Emit the `[[maybe_unused]]` attribute on anonymous objects to silence C++ compiler warnings

[bluetarpmedia]

C++ compilers at high warning levels warn about unused variables or class members.

This PR adds [[maybe_unused]] to anonymous objects (named _) at function and class scope (excludes namespace scope, since it's not needed there).

Using this code as an example:

foo: type = {
    _: double = 90.0;
}

_: foo = ();

ns: namespace = {
    _: bool = true;
}

main: () -> int = {
    _: int = 0;
    return 0;
}

Before this PR, Clang produces these warnings (with -Wall). Similar warnings with MSVC and GCC.

main.cpp2:12:9: warning: unused variable 'auto_1' [-Wunused-variable]
   12 |     int auto_1 {0}; 
      |         ^~~~~~
main.cpp2:2:21: warning: private field 'auto_1' is not used [-Wunused-private-field]
    2 |     private: double auto_1 {90.0}; 
      |                     ^

After this PR, there are no warnings, since the relevant code lowers to:

private: [[maybe_unused]] double auto_1 {90.0}; 

and

[[maybe_unused]] int auto_1 {0}; 

[bluetarpmedia]

I think you should change the tests instead. You shouldn't be making anonymous objects using a trivial constructor with trivial destruction.

The motivation wasn't to fix the warnings in the tests (that's just a byproduct), but rather for cases like this:

initialize: () -> int = {
    std::cout << "Setup\n";
    return 0;
}

main: () -> int = {
    _:int = initialize();
    return 0;
}

It's arguable that programmers should always check return values, but in this case the user has deliberately written a placeholder to say that they don't care about the return value. But this results in an unused variable warning, even though the user said they don't want it. Godbolt

[JohelEGP]

Well, you shouldn't make an object out of it, but instead discard it by leaving out the : and the type: _ = initialize();.

[bluetarpmedia]

Well, you shouldn't make an object out of it, but instead discard it by leaving out the : and the type: _ = initialize();.

Oh, of course, I completely missed that!

Hmm, so maybe cppfront could suggest to write _ = initialize(); instead of _:int = initialize();