Can C++ have code in the global scope?

Question

Is it legal to have code (which compiles to assembly instructions in the global scope of a C++ source file? Previously, I was under the impression that except for the Ch programming language (an interpreter for C/C++), you cannot have code in the global scope of a C++ program. Code/instructions can only be inside the body of a function [period]!

However, I found out that you can call functions before the main function in C++ by assigning them to a global variable! This would involve a call instruction in the assembly code. Also you can assign the sum of two variables into another global variable outside the assembly code. That would almost certainly involve an add and mov instructions. And if that code is in the global scope, outside of any function, when would it execute? If the + were an overloaded operator of a class type, if it had a print statement inside of it, when would that execute?

Also can you have loops and control structures in the global scope of a C++ program, and if so when are they executed? What about for other program constructs, are they allowed in the global scope, and under what circumstances, and when are they executed?

This question is in a response to a previous question that I posted: Why can't I assign values to global variables outside a function in C?

The answerer to the original question asserts that you cannot have code outside of the scope of a function. I think that I do not fully understand the rules for this, and what exactly is considered to be "code" or not.

int foo() {
    cout << "Inside foo()" << endl;
    return 5;
}

// is this not code?
int global_variable = foo();

// How does this statement work without generating code?
int a = 4;
int b = 5;
int c = a + b;

int main() {
    // The program behaves as if the statements above were executed from
    // top to bottom before entering the main() function.
    cout << "Inside main()" << endl;
    cout << "int c = " << c << endl;
    return 0;
}

Answer 1

The answer on the question you linked to was talking in a simple way, not using strict C++ naming for constructs.

Being more pedantic, C++ does not have "code". C++ has declarations, definitions, and statements. Statements are what you probably think of as "code": if, for, expressions, etc.

Only declarations and definitions can appear at global scope. Of course, definitions can include expressions. int a = 5; defines a global variable, initialized by an expression.

But you can't just have a random statement/expression at global scope, like a = 5;. That is, expressions can be part of definitions, but an expression is not a definition.

You can call functions before main of course. Global variable constructors and initializers which are too complex to be executed at compile time have to run before main. For example:

int b = []()
{
    std::cout << "Enter a number.\n";
    int temp;
    std::cin >> temp;
    return temp;
}();

The compiler can't do that at compile-time; it's interactive. And C++ requires that all global variables are initialized before main begins. So the compiler will have to invoke code pre-main. Which is perfectly legal.

Every C++ compilation and execution system has some mechanism for invoking code before and after main. Globals have to be initialized, and object constructors may need to be called to do that initialization. After main completes, global variables have to be destroyed, which means destructors need to be called.