I'm using C++11 and g++4.8 on Ubuntu Trusty.
Consider this snippet
class Parent {
public:
virtual ~Parent() = default;
virtual void f() = 0;
};
class Child: public Parent {
public:
void f(){}
};
Called using
{
Child o;
o.f();
}
{
Parent * o = new Child;
delete o;
}
{
Child * o = new Child;
delete o;
}
I use gcov to generate my code coverage report. It report that the destructor with symbol _ZN6ParentD0Ev
is never called, while _ZN6ParentD2Ev
is.
Answer Dual emission of constructor symbols and GNU GCC (g++): Why does it generate multiple dtors? reports that _ZN6ParentD0Ev
is the deleting constructor.
Is there any case where this "deleting destructor" is called on the Parent
class ?
Subsidiary question: if not, is there a way to get the gcov/lcov code coverage tool (used following answer of Detailed guide on using gcov with CMake/CDash?) ignore that symbol in its report ?
I think it's because you have the Child
object, not Parent
object.
{
Child o;
o.f();
} // 1
{
Parent * o = new Child;
delete o;
} // 2
{
Child * o = new Child;
delete o;
} // 3
In // 1
, o
is destroyed, and the complete object destructor of Child
is called. Since Child
inherits Parent
, it'll call the base object destructor, which is _ZN6ParentD2Ev
, of Parent
.
In // 2
, o
is dynamically allocated and deleted, and the deleting destructor of Child
is called. Then, it'll call the base object destructor of Parent
. In both, base object destructor is called.
// 3
is same. it's just equal to // 2
, except o
's type.
I've tested it on cygwin & g++ 4.8.3 & windows 7 x86 SP1. Here's my test code.
class Parent
{
public:
virtual ~Parent() { }
virtual void f() = 0;
};
class Child : public Parent
{
public:
void f() { }
};
int main()
{
{
Child o;
o.f();
}
{
Parent * o = new Child;
delete o;
}
{
Child * o = new Child;
delete o;
}
}
and compile & gcov option:
$ g++ -std=c++11 -fprofile-arcs -ftest-coverage -O0 test.cpp -o test
$ ./test
$ gcov -b -f test.cpp
Here's the result.
-: 0:Source:test.cpp
-: 0:Graph:test.gcno
-: 0:Data:test.gcda
-: 0:Runs:1
-: 0:Programs:1
function _ZN6ParentC2Ev called 2 returned 100% blocks executed 100%
2: 1:class Parent
-: 2:{
-: 3:public:
function _ZN6ParentD0Ev called 0 returned 0% blocks executed 0%
function _ZN6ParentD1Ev called 0 returned 0% blocks executed 0%
function _ZN6ParentD2Ev called 3 returned 100% blocks executed 75%
3: 4: virtual ~Parent() = default;
call 0 never executed
call 1 never executed
branch 2 never executed
branch 3 never executed
call 4 never executed
branch 5 taken 0% (fallthrough)
branch 6 taken 100%
call 7 never executed
-: 5: virtual void f() = 0;
-: 6:};
-: 7:
function _ZN5ChildD0Ev called 2 returned 100% blocks executed 100%
function _ZN5ChildD1Ev called 3 returned 100% blocks executed 75%
function _ZN5ChildC1Ev called 2 returned 100% blocks executed 100%
7: 8:class Child : public Parent
call 0 returned 100%
call 1 returned 100%
call 2 returned 100%
branch 3 taken 0% (fallthrough)
branch 4 taken 100%
call 5 never executed
call 6 returned 100%
-: 9:{
-: 10:public:
function _ZN5Child1fEv called 1 returned 100% blocks executed 100%
1: 11: void f() { }
-: 12:};
-: 13:
function main called 1 returned 100% blocks executed 100%
1: 14:int main()
-: 15:{
-: 16: {
1: 17: Child o;
1: 18: o.f();
call 0 returned 100%
call 1 returned 100%
-: 19: }
-: 20: {
1: 21: Parent * o = new Child;
call 0 returned 100%
call 1 returned 100%
1: 22: delete o;
branch 0 taken 100% (fallthrough)
branch 1 taken 0%
call 2 returned 100%
-: 23: }
-: 24: {
1: 25: Child * o = new Child;
call 0 returned 100%
call 1 returned 100%
1: 26: delete o;
branch 0 taken 100% (fallthrough)
branch 1 taken 0%
call 2 returned 100%
-: 27: }
1: 28:}
As you can see, _ZN6ParentD2Ev
, the base object destructur of Base
, is called while the others of Base
are not called.
However, _ZN5ChildD0Ev
, deleting destructor of Child
, is called twice and _ZN5ChildD1Ev
, complete object destructor of Child
, is called three times, since there's delete o;
and Child o;
.
But according my explanation, _ZN5ChildD0Ev
should be called twice and _ZN5ChildD1Ev
should be called once, shouldn't it? To figure out the reason, I did this:
$ objdump -d test > test.dmp
Result:
00403c88 <__ZN5ChildD0Ev>:
403c88: 55 push %ebp
403c89: 89 e5 mov %esp,%ebp
403c8b: 83 ec 18 sub $0x18,%esp
403c8e: a1 20 80 40 00 mov 0x408020,%eax
403c93: 8b 15 24 80 40 00 mov 0x408024,%edx
403c99: 83 c0 01 add $0x1,%eax
403c9c: 83 d2 00 adc $0x0,%edx
403c9f: a3 20 80 40 00 mov %eax,0x408020
403ca4: 89 15 24 80 40 00 mov %edx,0x408024
403caa: 8b 45 08 mov 0x8(%ebp),%eax
403cad: 89 04 24 mov %eax,(%esp)
403cb0: e8 47 00 00 00 call 403cfc <__ZN5ChildD1Ev>
403cb5: a1 28 80 40 00 mov 0x408028,%eax
403cba: 8b 15 2c 80 40 00 mov 0x40802c,%edx
403cc0: 83 c0 01 add $0x1,%eax
403cc3: 83 d2 00 adc $0x0,%edx
403cc6: a3 28 80 40 00 mov %eax,0x408028
403ccb: 89 15 2c 80 40 00 mov %edx,0x40802c
403cd1: 8b 45 08 mov 0x8(%ebp),%eax
403cd4: 89 04 24 mov %eax,(%esp)
403cd7: e8 a4 f9 ff ff call 403680 <___wrap__ZdlPv>
403cdc: a1 30 80 40 00 mov 0x408030,%eax
403ce1: 8b 15 34 80 40 00 mov 0x408034,%edx
403ce7: 83 c0 01 add $0x1,%eax
403cea: 83 d2 00 adc $0x0,%edx
403ced: a3 30 80 40 00 mov %eax,0x408030
403cf2: 89 15 34 80 40 00 mov %edx,0x408034
403cf8: c9 leave
403cf9: c3 ret
403cfa: 90 nop
403cfb: 90 nop
Yeah, since _ZN5ChildD0Ev
calls _ZN5ChildD1Ev
, _ZN5ChildD1Ev
was called three times. (1 + 2) I guess it's just GCC's implementation - for reduce duplication.