[Bug fortran/112772] New: Some issues with OPTIONAL, ALLOCATABLE dummy arguments

[Bug fortran/112772] New: Some issues with OPTIONAL, ALLOCATABLE dummy arguments

[anlauf at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112772

            Bug ID: 112772
           Summary: Some issues with OPTIONAL, ALLOCATABLE dummy arguments
           Product: gcc
           Version: 14.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: fortran
          Assignee: unassigned at gcc dot gnu.org
          Reporter: anlauf at gcc dot gnu.org
  Target Milestone: ---

While working on pr100651, the following issues surfaced (not discussed there):

program main
  call test_rank1 ()
contains
  subroutine test_rank1 (msg1)
    class(*), optional, allocatable :: msg1(:)
    if (present (msg1)) stop 77
    call assert_rank1 ()        ! <- no problem here
    call assert_rank1 (msg1)    ! <- problematic code path
  end

  subroutine assert_rank1 (msg2)
    class(*), optional, allocatable :: msg2(:)
  end
end

This crashes at runtime.  The dump for the problematic code path is:

  {
    struct __class__STAR_1_0a class.0;

    if (msg1 != 0B)
      {
        class.0._len = msg1->_len;
        class.0._data = msg1->_data;
        class.0._vptr = msg1->_vptr;
      }
    assert_rank1 (msg1 != 0B ? &class.0 : 0B);
    msg1->_data = class.0._data;
    msg1->_vptr = class.0._vptr;
    msg1->_len = class.0._len;
  }

We need to protect the copy-out for an absent optional argument.

Also, the testcase provided in

https://gcc.gnu.org/pipermail/fortran/2023-November/059969.html

which is similar to the above but uses deferred-length character
shows a questionable tree dump (excerpt):

void test_rank1 (struct array01_character(kind=1) * msg1, integer(kind=8) *
_msg1)
{
  integer(kind=8) D.4350;
  bitsizetype D.4351;
  sizetype D.4352;

  D.4350 = *_msg1;
  D.4351 = (bitsizetype) (sizetype) *_msg1 * 8;
  D.4352 = (sizetype) *_msg1;
  if (msg1 != 0B)
    {
      _gfortran_stop_numeric (77, 0);
    }
  L.2:;
  assert_rank1 (0B, &0);
  {
    struct array01_character(kind=1) * D.4348;
    integer(kind=8) * D.4349;

    D.4348 = msg1 != 0B ? msg1 : 0B;
    D.4349 = msg1 != 0B ? _msg1 : 0B;
    assert_rank1 (D.4348, D.4349);
  }
}

This emits warning messages with -fsanitize=undefined, likely due to the
unconditional dereferencing of _msg1 at function entry.

[Bug fortran/112772] Some issues with OPTIONAL, ALLOCATABLE dummy arguments

[anlauf at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112772

--- Comment #1 from anlauf at gcc dot gnu.org ---
Created attachment 56736
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=56736&action=edit
Fix for testcase 1

The attached rather obvious patch fixes the copy-out issue for class dummies
and regtests ok.

[Bug fortran/112772] Some issues with OPTIONAL, ALLOCATABLE dummy arguments

[anlauf at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112772

anlauf at gcc dot gnu.org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Assignee|unassigned at gcc dot gnu.org      |anlauf at gcc dot gnu.org

--- Comment #2 from anlauf at gcc dot gnu.org ---
Fix for first testcase submitted:

https://gcc.gnu.org/pipermail/fortran/2023-November/059976.html

[Bug fortran/112772] Some issues with OPTIONAL, ALLOCATABLE dummy arguments

[anlauf at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112772

--- Comment #3 from anlauf at gcc dot gnu.org ---
Created attachment 56758
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=56758&action=edit
Patch for testcase 2

This patch makes the initialization code seen in testcase 2
dependent on the presence of the optional argument.

Needs testing.

[Bug fortran/112772] Some issues with OPTIONAL, ALLOCATABLE dummy arguments

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112772

--- Comment #4 from GCC Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Harald Anlauf <anlauf@gcc.gnu.org>:

https://gcc.gnu.org/g:7317275497e10c4a0fb3fbaa6ca87f3463ac124d

commit r14-6066-g7317275497e10c4a0fb3fbaa6ca87f3463ac124d
Author: Harald Anlauf <anlauf@gmx.de>
Date:   Thu Nov 30 21:53:21 2023 +0100

    Fortran: copy-out for possibly missing OPTIONAL CLASS arguments [PR112772]

    gcc/fortran/ChangeLog:

            PR fortran/112772
            * trans-expr.cc (gfc_conv_class_to_class): Make copy-out
conditional
            on the presence of an OPTIONAL CLASS argument passed to an OPTIONAL
            CLASS dummy.

    gcc/testsuite/ChangeLog:

            PR fortran/112772
            * gfortran.dg/missing_optional_dummy_7.f90: New test.

[Bug fortran/112772] Some issues with OPTIONAL, ALLOCATABLE dummy arguments

[anlauf at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112772

anlauf at gcc dot gnu.org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         Resolution|---                         |FIXED
           Keywords|                            |wrong-code
   Target Milestone|---                         |14.0
             Status|UNCONFIRMED                 |RESOLVED

--- Comment #5 from anlauf at gcc dot gnu.org ---
Fixed testcase 1.

Testcase 2 is dealt with by the fix for pr100651.