[Bug fortran/102689] New: Segfault with RESHAPE of CLASS as actual argument

[Bug fortran/102689] New: Segfault with RESHAPE of CLASS as actual argument

[burnus at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102689

            Bug ID: 102689
           Summary: Segfault with RESHAPE of CLASS as actual argument
           Product: gcc
           Version: 12.0
            Status: UNCONFIRMED
          Keywords: wrong-code
          Severity: normal
          Priority: P3
         Component: fortran
          Assignee: unassigned at gcc dot gnu.org
          Reporter: burnus at gcc dot gnu.org
  Target Milestone: ---

Created attachment 51586
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=51586&action=edit
testcase - compile & run (fails with segfault at marked line)

Dummy and actual argument are CLASS, then:

   call class_bar (RESHAPE (B, [100]))

will segfault at runtime.  Besides the actual crash, the question is why there
is a call to the library at all:

      _gfortran_reshape_4 (&atmp.25, D.4357, D.4383, 0B, 0B);

At least for most common cases, it looks as if this should be inlined.

[Bug fortran/102689] Segfault with RESHAPE of CLASS as actual argument

[pault at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102689

Paul Thomas <pault at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2023-11-29
                 CC|                            |pault at gcc dot gnu.org
     Ever confirmed|0                           |1

--- Comment #1 from Paul Thomas <pault at gcc dot gnu.org> ---
Confirmed!

implicit none (type, external)
type t
  integer :: i
end type t
integer :: j
integer :: src(100) = [(j, j=1,100)]
class(t), allocatable :: B(:,:,:), D(:)
allocate(t :: B(10,10,1))
B(:,:,:)%i = RESHAPE(src, shape(B))
D = reshape(B, [100])
call class_bar(D)

contains

subroutine class_bar(x)
  class(t), intent(in) :: x(:)
  integer :: k
  if (any (x(:)%i /= [(k, k=1,100)])) error stop 1
end
end

Works as intended. Inspection of the original code reveals:
      atmp.19.dtype = {.elem_len=4, .rank=1, .type=5};
      atmp.19.dim[0].stride = 1;
      atmp.19.dim[0].lbound = 0;
      atmp.19.dim[0].ubound = 99;
      atmp.19.span = 4;
      atmp.19.data = (void * restrict) &A.20;
      atmp.19.offset = 0;
      _gfortran_reshape (&atmp.19, D.4442, D.4451, 0B, 0B);
      class_bar (atmp.19);

ie. The class formal argument is being fed an array descriptor. This is a
problem with gfc_conv_procedure_call.

The discussion about inlining or not is for another day :-)

Thanks for the report, Tobias.

Paul

[Bug fortran/102689] Segfault with RESHAPE of CLASS as actual argument

[pault at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102689

--- Comment #2 from Paul Thomas <pault at gcc dot gnu.org> ---
Created attachment 56730
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=56730&action=edit
Another failing testcase

With the attached, we should get:
[pault@pc30 pr87477]$ rm ./a.out;ifort ~/prs/pr102689/pr1*.f90 ;./a.out
 type is 's' at line            0
 type is 's' at line            2
 type is 's' at line            3
 type is 's' at line            4
 type is 's' at line            5

whereas gfortran gives
 type is 's' at line            0
 type is 't' at line            2
line  2 1275 should be 5050
 type is 't' at line            3
line  3 1275 should be 5050
 type is 't' at line            4
line  4 1275 should be 5050
 type is 't' at line            5
line  5 1275 should be 5050