Why does zeroing out a structure cause lost of access to virtual base class members?

Question

 struct base
 {
     int num;
     char ch;
     float fl;
 };

 struct ss : virtual base
 {
 };

 void main()
 {
     ss sa;

     sa.num = 100;

     cout << sa.num << endl;

     memset(&sa, 0, sizeof(sa));        // set all members to zero

     cout <<  sa.num << endl;           // Access violation reading location 0x00000004.


 }

I can access members from a struct that is derived from a virtual base. When zeroing out the struct it also causes the lost of access to the members. Does this mean we would have to clear members individually instead of using memset or ZeroMemory?

You have undefined behaviour since `ss` is not a POD type. You're making zeroing it harder than it has to be: `ss sa{};`. And `void main` is bad; use `int main`. — ghostofstandardspast, Jun 16 '14 at 18:06
@ghostofstandardspast What if you want to clear a struct after using it? — user3613229, Jun 16 '14 at 18:09
@Mooing Duck Too much overhead. Why create a new object, initialize to zero and then copy it to the existing object? — user3613229, Jun 16 '14 at 18:14
@user3613229, I think the overhead is much smaller than you imagine. By the way, mine also creates a new object and assigns that to `sa`. I really can't see it differing much in overhead from `ZeroMemory` or something, though. — ghostofstandardspast, Jun 16 '14 at 18:14
@ghostofstandardspast So it's not possible to clear the object in the constructor? Why does this language have so much limitations. — user3613229, Jun 16 '14 at 18:18
@user3613229 What about to provide a method `ss::clear()`, if you're worried about the copy/move _'overhead'_?? — πάντα ῥεῖ, Jun 16 '14 at 18:21
@user3613229: I can't imagine why you think it's not possible to clear the object in the constructor, it's totally possible (and easy) to do that, but that's not what you asked us. — Mooing Duck, Jun 16 '14 at 18:23
@Mysticial Someone should invent a new language simliar to C++ but without classess. If such a language does exist I would be so happy. — user3613229, Jun 16 '14 at 18:27
Or, you could just refrain from using classes. @Mysticial: There's a lot more than classes which C++ has that C doesn't. — Benjamin Lindley, Jun 16 '14 at 18:29
Just don't do `memset`. It's a sledgehammer. We do not do that any more. There are little implementation details here that you cannot see, and you're bulldozing right over them with no regard whatsoever. — Lightness Races in Orbit, Jun 16 '14 at 18:32
@DeadMG: No, neither do I. The _code_ needs downvoting (and so, possibly, does the author), but the question is fairly reasonable. — Lightness Races in Orbit, Jun 16 '14 at 18:33
@BenjaminLindley *shhhhh... I think your sarcasm detector is failing.* :) — Mysticial, Jun 16 '14 at 18:33
@user3613229 Try template meta programming. Fun language I assure you. — Shoe, Jun 16 '14 at 18:34
@ghostofstandardspast That's the point. With template meta programming you get almost `0` runtime overhead. — Shoe, Jun 16 '14 at 18:38
@Jefffrey, [Not you, too.](http://stackoverflow.com/questions/24249598/why-does-zeroing-out-a-structure-cause-lost-of-access-to-virtual-base-class-memb#comment37457666_24249598) — ghostofstandardspast, Jun 16 '14 at 18:39
@LightnessRacesinOrbit, I don't even know anymore. Mine blew up and now I'm just hopelessly bluffing. — ghostofstandardspast, Jun 16 '14 at 18:45
Also see: http://stackoverflow.com/a/38103250/3223828 (particularly for non C++11 solutions). — Stuart Gillibrand, Jun 29 '16 at 18:19

score 5 · Answer 1 · answered Jun 16 '14 at 18:13

5

Short answer: Undefined behaviour. Full stop.

Long answer: base is a virtual base of ss. Thus, it is unknown at compile time at which offset from the beginning of ss the parent base is located. This offset information must be retrieved at runtime. The information is stored in the vtable of the object.

So in order to know the offset of fields in base, the vtable pointer (which resides inside the object) has to be considered. You just nulled this vtable pointer by nulling the whole struct. So when it is dereferenced to look up the offset, everything is about to blow up. The offset of base seems to be at offset 4 from the beginning of the vtable, this is why you get an access violation at address 4.

answered Jun 16 '14 at 18:13

gexicide

38,535
21
92
152

So... `memset((char*)&sa + sizeof(void*), 0, sizeof(sa) - sizeof(void*))`? ;) – fredoverflow Jun 16 '14 at 18:40
@FredOverflow Why not simply `sa.num = 0; sa.ch = '\0'; sa.fl = 0.0; `? – chux - Reinstate Monica Jun 16 '14 at 19:00
@FredOverflow: That assumes that the vptr is based at 0. – Puppy Jun 16 '14 at 19:12

Why does zeroing out a structure cause lost of access to virtual base class members?

1 Answers1