[Bug fortran/100650] New: Passing a derived-type array constructor to the reshape intrinsic gives incorrect results

[Bug fortran/100650] New: Passing a derived-type array constructor to the reshape intrinsic gives incorrect results

[here.is.a.gcc.bug at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100650

            Bug ID: 100650
           Summary: Passing a derived-type array constructor to the
                    reshape intrinsic gives incorrect results
           Product: gcc
           Version: 11.1.1
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: fortran
          Assignee: unassigned at gcc dot gnu.org
          Reporter: here.is.a.gcc.bug at gmail dot com
  Target Milestone: ---

Created attachment 50836
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=50836&action=edit
A minimum broken example. Needs no special compiler flags.

A type `String` has a `character(:), allocatable` component.

When an array constructor of type `String` is passed to the reshape intrinsic,
the components of the resulting array are incorrect.

The bug does not trigger if the array constructor is used to create a named
array and this named array is passed to the reshape intrinsic.

This affects at least versions 7.4.0, 10.1.0 and 11.1.1 of gfortran.

[Bug fortran/100650] Passing a derived-type array constructor to the reshape intrinsic gives incorrect results

[damian at archaeologic dot codes]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100650

Damian Rouson <damian at archaeologic dot codes> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |damian at archaeologic dot codes

--- Comment #1 from Damian Rouson <damian at archaeologic dot codes> ---
Here's a shorter reproducer:

% cat reshape-array-constructor.f90 
  type foo_t
    integer, allocatable :: bar(:)
  end type

  type(foo_t) foo(2,1)
  integer :: i, j,  n(2,2,1) = reshape([-1,-1,1,1], [2,2,1])

  foo = reshape([([(foo_t(merge(0, 1, n(:,i,j) > 0)), i=1,2)], j=1,1)], [2,1])
  print *, merge(0, 1, n(:,1,1) > 0), foo(1,1)%bar
  print *, merge(0, 1, n(:,2,1) > 0), foo(2,1)%bar
end
% gfortran reshape-array-constructor.f90 
% ./a.out
           1           1  -358465392       24440
           0           0  -358465392       24440
% gfortran --version
GNU Fortran (Homebrew GCC 13.1.0) 13.1.0

Another workaround in the above example is to make n a constant array via

integer, parameter :: n(*,*,*) = reshape([-1,-1,1,1], [2,2,1])

[Bug fortran/100650] Passing a derived-type array constructor to the reshape intrinsic gives incorrect results

[damian at archaeologic dot codes]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100650

--- Comment #2 from Damian Rouson <damian at archaeologic dot codes> ---
Is this related to PR 49324?

[Bug fortran/100650] Passing a derived-type array constructor to the reshape intrinsic gives incorrect results

[pault at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100650

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Last reconfirmed|                            |2023-06-21
                 CC|                            |pault at gcc dot gnu.org
     Ever confirmed|0                           |1
             Status|UNCONFIRMED                 |NEW

--- Comment #3 from Paul Thomas <pault at gcc dot gnu.org> ---
(In reply to Damian Rouson from comment #1)
> Here's a shorter reproducer:
> 
> % cat reshape-array-constructor.f90 
>   type foo_t
>     integer, allocatable :: bar(:)
>   end type
> 
>   type(foo_t) foo(2,1)
>   integer :: i, j,  n(2,2,1) = reshape([-1,-1,1,1], [2,2,1])
> 
>   foo = reshape([([(foo_t(merge(0, 1, n(:,i,j) > 0)), i=1,2)], j=1,1)],
> [2,1])
>   print *, merge(0, 1, n(:,1,1) > 0), foo(1,1)%bar
>   print *, merge(0, 1, n(:,2,1) > 0), foo(2,1)%bar
> end
> % gfortran reshape-array-constructor.f90 
> % ./a.out
>            1           1  -358465392       24440
>            0           0  -358465392       24440
> % gfortran --version
> GNU Fortran (Homebrew GCC 13.1.0) 13.1.0
> 
> Another workaround in the above example is to make n a constant array via
> 
> integer, parameter :: n(*,*,*) = reshape([-1,-1,1,1], [2,2,1])

I have played with this a bit. It's a complete mess. Breaking out the inner
array constructor to a temporary produces the same result.

I have put it on my list of TODOs. First finish associate, gather up the last
of the select type bugs and attack PDTs.

Cheers

Paul

[Bug fortran/100650] Passing a derived-type array constructor to the reshape intrinsic gives incorrect results

[damian at archaeologic dot codes]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100650

--- Comment #4 from Damian Rouson <damian at archaeologic dot codes> ---
Thanks, Paul.  No urgency.  The workaround is simple and painless.