in Clang!
}

The one in calling func is detected by Clang, GCC and MSVC with appropriate command line options -Wall -Wextra for Clang/GCC and /Wall for MSVC.

But the other one in calling A::A is not detected by Clang. Online demo: https://gcc.godbolt.org/z/6vGT33q1W

This is just a Clang bug, which is being fixed right now: https://github.com/llvm/llvm-project/issues/58944