An argument pro liberal use of [[nodiscard]]

Consider the following C++03 code (recently written by a student in a class of mine):

#include <iostream>
#include <vector>

int main() {
    std::vector<int> myVector;


    std::vector<int>::const_iterator it;
    for (myVector.begin(); it != myVector.end(); ++it) {
        std::cout << *it << std::endl;

When you compile and run this program, you get a segfault.

How long did it take you to spot the bug?

The original sin in this code snippet was actually the definition of variable it without any initializer. If we’d written the definition right at the point of initialization —

    typedef std::vector<int>::const_iterator CIT;
    for (CIT it = myVector.begin(); it != myVector.end(); ++it) {
        std::cout << *it << std::endl;

— then there’d have been no bug. Even better, we could have used C++11’s auto or ranged for loop:

    for (auto it = myVector.begin(); it != myVector.end(); ++it) {
        std::cout << *it << std::endl;

    for (auto&& elt : myVector) {
        std::cout << elt << std::endl;

Lessons learned:

  • C++11 does make a lot of things safer.

  • Not all segfaults are due to subtle arcane issues — every so often, it’s the simple stuff that gets you.

Posted 2018-11-06