std::exclusive_scan narrows to type of initial value

[upsj]

#include <vector>
#include <numeric>
#include <iostream>
#include <limits>
#include <cstdint>

int main() {
        std::vector<std::int64_t> vec{1LL << 32, 0};
        std::exclusive_scan(vec.begin(), vec.end(), vec.begin(), 0);
        std::cout << vec[0] << '\n';
        std::cout << vec[1] << '\n';
}

I would expect this to output 0 and 1LL<<32, but it outputs 0, 0, because T in std::exclusive_scan is deduced to int, which is the computational type that will be used in exclusive_scan. Not sure if this is a library or standard defect, but I think this is pretty bug-prone otherwise. Related to NVIDIA/cccl#836

https://godbolt.org/z/zMf5M9hGK

[jwakely]

This is the subject of https://www.open-std.org/jtc1/sc22/wg21/docs/papers/2018/p0571r2.html#intermediate_unordered which proposes to make it explicit that the summation is done with the type of the init parameter.