[fparser2 frontend] Unbounded loops not correctly reproduced

[arporter]

Simon has reported:

"Apparently unbounded loops with a nested inner loop are incorrectly reproduced: the nested loops of

PROGRAM x
  INTEGER :: y
  DO
    DO y=1,2
    END DO
  END DO
END PROGRAM x

are currently transformed into two nested loops with identical loop variables and parameters."

I would have expected the unbounded loop to result in a CodeBlock so I'm not sure what's going wrong here.

[arporter]

Victoria has also hit a related problem in NEMOVAR where there are similar unbound loops that then contain loops that we want to parallelise:

  CBV_loop : DO

     CALL soladj_cbv_main( &
        &  piom_ctl, &
        &  p_dif%fsol%fcbv, &
        &  p_dif%fsol%fcbv%cb_b, &
        &  p_dif%fsol%fcbv%cb_x, &
        &  i_nsolitmax, &
        &  p_dif%fsol%nsolit_k, &
        &  i_flag_rc, &
        &  z_arginp, &
        &  z_argout, &
        &  ld_gpu=ll_gpu_override )

     SELECT CASE( i_flag_rc )

     CASE ( jp_cbv_A_x0, jp_cbv_A_p )

        ! Point to output vector

        CALL pointval_ctlvec( &
           &  piom_ctl, &
           &  p_dif%fgrid%fmpp, &
           &  1, &
           &  z_ptr2, &
           &  z_argout )
        !$ACC ENTER DATA IF(ll_transfer) ATTACH ( z_ptr2 )

        ! Retrieve input from the control vector

        IF ( .NOT. ASSOCIATED( z_argout, z_arginp ) ) THEN

           CALL getval_ctlvec( &
              &  piom_ctl, &
              &  p_dif%fgrid%fmpp, &
              &  1, &
              &  z_ptr2(:,:,:), &
              &  z_arginp, &
              &  ld_gpu=ll_gpu_override )

        END IF

        IF ( p_dif%fsol%ldiaprc ) THEN

           ! Multiply by the square root of the diagonal preconditioner
           IF(ll_gpu) THEN
              !$ACC KERNELS
              !$ACC LOOP COLLAPSE(3) GANG VECTOR INDEPENDENT
              DO jk=1,p_dif%npk
                 DO jj=1,p_dif%npj
                    DO ji=1,p_dif%npi
                       z_ptr2(ji,jj,jk) = z_ptr2(ji,jj,jk) * p_dif%fsol%diaprciijj(ji,jj,jk,i_nbc)
                    END DO
                 END DO
              END DO
              !$ACC END KERNELS
           ELSE
              z_ptr2(:,:,:) = z_ptr2(:,:,:) * p_dif%fsol%diaprciijj(:,:,:,i_nbc)
           END IF
        END IF

    ...
  END DO CBV_loop

Currently this whole loop is (correctly) put into a CodeBlock but obviously that prevents transformation of the loops nested within it.