[Bug fortran/61281] New: Memory corruption on assigning a polymorphic variable to a non-polymorphic one

[Bug fortran/61281] New: Memory corruption on assigning a polymorphic variable to a non-polymorphic one

[rouson at stanford dot edu]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61281

            Bug ID: 61281
           Summary: Memory corruption on assigning a polymorphic variable
                    to a non-polymorphic one
           Product: gcc
           Version: 4.9.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: fortran
          Assignee: unassigned at gcc dot gnu.org
          Reporter: rouson at stanford dot edu

The code below demonstrates a bug isolated from code from my collaborator Rob
Rosenberg of the Naval Research Laboratory.  The code exhibits an apparent
memory corruption issue with gfortran 4.9 dated 20140416.  Gfortran 4.8.2
rejects the code with the message below.

Damian


$ cat rob.f90 
 type employees
   character(3) :: name="rob"
 end type 
 type companies
   type(employees) :: list
 end type 
 type(companies) :: nrl
 print *,nrl%list
 call map
 print *,nrl%list
contains
 type (employees) function give_raise(e) 
   class(employees) :: e ! Switch to "type" to hide bug
   give_raise=e
 end function 
 subroutine map() 
   nrl%list=give_raise(nrl%list)
 end subroutine
end 
$ gfortran rob.f90 
$ ./a.out
rob
8??
$ gfortran --version
GNU Fortran (MacPorts gcc49 4.9-20140416_0) 4.9.0 20140416 (prerelease)
Copyright (C) 2014 Free Software Foundation, Inc.
$ sudo port select --set gcc mp-gcc48
Selecting 'mp-gcc48' for 'gcc' succeeded. 'mp-gcc48' is now active.
$ gfortran rob.f90 
rob.f90:14.15:

    give_raise=e
               1
Error: Can't convert CLASS(employees) to TYPE(employees) at (1)
localhost:Desktop rouson$ gfortran --version
GNU Fortran (MacPorts gcc48 4.8.2_0) 4.8.2

[Bug fortran/61281] Memory corruption on assigning a polymorphic variable to a non-polymorphic one

[dominiq at lps dot ens.fr]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61281

Dominique d'Humieres <dominiq at lps dot ens.fr> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2014-05-23
     Ever confirmed|0                           |1

--- Comment #1 from Dominique d'Humieres <dominiq at lps dot ens.fr> ---
Confirmed. The error disappeared between r201266 (2013-07-26) and r201631
(2013-08-09).

[Bug fortran/61281] Memory corruption on assigning a polymorphic variable to a non-polymorphic one

[rouson at stanford dot edu]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61281

--- Comment #2 from Damian Rouson <rouson at stanford dot edu> ---
I assume you're saying it disappeared and then reappeared and you're confirming
the problem exists on the current trunk.  Correct?

[Bug fortran/61281] Memory corruption on assigning a polymorphic variable to a non-polymorphic one

[dominiq at lps dot ens.fr]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61281

--- Comment #3 from Dominique d'Humieres <dominiq at lps dot ens.fr> ---
> I assume you're saying it disappeared and then reappeared and you're confirming
> the  problem exists on the current trunk.  Correct?

Sorry for the confusion. I meant
(1) I confirm the memory corruption after r201631,
(2) at least up to r201266, but not after r201630 (no archive in the gap),
gfortran used to emit the error

Error: Can't convert CLASS(employees) to TYPE(employees) at (1)

[Bug fortran/61281] Memory corruption on assigning a polymorphic variable to a non-polymorphic one

[dominiq at lps dot ens.fr]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61281

Dominique d'Humieres <dominiq at lps dot ens.fr> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |burnus at gcc dot gnu.org,
                   |                            |janus at gcc dot gnu.org

--- Comment #4 from Dominique d'Humieres <dominiq at lps dot ens.fr> ---
Valgrind nor -fsanitize=address don't report any error.

The "memory corruption" is not limited to character variable and it occurs
already in map, i.e., give_raise(nrl%list) does not return 'rob'.

The error "Can't convert CLASS(employees) to TYPE(employees) at (1)" may have
been removed by r201328.

The original dump for type is:

map ()
{
  nrl.list = give_raise (&nrl.list);
}


give_raise (struct employees & restrict e)
{
  struct employees __result_give_raise;

  {
    struct employees employees.0;

    (MEM[(c_char * {ref-all})&employees.0] = MEM[(c_char * {ref-all})"rob"];,
(void *) &employees.0.name;);
    __result_give_raise = employees.0;
  }
  __result_give_raise = *e;
  return __result_give_raise;
}


MAIN__ ()
{
  static struct companies nrl = {.list={.name="rob"}};
  static struct employees give_raise (struct employees & restrict);
  static void map (void);

and with class it is:

__copy_MAIN___Employees (struct employees & restrict src, struct employees &
restrict dst)
{
  *dst = *src;
}


map ()
{
  {
    struct __class_MAIN___Employees_t class.0;

    class.0._vptr = (struct __vtype_MAIN___Employees * {ref-all})
&__vtab_MAIN___Employees;
    class.0._data = &nrl.list;
    nrl.list = give_raise (&class.0);
  }
}


give_raise (struct __class_MAIN___Employees_t & restrict e)
{
  struct employees __result_give_raise;

  {
    struct employees employees.1;

    (MEM[(c_char * {ref-all})&employees.1] = MEM[(c_char * {ref-all})"rob"];,
(void *) &employees.1.name;);
    __result_give_raise = employees.1;
  }
  __result_give_raise = VIEW_CONVERT_EXPR<struct employees>(*e);
  return __result_give_raise;
}


MAIN__ ()
{
  static struct __vtype_MAIN___Employees __vtab_MAIN___Employees =
{._hash=65382630, ._size=3, ._extends=0B,
._def_init=&__def_init_MAIN___Employees, ._copy=__copy_MAIN___Employees,
._final=0B};
  static struct employees __def_init_MAIN___Employees = {.name="rob"};
  static struct companies nrl = {.list={.name="rob"}};
  static struct employees give_raise (struct __class_MAIN___Employees_t &
restrict);
  static void map (void);
  static void __copy_MAIN___Employees (struct employees & restrict, struct
employees & restrict);
...

[Bug fortran/61281] Memory corruption on assigning a polymorphic variable to a non-polymorphic one

[janus at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61281

--- Comment #5 from janus at gcc dot gnu.org ---
I can confirm the misbehavior with 4.9.1, but it seems that the problem is gone
on current trunk:

$ ./a.out
 rob
 rob

gcc version 5.0.0 20141213 (experimental) [trunk revision 218705] (GCC)

[Bug fortran/61281] Memory corruption on assigning a polymorphic variable to a non-polymorphic one

[dominiq at lps dot ens.fr]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61281

--- Comment #6 from Dominique d'Humieres <dominiq at lps dot ens.fr> ---
> I can confirm the misbehavior with 4.9.1, but it seems that the problem
> is gone on current trunk:

Confirmed. Likely fixed by r 216427 (pr63553).

[Bug fortran/61281] Memory corruption on assigning a polymorphic variable to a non-polymorphic one

[janus at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61281

janus at gcc dot gnu.org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |wrong-code
             Status|NEW                         |RESOLVED
      Known to work|                            |5.0
         Resolution|---                         |FIXED
      Known to fail|                            |4.9.0

--- Comment #7 from janus at gcc dot gnu.org ---
(In reply to Dominique d'Humieres from comment #6)
> Confirmed. Likely fixed by r 216427 (pr63553).

Probably. Closing as fixed.