C++ Style: Prefixing virtual keyword to overridden methods

Question

I've been having a discussion with my coworkers as to whether to prefix overridden methods with the virtual keyword, or only at the originating base class.

I tend to prefix all virtual methods (that is, methods involving a vtable lookup) with the virtual keyword. My rationale is threefold:

Given that C++ lacks an override keyword, the presence of the virtual keyword at least notifies you that the method involves a lookup and could theoretically be overridden by further specializations, or could be called through a pointer to a higher base class.
Consistently using this style means that, when you see a method (at least within our code) without the virtual keyword, you can initially assume that it is neither derived from a base nor specialized in subclass.
If, through some error, the virtual were removed from IFoo, all children will still function (CFooSpecialization::DoBar would still override CFooBase::DoBar, rather than simply hiding it).

The argument against the practice, as I understood it, was, "But that method isn't virtual" (which I believe is invalid, and borne from a misunderstanding of virtuality), and "When I see the virtual keyword, I expect that means someone is deriving from it, and go searching for them."

The hypothetical classes may be spread across several files, and there are several specializations.

class IFoo {
public:
    virtual void DoBar() = 0;
    void DoBaz();
};

class CFooBase : public IFoo {
public:
    virtual void DoBar(); // Default implementation
    void DoZap();
};


class CFooSpecialization : public CFooBase {
public:
    virtual void DoBar(); // Specialized implementation
};

Stylistically, would you remove the virtual keyword from the two derived classes? If so, why? What are Stack Overflow's thoughts here?

I'd be interested to see Bjarne's (or the C++ committee's) rationale for why virtual is even allowed to be omitted in the derived class. The rationale might provide more compelling reasons for actually omitting it than does your colleague. Perhaps only in certain cases, though. — Steve Jessop
Unfortunately there is no rationale included in the spec. It just says it is "legal but redundant (has empty semantics)". Would be interesting to hear, though, if there is any record of what the rationale was. — Tyler McHenry
I don't have a copy of "design and evolution", which may or may not contain anything about it. — Steve Jessop
I'd like to point out that as of C++11, you can also use the override specifier to, well, specify that a function is virtual and overrides another function; doing so also allows the compiler to give you an error message if the function doesn't override anything. With your example, CFooBase and CFooSpecialization would have virtual void DoBar() override;. — Justin Time - Reinstate Monica
@JustinTime I think what people want, as impractical as it is, is a Standard-enforced way to require descriptive vfunc declarations, rather than having to remember to deliberately type yet another keyword that is totally optional without having to turn on compiler-dependent warnings. But I think it's too late for that now in terms of what it'd break and the customary backlash that would result. That said, here's the trick for g++: stackoverflow.com/questions/29145476/… — underscore_d

Tyler McHenry Tyler McHenry · Accepted Answer · 2009-09-03T01:12:56

I completely agree with your rationale. It's a good reminder that the method will have dynamic dispatch semantics when called. The "that method isn't virtual" argument that you co-worker is using is completely bogus. He's mixed up the concepts of virtual and pure-virtual.

C++ Style: Prefixing virtual keyword to overridden methods

6 Answers