I am a teaching assistant for the Data Structures course at my school. I wrote the reference implementation and a data structure that the students could use for their first project: Datum. Datum is supposed to be able to hold integers, booleans and “codeblocks” (glorified strings) for the small postfix calculator students are building. The idea is that they have a minimal runtime type system.

Since this class is taught in C++ and currently the students are trying to grasp interface / implementation boundaries, I wrote the interface like this:

class Datum {
    public:
        Datum(const Datum &d);  /* Copy constructor */
        Datum(int i);           /* Make a Datum from an int, bool, or string */
        Datum(bool b);
        Datum(const char *s);
        Datum(std::string s);

        /* These throw exceptions if they are called on the wrong type. */
        int getInt();
        bool getBool();
        std::string getCodeBlock();
        std::string toString();

    private:
    /* Some private variables. */
};

so that students could build Datums with expressions like Datum(true), Datum(5), Datum("Hello, world!").

The first students started posting on the online course forum with problems like “I am trying to print the top Datum on the stack but no matter its value, the program always prints true”. Then people came into office hours because they could not fix their bugs, and I noticed that at least four people had the same problem. I grabbed Aubrey, another TA, and we started investigating what was going wrong.

After around 20 minutes, we managed to narrow the problem down to a very minimal test case:

#include <iostream>
#include "Datum.h";

int main() {
    Datum d = new Datum(5);
    std::cout << d.toString() << endl;
}

People who read too much code for a living will have immediately stopped and thought, “But Max, that won’t even compile. The right side of that expression returns a pointer, and the left side is not even close to a pointer.”

That’s true! But as we discovered, the C++11 standard (PDF) has implicit type conversions:

4.12 Boolean conversions

A prvalue of arithmetic, unscoped enumeration, pointer, or pointer to member type can be converted to a prvalue of type bool. A zero value, null pointer value, or null member pointer value is converted to false; any other value is converted to true. For direct-initialization (8.5), a prvalue of type std::nullptr_t can be converted to a prvalue of type bool; the resulting value is false.

and in our case this happens in the copy-initialization of d.

So what does that mean? It means that the right side (of type Datum *), did not match against Datum(int), Datum(std::string), or Datum(const char *), but instead matched against Datum(bool). Since it found something to match against, it did not generate so much as a warning (yes, even with -Wall -Wextra -Weverything -pedantic) and silently confused the heck out of the poor Data Structures students. The students, who would normally have seen a compiler error, adjusted the left side to match, and gone on their merry ways were now left with broken code.

The solution

The solution is one word long: add the keyword explicit before the bool constructor:

class Datum {
    public:
        Datum(const Datum &d);  /* Copy constructor */
        Datum(int i);           /* Make a Datum from an int, bool, or string */
        explicit Datum(bool b);
        Datum(const char *s);
        Datum(std::string s);

    /* Some other stuff. */
};

and call it a day. explicit ensures that the type must match the constructor parameter exactly. This way, a compile-error is raised when students attempt the above code. Another solution, I suppose, is to stop using C++ entirely.