[Bug fortran/103283] New: ICE in gfc_trans_array_constructor_subarray, at fortran/trans-array.c:1972

[Bug fortran/103283] New: ICE in gfc_trans_array_constructor_subarray, at fortran/trans-array.c:1972

[gscfq@t-online.de]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103283

            Bug ID: 103283
           Summary: ICE in gfc_trans_array_constructor_subarray, at
                    fortran/trans-array.c:1972
           Product: gcc
           Version: 12.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: fortran
          Assignee: unassigned at gcc dot gnu.org
          Reporter: gscfq@t-online.de
  Target Milestone: ---

Affects versions down to at least r5 :


$ cat z1.f90
program p
   type t
      integer :: a(1) = 2
   end type
   type(t), parameter :: x(1) = [t(3)]
   integer :: y(1)
   y = [(x(1)%a)]
   print *, y(1)
end


$ cat z2.f90
program p
   type t
      integer :: a(1) = 2
   end type
   type(t), parameter :: x(1) = [t(3)]
   associate (y => [(x(1)%a)])
      print *, y(1)
   end associate
end


$ gfortran-12-20211114 -c z1.f90
z1.f90:7:17:

    7 |    y = [(x(1)%a)]
      |                 1
internal compiler error: in gfc_trans_array_constructor_subarray, at
fortran/trans-array.c:1972
0x85c45d gfc_trans_array_constructor_subarray
        ../../gcc/fortran/trans-array.c:1972
0x85c45d gfc_trans_array_constructor_value
        ../../gcc/fortran/trans-array.c:2100
0x85cdaa trans_array_constructor
        ../../gcc/fortran/trans-array.c:2900
0x85cdaa gfc_add_loop_ss_code
        ../../gcc/fortran/trans-array.c:3180
0x85d475 gfc_conv_loop_setup(gfc_loopinfo*, locus*)
        ../../gcc/fortran/trans-array.c:5430
0x890fad gfc_trans_assignment_1
        ../../gcc/fortran/trans-expr.c:11642
0x84ed67 trans_code
        ../../gcc/fortran/trans.c:1916
0x87801e gfc_generate_function_code(gfc_namespace*)
        ../../gcc/fortran/trans-decl.c:7644
0x7fb32e translate_all_program_units
        ../../gcc/fortran/parse.c:6638
0x7fb32e gfc_parse_file()
        ../../gcc/fortran/parse.c:6925
0x84806f gfc_be_parse_file
        ../../gcc/fortran/f95-lang.c:216

[Bug fortran/103283] ICE in gfc_trans_array_constructor_subarray, at fortran/trans-array.c:1972

[gscfq@t-online.de]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103283

G. Steinmetz <gscfq@t-online.de> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |ice-on-valid-code

--- Comment #1 from G. Steinmetz <gscfq@t-online.de> ---

With simplified array constructor :


$ cat z3.f90
program p
   type t
      integer :: a(1) = 2
   end type
   type(t), parameter :: x(1) = [t(3)]
   integer :: y(1)
   y = [x(1)%a]
   print *, y(1)
end


$ gfortran-12-20211114 z3.f90 && ./a.out
           3
$

[Bug fortran/103283] ICE in gfc_trans_array_constructor_subarray, at fortran/trans-array.c:1972

[anlauf at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103283

anlauf at gcc dot gnu.org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever confirmed|0                           |1
   Last reconfirmed|                            |2021-11-30
                 CC|                            |anlauf at gcc dot gnu.org

--- Comment #2 from anlauf at gcc dot gnu.org ---
Reduced testcase:

program p
  type t
     integer :: a(1) = 2
  end type
  type(t), parameter :: x(1) =  t(3)
  print *, (x(1)%a) ! ICE
end

[Bug fortran/103283] ICE in gfc_trans_array_constructor_subarray, at fortran/trans-array.c:1972

[anlauf at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103283

--- Comment #3 from anlauf at gcc dot gnu.org ---
There's more to it:

program p
  type t
     integer :: a(1) = 2
  end type
  type(t), parameter :: x(1) = t(3)
  integer, parameter :: k(*) = x(1)%a
end

This fails with:

pr103283-z1.f90:6:28:

    6 |   integer, parameter :: k(*) = x(1)%a
      |                            1
Error: Cannot initialize implied-shape array at (1) with scalar

Funny.

[Bug fortran/103283] ICE in gfc_trans_array_constructor_subarray, at fortran/trans-array.c:1972

[anlauf at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103283

anlauf at gcc dot gnu.org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |wrong-code

--- Comment #4 from anlauf at gcc dot gnu.org ---
(In reply to anlauf from comment #3)
>   integer, parameter :: k(*) = x(1)%a

> 
> This fails with:
> 
> pr103283-z1.f90:6:28:
> 
>     6 |   integer, parameter :: k(*) = x(1)%a
>       |                            1
> Error: Cannot initialize implied-shape array at (1) with scalar

Doing some debugging, we seem to lose the shape when simplifying the r.h.s.
near the end of expr.c::simplify_parameter_variable:

  /* Do not copy subobject refs for constant.  */
  if (e->expr_type != EXPR_CONSTANT && p->ref != NULL)
    e->ref = gfc_copy_ref (p->ref);
  t = gfc_simplify_expr (e, type);

Initially the ref looks fine, but it might be evaluated improperly.
We end up with a scalar instead of a rank-1 array.

[Bug fortran/103283] ICE in gfc_trans_array_constructor_subarray, at fortran/trans-array.c:1972

[anlauf at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103283

--- Comment #5 from anlauf at gcc dot gnu.org ---
I did get some progress with the attempt:

diff --git a/gcc/fortran/expr.c b/gcc/fortran/expr.c
index 87089321a3b..d5bbcd493b6 100644
--- a/gcc/fortran/expr.c
+++ b/gcc/fortran/expr.c
@@ -1929,6 +1929,8 @@ simplify_const_ref (gfc_expr *p)
            /* Fall through.  */

            case AR_FULL:
+             if (p->ref->next == NULL)
+               return true;
              if (p->ref->next != NULL
                  && (p->ts.type == BT_CHARACTER || gfc_bt_struct
(p->ts.type)))
                {

which fixes testcase z1.f90 as well as comment#3, but has a lot of regressions
in the testsuite.

But then I have to admit I do not understand that function well enough to fix
it.