[Bug fortran/41951] New: [OOP] ICE in gfc_match_varspec, at fortran/primary.c:1815

[Bug fortran/41951] New: [OOP] ICE in gfc_match_varspec, at fortran/primary.c:1815

[burnus at gcc dot gnu dot org]

The following program is an extension to the one in PR 41539. It finally
requires to many features that ifort 11.1 does not like it - nor the dated NAG
f95 5.1 (current is 5.2). And while gfortran should have everything needed, it
still does not work.

I have no idea whether the code is correct - probably not - but an ICE is
definitely wrong. (For other issues see PR 41539.)

$ gfortran sort.f90
f951: internal compiler error: in gfc_match_varspec, at fortran/primary.c:1815
Please submit a full bug report


-- 
           Summary: [OOP] ICE in gfc_match_varspec, at
                    fortran/primary.c:1815
           Product: gcc
           Version: 4.5.0
            Status: UNCONFIRMED
          Keywords: ice-on-valid-code
          Severity: normal
          Priority: P3
         Component: fortran
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: burnus at gcc dot gnu dot org
 BugsThisDependsOn: 41539


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41951

[Bug fortran/41951] [OOP] ICE in gfc_match_varspec, at fortran/primary.c:1815

[burnus at gcc dot gnu dot org]

------- Comment #1 from burnus at gcc dot gnu dot org  2009-11-05 17:01 -------
Created an attachment (id=18974)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=18974&action=view)
Test case - see marked line for the ICE


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41951

[Bug fortran/41951] [OOP] ICE in gfc_match_varspec, at fortran/primary.c:1815

[janus at gcc dot gnu dot org]

------- Comment #2 from janus at gcc dot gnu dot org  2009-11-06 10:35 -------
Created an attachment (id=18978)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=18978&action=view)
legalized test case

The test case in comment #1 is clearly illegal (and rejected by NAG 5.2). This
version should be valid.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41951

[Bug fortran/41951] [OOP] ICE in gfc_match_varspec, at fortran/primary.c:1815

[tkoenig at gcc dot gnu dot org]

------- Comment #3 from tkoenig at gcc dot gnu dot org  2009-12-05 22:14 -------
Reduced test case, from c2.f90:

module m_sort
  implicit none
  type, abstract :: sort_t
  contains
    procedure(gt_cmp), deferred :: gt_cmp
  end type sort_t
contains
  subroutine bsort(a)
    class(sort_t), intent(inout) :: a(:)
    if (a(1)%gt_cmp(a(2))) then ! ICE
    end if
  end subroutine bsort
end module m_sort


-- 

tkoenig at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|0                           |1
   Last reconfirmed|0000-00-00 00:00:00         |2009-12-05 22:14:52
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41951

[Bug fortran/41951] [OOP] ICE in gfc_match_varspec, at fortran/primary.c:1815

[burnus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41951

--- Comment #9 from Tobias Burnus <burnus at gcc dot gnu.org> 2011-12-16 14:30:15 UTC ---
(In reply to comment #8)
> I failed to nail it in the standard, but I am sure that "12.4.3.4.5
> Restrictions on generic declarations" somehow must apply.

Thus I have now asked at
http://j3-fortran.org/pipermail/j3/2011-December/004946.html

[Bug fortran/41951] [OOP] ICE in gfc_match_varspec, at fortran/primary.c:1815

[burnus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41951

--- Comment #8 from Tobias Burnus <burnus at gcc dot gnu.org> 2011-12-13 11:40:51 UTC ---
(In reply to comment #6)
> Given that NAG 5.2 compiles it and as "assign" is only called via
> "assignment(=)", there must be some issue with the assignment resolution.

Actually, I wonder whether the code is valid. One has:

  type, abstract :: sort_t
  contains
    generic :: assignment(=) => assign
    ...

  interface assignment(=)
    procedure assign
  end interface

Thus, one defines twice a generic name with the same (= identical) interface,
once as TBP and once as external procedure. Thus, when one encounters:
    tmp = a(i)
the compiler has the choice between "assign_" and "tmp->_vtab->assign".

I failed to nail it in the standard, but I am sure that "12.4.3.4.5
Restrictions on generic declarations" somehow must apply.

[Bug fortran/41951] [OOP] ICE in gfc_match_varspec, at fortran/primary.c:1815

[burnus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41951

--- Comment #7 from Tobias Burnus <burnus at gcc dot gnu.org> 2011-12-12 13:14:24 UTC ---
(In reply to comment #6)
> However, while the example of comment 2 (attachment 18978 [details]) compiles,
> it fails to link with:   undefined reference to `assign_'

The problem is in "subroutine bsort(a,tmp)": The line
    tmp = a(i)
should call tmp->_vtr->assign. However, it calls a simple external "assign"
procedure. If one uses:
    call tmp%assign(a(i))
one finds the correct dump, namely:
    tmp->_vptr->assign ((struct __class_m_sort_Sort_t *) tmp, &class.2);


The "assign" is found when resolve_ordinary_assign calls gfc_extend_assign,
which runs:
  3452    sym = gfc_search_interface (ns->op[INTRINSIC_ASSIGN], 1, &actual);

That returns the INTERFACE instead of the type bound procedure:

(gdb) p sym->name
$9 = 0x2aaaacde8fd0 "assign"
(gdb) p sym->attr.external
$11 = 1

If one changes the interface to an abstract one, one gets 
   ERROR: ABSTRACT INTERFACE 'assign' must not be referenced


Using "assign" instead of the type-bound "assign" is definitely a bug as no 
"INTERFACE ASSIGNMENT(=)" exists - there is only the type-based "generic ::
assignment(=) => assign".

[Bug fortran/41951] [OOP] ICE in gfc_match_varspec, at fortran/primary.c:1815

[burnus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41951

--- Comment #6 from Tobias Burnus <burnus at gcc dot gnu.org> 2011-12-12 08:03:41 UTC ---
STATUS:
With the commit for PR 41539, there is no longer an ICE.

However, while the example of comment 2 (attachment 18978) compiles, it fails
to link with:   undefined reference to `assign_'

Given that NAG 5.2 compiles it and as "assign" is only called via
"assignment(=)", there must be some issue with the assignment resolution.

[Bug fortran/41951] [OOP] ICE in gfc_match_varspec, at fortran/primary.c:1815

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41951

--- Comment #5 from Dominique d'Humieres <dominiq at lps dot ens.fr> 2011-01-23 10:57:05 UTC ---
> Another test case, from ...

It is pr47399.

[Bug fortran/41951] [OOP] ICE in gfc_match_varspec, at fortran/primary.c:1815

[tkoenig at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41951

Thomas Koenig <tkoenig at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |tkoenig at gcc dot gnu.org

--- Comment #4 from Thomas Koenig <tkoenig at gcc dot gnu.org> 2011-01-23 10:10:33 UTC ---
Another test case, from

http://groups.google.com/group/comp.lang.fortran/msg/e6a865eda59e86db?hl=de


module mytypes
   implicit none
   private
   public :: mytype, get_i

   integer, save :: i_priv = 13

   type :: mytype
      integer :: dummy
    contains
      procedure, nopass :: i => get_i
   end type mytype

 contains

   pure function get_i () result (i)
     integer :: i
     i = i_priv
   end function get_i

end module mytypes


program test
   use mytypes
   implicit none

   type(mytype) :: a
   type(mytype), parameter :: a_const = mytype (0)

!   integer, dimension (get_i()) :: x            ! #1
!   integer, dimension (a%i()) :: x              ! #2
   integer, dimension (a_const%i()) :: x        ! #3

   print *, size (x)

end program test