Dealing With Exception From Initialization

Okay. That title was wierd to phrase. Really it's more like this. I have an object that is a field of my class that I construct using an initializer list, and it's constructor has the potential of throwing an exception. I don't have any idea how I would deal with handling that exception.

It's really akin to something like:

Code:

struct s
{
        s() : p(new int()) { }
        ~s() { delete p; }

        int * p;
}

How would you deal with new throwing an exception?

There is no need to deal with it inside that class (in fact it's impossible to do it). You'd need to catch that exception wherever you make an object of that class.

Oh, and the big three rule applies here: If you make any one of: copy constructor, assignment operator, or destructor, you generally need all three.

Do you mean you're worried that 'new' will throw a bad_alloc exception? If that's the only possible exception to be thrown from the constructor, then there isn't much to do.

On the other hand, if your constructor throws another exception then you'll potentially have a memory leak (since the destructor won't be called).

I was actually just making a wierd solution around the fact that I can't install the dxsdk for real. Here's this little nested class I made to handle raii of a loaded library. It has a totally wierd ass design, it's just making up wierd solutions as I go along.

Code:

    struct f
    {
        HMODULE l;

        f( LPCTSTR lib, LPCSTR foo, LPCSTR bar ) throw(...)
        {
            if( l = ::LoadLibrary( lib ) )
            {
                XD3DXMatrixTranslation      = reinterpret_cast<D3DMATRIX * ( WINAPI * )( D3DMATRIX *, float, float, float)>         ( ::GetProcAddress( l, foo ) );
                XD3DXMatrixPerspectiveFovLH = reinterpret_cast<D3DMATRIX * ( WINAPI * )( D3DMATRIX *, float, float, float, float )>( ::GetProcAddress( l, bar ) );

                if( !XD3DXMatrixTranslation || !XD3DXMatrixPerspectiveFovLH )
                {
                    throw std::exception();
                }
            }
            else 
                throw std::exception();
        }
        ~f(void)
        {
            ::FreeLibrary( l );
        }
    } t;

Then I construct it something like

Code:

t( TEXT("H:\\d3dx9d_31.dll"), "D3DXMatrixTranslation", "D3DXMatrixPerspectiveFovLH" )

XD3DXMatrixTranslation and XD3DXMatrixPerspectiveFovLH are static class fields and usage of the class kind of depends on them. Oh wait, I just realized, the exception would fall through and the construction of the object would fail and I would be able to deal with things. I think that's good. Okay good.

>> I just realized, the exception would fall through and the construction of the object would fail and I would be able to deal with things.

That's how the handling of the exception works, but the issue is still about what to do with resources that have already been acquired when an exception is thrown inside the constructor. In your example, both times you throw std::exception(), ::FreeLibrary(l) will never be called. This might be ok, as in the second throw where apparently the call to LoadLibrary failed. However, the first throw inside the nested if looks like FreeLibrary should be called.

A simple RAII class should be created just for the call to LoadLibrary and FreeLibrary.

BTW, the t at the end of the struct declaration is probably not needed, and if you really did want an instance of your struct in that scope you should declare it separately to be more clear. As is it is a different instance than the one you construct with the constructor.

C++ should really allow aggregate and array initializers in the initializer list. And also, they should allow variable length template parameter lists. Okay, so this is safe?

Code:

    struct Library
    {
        HMODULE lib;

        Library( LPCTSTR szlib )
        {
            lib = ::LoadLibrary(szlib);

            if( !lib )
            {
                throw std::exception();
            }
        }

        operator HMODULE ()
        {
            return lib;
        }

        ~Library( void )
        {
            ::FreeLibrary( lib );
        }
    };

    template <typename T, typename U>

    struct Stuff
    {
        Library l;

        Stuff( LPCTSTR lib, LPCSTR foo, LPCSTR bar, T * t, U * u ) 
            : l(lib)
        {
            *t = reinterpret_cast
                <D3DMATRIX * ( WINAPI * )(D3DMATRIX *, float, float, float)>
                    ( ::GetProcAddress( l, foo ) );

            *u = reinterpret_cast
                <D3DMATRIX * ( WINAPI * )( D3DMATRIX *, float, float, float, float )>
                    ( ::GetProcAddress( l, bar ) );

            if( !*t || !*u ) 
                throw std::exception();
        }
    };

>> Okay, so this is safe?
As far as the automatic call to FreeLibrary happening whenever necessary, it looks safe to me. You could also initialize the Library struct's lib member in an intializer list for consistency, although it really doesn't matter.

>> C++ should really allow aggregate and array initializers in the initializer list.
I think they're trying to get that into the next standard.

> And also, they should allow variable length template parameter lists.

http://www.osl.iu.edu/~dgregor/cpp/v...templates.html


The code is safe, though you might want to throw something more expressive than a std::exception().

That is way cool. Thank you peoples.