This PR affirms that all virtual functions, *including destructors*,
should be declared exactly one of `virtual`, `override`, or `final`, and
takesa pass through the document to make the examples and guidance
consistent with that.

Of course a virtual destructor is a virtual function: It behaves
polymorphically, and it has a vtable entry that can be overwritten ==
overridden in a derived class exactly the same as any other derived
virtual override. See also [class.virtual]/7: "Even though destructors
are not inherited, a destructor in a derived class overrides a base
class destructor declared virtual; see [class.dtor] and [class.free]."

However, the following exception text currently appears in C.128:

> If a base class destructor is declared `virtual`, one should avoid
declaring derived class destructors  `virtual` or `override`. Some code
base and tools might insist on `override` for destructors, but that is
not the recommendation of these guidelines.

... but this exception is (a) not well-founded, and (b) inconsistent
with the Guidelines' practice in other examples and with the rationale a
few lines earlier for C.128 itself.

Re (a):

- The exception is overly broad: The rationale given for this exception
is entirely against marking destructors `override` (not `virtual`). So
clearly the exception to write neither keyword is too broad: At most,
the exception should be to write `virtual` rather than `override`.
- Explicit `virtual` is primarily for class users, not class authors:
The arguments given in #721 favoring this exception are from the
viewpoint of the implementation of the function (even then, the
arguments are debatable and debated). But `virtual`, `override`, and
`final` are primarily for the far larger audience of *class users and
call sites* of the function, for whom of course we should document each
declared function that is polymorphic, *especially* the destructor --
this tells calling code whether the function is safe to call through a
(smart or built-in) pointer or reference to base, which will nearly
always be the case for such types. We should not make the reader of the
code go way to look in the base classes to figure out whether a function
declared in this class is virtual or not -- the reason this Item exists
is primarily to avoid that implicit virtual antipattern via convention
and automated enforcement. For class users, all virtual functions
including destructors are equally polymorphic.

Re (b): The Guidelines already don't follow this. For instance, two
Items later (in C.130) we have this example that correctly uses
`override`:

~~~
virtual ~D() override;
~~~

... though per C.128 it should not also specify `virtual` (also fixed in
this PR).

Finally, the exception also contradicts the rationale given earlier in
the same Item.
This commit is contained in:
hsutter 2019-06-16 13:35:52 -07:00
parent 4b414458cf
commit fcb2960793

View File

@ -1,6 +1,6 @@
# <a name="main"></a>C++ Core Guidelines
May 2, 2019
June 16, 2019
Editors:
@ -6305,6 +6305,7 @@ Worse, a direct or indirect call to an unimplemented pure virtual function from
virtual void f() = 0; // not implemented
virtual void g(); // implemented with Base version
virtual void h(); // implemented with Base version
virtual ~Base(); // implemented with Base version
};
class Derived : public Base {
@ -6977,8 +6978,6 @@ It's simple and clear:
* `override` means exactly and only "this is a non-final overrider."
* `final` means exactly and only "this is a final overrider."
If a base class destructor is declared `virtual`, one should avoid declaring derived class destructors `virtual` or `override`. Some code base and tools might insist on `override` for destructors, but that is not the recommendation of these guidelines.
##### Example, bad
struct B {
@ -7260,7 +7259,7 @@ Copying a polymorphic class is discouraged due to the slicing problem, see [C.67
class B {
public:
virtual owner<B*> clone() = 0;
virtual ~B() = 0;
virtual ~B() = default;
B(const B&) = delete;
B& operator=(const B&) = delete;
@ -7269,7 +7268,7 @@ Copying a polymorphic class is discouraged due to the slicing problem, see [C.67
class D : public B {
public:
owner<D*> clone() override;
virtual ~D() override;
~D() override;
};
Generally, it is recommended to use smart pointers to represent ownership (see [R.20](#Rr-owner)). However, because of language rules, the covariant return type cannot be a smart pointer: `D::clone` can't return a `unique_ptr<D>` while `B::clone` returns `unique_ptr<B>`. Therefore, you either need to consistently return `unique_ptr<B>` in all overrides, or use `owner<>` utility from the [Guidelines Support Library](#SS-views).
@ -7545,6 +7544,7 @@ Without a using declaration, member functions in the derived class hide the enti
public:
virtual int f(int i) { std::cout << "f(int): "; return i; }
virtual double f(double d) { std::cout << "f(double): "; return d; }
virtual ~B() = default;
};
class D: public B {
public:
@ -7632,6 +7632,7 @@ That can cause confusion: An overrider does not inherit default arguments.
class Base {
public:
virtual int multiply(int value, int factor = 2) = 0;
virtual ~Base() = default;
};
class Derived : public Base {
@ -7659,7 +7660,7 @@ If you have a class with a virtual function, you don't (in general) know which c
##### Example
struct B { int a; virtual int f(); };
struct B { int a; virtual int f(); virtual ~B() = default };
struct D : B { int b; int f() override; };
void use(B b)
@ -7702,6 +7703,7 @@ Flag all slicing.
struct B { // an interface
virtual void f();
virtual void g();
virtual ~B();
};
struct D : B { // a wider interface
@ -21507,6 +21509,8 @@ Here is an example of the last option:
p->post_initialize();
return p;
}
// ...
};