segfault with dynamic_cast

[Markus Dehmann]

I get a segfault in the following scenario (I am not really used to
pointers...):

I have a base class Animal and a derived class Dog. Now I construct
two Dogs and store pointers to them in a set<Dog*>. Beforing storing,
the Dog pointers are passed through a function taking Animal pointers
and dynamic_cast'in g them to Dog pointers.

Looks good to me, but it segfaults. Why? Here is my code:
#include <iostream>
#include <stdexcept>
#include <set>
using namespace std;

class Animal {
public:
Animal(const string& theName):name(t heName){}
virtual ~Animal(){}
virtual void speak(){
cout << "I do not speak." << endl;
};
const string& name;
const string& getName(){retur n name;}
};

class Dog : public Animal {
public:
Dog(const string& n):Animal(n){}
virtual ~Dog(){}
virtual void speak(){
cout << "Woof. I am " << getName() << endl;
};
};

class Zoo {
public:
set<Dog*> dogs;
void addAnimal(Anima l* a){
Dog* d;
if((d = dynamic_cast<Do g*>(a))){
dogs.insert(d);
}else throw runtime_error(" Unknown animal");
}
};

int main(){
Zoo zoo;
Dog d1("Rex");
Dog d2("Rox");
zoo.addAnimal(& d1);
zoo.addAnimal(& d2);
for(set<Dog*>:: iterator it = zoo.dogs.begin( ); it !=
zoo.dogs.end(); ++it){
(*it)->speak();
}
}

// (I know that zoo::dogs should be private, I just made it public
here to make code shorter)

Thanks for your comments!
Markus

[Ron Natalie]

Markus Dehmann wrote:

I get a segfault in the following scenario (I am not really used to
pointers...):

I have a base class Animal and a derived class Dog. Now I construct
two Dogs and store pointers to them in a set<Dog*>. Beforing storing,
the Dog pointers are passed through a function taking Animal pointers
and dynamic_cast'in g them to Dog pointers.

Looks good to me, but it segfaults. Why? Here is my code:

First make sure your compiler doesn't have some bogus mode to disable
RTTI.

Otherwise it looks ok to me. However, your code uses one of my peevish
constructs. I hate assignments as side effects of comparisons. This
is further aggravated by the fact that you don't initialize the variable
d. It would be better style to write it
Dog* d = dynmaic_cast<Do g*>(a);
if(d) ...

-Ron

Isn't Sirius the Dog*?

[Thomas Maier-Komor]

Markus Dehmann wrote:

I get a segfault in the following scenario (I am not really used to
pointers...):

I have a base class Animal and a derived class Dog. Now I construct
two Dogs and store pointers to them in a set<Dog*>. Beforing storing,
the Dog pointers are passed through a function taking Animal pointers
and dynamic_cast'in g them to Dog pointers.

Looks good to me, but it segfaults. Why? Here is my code:
#include <iostream>
#include <stdexcept>
#include <set>
using namespace std;

class Animal {
public:
Animal(const string& theName):name(t heName){}
virtual ~Animal(){}
virtual void speak(){
cout << "I do not speak." << endl;
};
const string& name;
const string& getName(){retur n name;}
};

class Dog : public Animal {
public:
Dog(const string& n):Animal(n){}
virtual ~Dog(){}
virtual void speak(){
cout << "Woof. I am " << getName() << endl;
};
};

class Zoo {
public:
set<Dog*> dogs;
void addAnimal(Anima l* a){
Dog* d;
if((d = dynamic_cast<Do g*>(a))){
dogs.insert(d);
}else throw runtime_error(" Unknown animal");
}
};

int main(){
Zoo zoo;
Dog d1("Rex");
Dog d2("Rox");
zoo.addAnimal(& d1);
zoo.addAnimal(& d2);
for(set<Dog*>:: iterator it = zoo.dogs.begin( ); it !=
zoo.dogs.end(); ++it){
(*it)->speak();
}
}

// (I know that zoo::dogs should be private, I just made it public
here to make code shorter)

Thanks for your comments!
Markus

you pass "Rex" and "Rox" to Dog::Dog(const string &) which
create a temporary string. g++ does not seem to find out
that this temporary is needed until d1 and d2 go out of scope.
I would say this is a bug in g++, as Sun's Studio compiler does
what you would expect it to do...

Tom

[Shezan Baig]

Markus Dehmann wrote:

I get a segfault in the following scenario (I am not really used to
pointers...):

I have a base class Animal and a derived class Dog. Now I construct
two Dogs and store pointers to them in a set<Dog*>. Beforing storing, the Dog pointers are passed through a function taking Animal pointers
and dynamic_cast'in g them to Dog pointers.

Looks good to me, but it segfaults. Why? Here is my code:
#include <iostream>
#include <stdexcept>
#include <set>
using namespace std;

class Animal {
public:
Animal(const string& theName):name(t heName){}
virtual ~Animal(){}
virtual void speak(){
cout << "I do not speak." << endl;
};
const string& name;

Change this to:

const string name;

(i.e., hold a copy, not a reference). After this, you should be fine.

Hope this helps,
-shez-

const string& getName(){retur n name;}
};

class Dog : public Animal {
public:
Dog(const string& n):Animal(n){}
virtual ~Dog(){}
virtual void speak(){
cout << "Woof. I am " << getName() << endl;
};
};

class Zoo {
public:
set<Dog*> dogs;
void addAnimal(Anima l* a){
Dog* d;
if((d = dynamic_cast<Do g*>(a))){
dogs.insert(d);
}else throw runtime_error(" Unknown animal");
}
};

int main(){
Zoo zoo;
Dog d1("Rex");
Dog d2("Rox");
zoo.addAnimal(& d1);
zoo.addAnimal(& d2);
for(set<Dog*>:: iterator it = zoo.dogs.begin( ); it !=
zoo.dogs.end(); ++it){
(*it)->speak();
}
}

// (I know that zoo::dogs should be private, I just made it public
here to make code shorter)

Thanks for your comments!
Markus

[Thomas Maier-Komor]

Thomas Maier-Komor wrote:

you pass "Rex" and "Rox" to Dog::Dog(const string &) which
create a temporary string. g++ does not seem to find out
that this temporary is needed until d1 and d2 go out of scope.
I would say this is a bug in g++, as Sun's Studio compiler does
what you would expect it to do...

Tom

OK, I looked into the standard and it says that the temporary
may be destructed after the execution of the ctor. So g++ does
the right thing. Simply create a string variable and pass it
to the Dog constructor and you will be fine. As an alternative
you can change Animal::name to a normal variable (not a reference)
and everything will be OK, too. This has the advantage that
you don't have to bother around with the lifetime of the string
you pass to the constructor...

Tom

[puzzlecracker]

Change this to:

const string name;

(i.e., hold a copy, not a reference). After this, you should be fine.
Hope this helps,
-shez-

Can you really do that, copy the temp? let's say:

char * p="temp"; // changing anything in p is an error
char arr[5];
strpcpy(p,arr);

what is the storaga nature of above temp?

[Ron Natalie]

puzzlecracker wrote:

Change this to:

const string name;

(i.e., hold a copy, not a reference). After this, you should be
fine.

Hope this helps,
-shez-

Can you really do that, copy the temp? let's say:

Sure, why not. The temp stays around long enough for name to get
initialized. String doesn't care what the char* it's initialized is
as long as it:
1. Is null terminated (or you pass the length)
2. It's valid while the consturctor is running.
char * p="temp"; // changing anything in p is an error
char arr[5];
strpcpy(p,arr);

Why would you want to do this?

[Karl Heinz Buchegger]

puzzlecracker wrote:

Change this to:

const string name;

(i.e., hold a copy, not a reference). After this, you should be fine.

Hope this helps,
-shez-

Can you really do that, copy the temp?

Sure, why not?
The temporary gets destroyed long after the copy has been made.
let's say:

char * p="temp"; // changing anything in p is an error
char arr[5];
strpcpy(p,arr);
that's completely different.
No temporary is involved.

what is the storaga nature of above temp?

There is no temporary.
If you are referring to "temp" in the above. This
is a string literal which gets created at the beginning
of your program and it lifes on until the program shuts
down. A completely different case.

--
Karl Heinz Buchegger
kb******@gascad .at