c++ - Implicitly deleted function overrides deleted virtual function - Stack Overflow

admin管理员组

文章数量:1353311

This is more of a theoretical question about the interpretation of the standard in one particular case, given below:

struct B;

struct A {
    virtual bool operator ==(const B&) const = delete;
};

struct B : A {
    bool operator ==(const B&) const = default;
};

This program is rejected by GCC, EDG, MSVC with the

error: nondeleted function overrides deleted function "A::operator=="

which looks reasonable according to :

A deleted function can only be overridden by deleted functions

But Clang actually admits the program with the

note: defaulted 'operator==' is implicitly deleted because there is no viable 'operator==' for base class 'A'

Online demo:

Which implementation is correct here?

This is more of a theoretical question about the interpretation of the standard in one particular case, given below:

struct B;

struct A {
    virtual bool operator ==(const B&) const = delete;
};

struct B : A {
    bool operator ==(const B&) const = default;
};

This program is rejected by GCC, EDG, MSVC with the

error: nondeleted function overrides deleted function "A::operator=="

which looks reasonable according to https://en.cppreference/w/cpp/language/function :

A deleted function can only be overridden by deleted functions

But Clang actually admits the program with the

note: defaulted 'operator==' is implicitly deleted because there is no viable 'operator==' for base class 'A'

Online demo: https://gcc.godbolt./z/n83nGE7dW

Which implementation is correct here?

• c++
• language-lawyer
• c++20
• virtual-functions
• deleted-functions

Share Improve this question Follow asked Mar 31 at 18:40 FedorFedor 21.6k3737 gold badges5555 silver badges168168 bronze badges 5

• 3 Aren't they all correct? They are all issuing a diagnostic which is all that is really required for ill formed code IIRC. – NathanOliver Commented Mar 31 at 18:45
• While I suspect Clang is, strictly speaking, conforming by just issuing a diagnostic, the program should definitely be rejected. Even the warning text isn't really correct: the defaulted operator== can't be implicitly deleted, since it simply can't be overriden to be defaulted. – cigien Commented Mar 31 at 19:06
• @cigien — the point of only requiring a diagnostic is to allow compiler-specific extensions. Making B::operator== implicitly deleted sounds like a reasonable extension. – Pete Becker Commented Mar 31 at 19:24
• @PeteBecker Hmm, in general, yes, but this one being intentional seems odd. I would be curious to see what the response is to an [accepts-invalid] bug report. – cigien Commented Mar 31 at 19:33
• Doesn't defaulting it simply make it deleted? It's not like it's actually making it callable. If one tries that, clang refuses too. – Ted Lyngmo Commented Mar 31 at 20:45

Add a comment  | 

1 Answer 1

Sorted by: Reset to default 3

The standard is just vauge enough on this to make me accept both interpretations (of Clang vs the others) as valid. Because as written we have

[class.virtual]

17 A deleted function ([dcl.fct.def]) shall not override a function that is not deleted. Likewise, a function that is not deleted shall not override a deleted function.

The standard only says that we can't have a mismatch of non-deleted to deleted (implicitly or explicitly - [dcl.fct.def.delete]/1) between base definition and overrider (or vice-versa). What's absent is that there's no explicit requirement for the two functions to be deleted in the same way. I.e. nothing says both have to be explicitly deleted, or implicitly deleted; nothing says we can't have them deleted in different ways, like the OP does.

Now, is it reasonable to allow it as Clang does, with a warning? Yes, it is. We don't violate anything in the standard to warrant a diagnostic (to my reading), and there's an "obvious" meaning to the code since the function is deleted either-way. Furthermore, the warning tells us we might have done somthing we didn't mean to. Clang encourages us to delete explicitly if we wish to keep the overrider and silence the warning.

Is it reasonable to reject the code until we explicitly delete? I'd say "yes" to that as well, it's a small change at the end of the day, that Clang is suggesting as well. Not unreasonable to require the programmer to be explicit.

On the whole, this might warrant a small core issue to get all compilers to the same baseline. But there isn't really a whole lot broken here.

本文标签： cImplicitly deleted function overrides deleted virtual functionStack Overflow