[Bug c++/22228] New: Arguments shift when print/call (sometimes causing gdb crash)

["sillyduckyang at gmail dot com" <sourceware-bugzilla at sourceware dot org>]

https://sourceware.org/bugzilla/show_bug.cgi?id=22228

            Bug ID: 22228
           Summary: Arguments shift when print/call (sometimes causing gdb
                    crash)
           Product: gdb
           Version: 8.0.1
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: c++
          Assignee: unassigned at sourceware dot org
          Reporter: sillyduckyang at gmail dot com
  Target Milestone: ---

Description:
------------
If a function is returning a inheritance class object,
the print/call of that function will have an argument shift.


Testcase:
---------
#include <iostream>
using namespace std;
class Dad {
    public:
        Dad(int i): _value(i) {}
    private:
        int _value ;
};
class Son : public Dad {
    public:
        Son() : Dad(777) {}
};
class B{
    public:
        Son f(int a, int b, B* c){
            cout << this << " " << a << " " << b << " " << c << endl;
            cout << (this == c) << endl;
            return Son();
        }
};
int main(){
    B* p = new B();
    p->f(1,2,p);
    return 0;
}


How to reproduce:
-----------------
Compile the above mentioned testcase 

(gdb) b main
Breakpoint 1 at 0x4009ee: file test.cpp, line 27.
(gdb) r
Starting program: /home/SillyDuck/bad 

Breakpoint 1, main () at test.cpp:27
27          B* p = new B();
(gdb) n
28          p->f(1,2,p);
(gdb) p p->f(1,2,p)
0x7fffffffda00 6376480 1 0x2
0
$1 = {<Dad> = {_value = 0}, <No data fields>}


Environment:
------------
GNU gdb (GDB) 8.0.1
gcc version 5.4.0 20160609 (Ubuntu 5.4.0-6ubuntu1~16.04.4) 

Also appears in the gcc 4.8.3 with gdb 7.12 in CentOS 6.6
(In fact it appears in most linux system)


Discussion:
-----------

After tracing the src,
I think the root problem may be in the
gnu-v3-api.c --> gnuv3_pass_by_reference

the assembly of test.cpp did not has a pointer passed as the first argument for
the return value.
But gdb is somehow inconsist with gcc's behavior...

If the function is calling some other function with incorrect "this" pointer,
it sometimes also crashes the gdb...

00000000004009e6 <main>:
  4009e6:       55                      push   %rbp
  4009e7:       48 89 e5                mov    %rsp,%rbp
  4009ea:       48 83 ec 10             sub    $0x10,%rsp
  4009ee:       bf 01 00 00 00          mov    $0x1,%edi
  4009f3:       e8 d8 fe ff ff          callq  4008d0 <_Znwm@plt>
  4009f8:       48 89 45 f8             mov    %rax,-0x8(%rbp)
  4009fc:       48 8b 55 f8             mov    -0x8(%rbp),%rdx
  400a00:       48 8b 45 f8             mov    -0x8(%rbp),%rax
  400a04:       48 89 d1                mov    %rdx,%rcx
  400a07:       ba 02 00 00 00          mov    $0x2,%edx
  400a0c:       be 01 00 00 00          mov    $0x1,%esi
  400a11:       48 89 c7                mov    %rax,%rdi
  400a14:       e8 cf 00 00 00          callq  400ae8 <_ZN1B1fEiiPS_>
  400a19:       b8 00 00 00 00          mov    $0x0,%eax
  400a1e:       c9                      leaveq 
  400a1f:       c3                      retq   

Interesting thing is that if we add a copy constructor in Dad, this problem is
gone...

-- 
You are receiving this mail because:
You are on the CC list for the bug.