C. Finegan C. Finegan - 1 month ago 10
C++ Question

Deriving class from virtual base with no default constructor

I'm writing a small hierarchy of exception classes for a C++ application I'm developing, and I'm having trouble deriving indirectly from

std::runtime_error
. Here is code analogous to what I've written so far:

class RuntimeException : public virtual boost::exception, public virtual std::runtime_error {
public:
virtual ~RuntimeException() {}
RuntimeException() : runtime_error("A RuntimeException occurred.") {}
RuntimeException(const std::string& what) : runtime_error(what) {}
};

class IllegalArgumentException : public virtual RuntimeException {
public:
IllegalArgumentException() : RuntimeException("An IllegalArgumentException occurred.") {}
IllegalArgumentException(const std::string& what) : RuntimeException(what) {}
};


The
RuntimeException
class compiles without issue, but
IllegalArgumentException
refuses to compile on VS2015, generating the error:
no default constructor exists for class "std::runtime_error"
for both constructors of
IllegalArgumentException
. This challenges my understanding of C++ inheritance hierarchies, as I expected this code to compile fine.

My understanding is that
IllegalArgumentException
should compile because, although it is true that
std::runtime_error
does not have a default constructor, its constructor is being called by the constructor for
RuntimeException
. But obviously this must must be false, as the compiler is rejecting it. It seems to want me to call the
std::runtime_error
constructor directly from the
IllegalArgumentException
constructor (the compiler error goes away when I do so), but this seems wrong because then I would be calling the constructor for
std::runtime_error
twice: once in the constructor for
RuntimeException
, and again in the constructor for
IllegalArgumentException
.

Is this safe and/or efficient to do? If not, why does the compiler seem to encourage it? I could just derive from
std::exception
and implement the
std::string
myself as a member variable, but I thought it would be easier to derive from a standard class that has already implemented this. Is this the wrong approach to take? Additionally, is the fact that I'm deriving virtually from both
boost:exception
and
std::runtime_error
contributing to this issue?

Answer

When using virtual inheritance the constructor call of the virtual base is the responsibility of the most derived class rather than the responsibility of any intermediate class. The reason is obvious: the use of virtual inheritance indicates that there is an expectation that there are actually multiple derived classes using the base class. Which one of these derived classes would be responsible for constructing the virtual base?

So, the constructor of any of the derived classes needs to provide an argument to the virtual base, e.g.:

IllegalArgumentException::IllegalArgumentException(std::string const& what)
    : std::runtime_error(what)
    , RuntimeException(what) {
}

To avoid having intermediate bases call the constructor of the virtual base classes intended for virtual inheritance often provide a default constructor. Of course, that opens up the possibility that the most derived class incorrectly relies on the proper constructor being called by one of its bases.

Comments