How can one branch of class tree have all its Zero initialized members be rubbish?

Question

(not rubbish only if zero initialized as static scope member.) and it works as expected on GCC!(((

So for code like:

#include <iostream>
#include <boost/shared_ptr.hpp>

namespace SP
{
    // enums
    enum EnumMessage {
        EnumMessage_EventBase = 0,
        EnumMessage_EventServiceBase = 1,
        EnumMessage_OperationBase = 2
    };

    class Message;

    // define proxy class
    class Message  {
    public:
        EnumMessage _MessageCode;
    };

    class Parent : public Message {
    public:
        int _ServiceId;
        int _CallbackId;
    };

    class Child : public Parent {
    public:
        std::string _Debug;
    };

    class AnotherChild : public Parent {
    public:
        int _UserId;
    };
}

using namespace SP;

int main() {
    //OK
    static Child staticScopeChild = Child();
    boost::shared_ptr<Parent> ptrParent(new Parent());
    boost::shared_ptr<AnotherChild> ptrChild2(new AnotherChild());

    //Bad
    Child scopeChild = Child();
    boost::shared_ptr<Child> ptrChild(new Child());


    std::cout << "static " << staticScopeChild._MessageCode 
        << std::endl << "vs scope " << scopeChild._MessageCode 
        << std::endl << "vs pointer " << ptrChild->_MessageCode 
        << std::endl << "vs parent class pointer: " << ptrParent->_MessageCode 
        << std::endl << "vs another parent child: " << ptrChild2->_MessageCode <<std::endl;
    std::cin.get();
    return 0;
}

where all classes are generally PODs (ints, enums) I get next output:

static 0
vs scope -858993460
vs pointer -842150451
vs parent class pointer: 0
vs another parent child: 0

while I expected all to be 0!

Why can such thing happen?

And I have such bad zero initialization for all classes in that branch up that child... While all goes Ok in all other branches which seem not diferent from that broken one.((( — myWallJSON, Jan 24 '13 at 12:06
It sounds like the constructor of `Child` doesn't initialise `$MessageCode`; so static objects will be zero-initialised and others will have indeterminate values. But without seeing the constructor, I can only guess. — Mike Seymour, Jan 24 '13 at 12:10
(By the way, it's not portable to use `$`, or any non-alphanumeric character except `_`, in an identifier.) — Mike Seymour, Jan 24 '13 at 12:12
@MikeSeymour there is no Child or Parent or AnotherChild constructor - we rely on ones VS generates for us. also seems like It is possible to use `$` - our code compiles and even works to some extent...) — myWallJSON, Jan 24 '13 at 12:13
@myWallJSON: Indeed, some compilers accept "other implementation-defined characters", and if you don't need portability then that's fine. I was just pointing out that it wasn't portable, in case that's an issue for you. — Mike Seymour, Jan 24 '13 at 12:15
@myWallJSON: Perhaps you could post the definition of `Child`, perhaps removing as much as you can while still exhibiting the problem. *If* the classes are POD, then `Child()` should value-initialise all the members to zero; but without seeing the code, we can only guess why that's not happening. — Mike Seymour, Jan 24 '13 at 12:18
I wouldn't expect enums to be 0. Maybe it's valid C++, and maybe I knew this long ago, but from all the weirdness of VC++ you should start forgetting "tricks" and just go and stuff the compiler's mouth with code of YOUR choice... *angry rant* — ActiveTrayPrntrTagDataStrDrvr, Jan 24 '13 at 12:39
Remember that enums are compile-time values, and they start at 0 by default. — Qix - MONICA WAS MISTREATED, Jan 24 '13 at 12:47
@ActiveTrayPrntrTagDataStrDrvr: enums, like any other scalar type, should be initialised to zero (converted to the enumeration type) by value-initialisation. — Mike Seymour, Jan 24 '13 at 13:06
Look at the answers... it's the "VC++ weirdness" I'm talking about. If you let VC2010 to handle anything that you forgot to specify, get ready to stumble hard. I don't know about VC2012. I've been in this boat for 8 years. — ActiveTrayPrntrTagDataStrDrvr, Jan 24 '13 at 13:38

Mike Seymour · Accepted Answer · 2013-01-24T14:52:36.887

4

It appears to be a compiler bug: according to this report members of base classes are not zero-initialised during value-initialisation of the derived class.

As far as I can see, there is nothing wrong with your code, and all class members should be zero-initialised on a conforming C++03 or C++11 compiler.

I guess your options are:

make the classes more POD-like by avoiding inheritance; or
add default constructors to any base classes to explicitly zero-initialise all members; or
use a less broken compiler.

edited Jan 24 '13 at 14:52

answered Jan 24 '13 at 13:11

Mike Seymour

249,747
28
448
644

Avoiding inheritance won't make the classes POD, as `Child` contains a member of type `std::string` that is not POD. I like the last suggestion though :) -- +1 as the answer is correct in that the members should be zero-initialized before the default constructor is called. – David Rodríguez - dribeas Jan 24 '13 at 14:50
@DavidRodríguez-dribeas: That's true. But without inheritance it might be POD-like enough for value-initialisation to work, though. – Mike Seymour Jan 24 '13 at 14:52
Please submit a bug report and lets see official answer. – myWallJSON Feb 04 '13 at 20:48
@myWallJSON: Someone already has - see the [link](https://connect.microsoft.com/VisualStudio/feedback/details/746973/incorrect-c-11-value-initialization-for-type-with-implicitly-declared-but-non-trivial-default-constructor) I included in the first sentence of the answer. – Mike Seymour Feb 05 '13 at 11:39

How can one branch of class tree have all its Zero initialized members be rubbish?

1 Answers1

Linked