[Patch, fortran] PR29284 - [4.1/4.2 Regression] ICE for optional subroutine argument

[FX Coudert]

Hi Paul,

> In testing for an actual argument that is an assumed character  
> length procedure, I did not test if the expression for the argument  
> is NULL.  This, of course, happens if an optional argument is not  
> present and an ICE ensues for not doing the check.  The testcase is  
> that provided by the author.

Doesn't the same problem happen with the block of code above the one  
you fixed, ie

       /* If an INTENT(OUT) dummy of derived type has a default
          initializer, it must be (re)initialized here.  */
       if (fsym && fsym->attr.intent == INTENT_OUT && fsym->ts.type  
== BT_DERIVED
           && fsym->value)
         {
           gcc_assert (!fsym->attr.allocatable);
           tmp = gfc_trans_assignment (e, fsym->value);
           gfc_add_expr_to_block (&se->pre, tmp);
         }

I think the following code exhibit this behaviour:

        MODULE foo
          TYPE T
            INTEGER :: I = 1
          END TYPE T

          TYPE(T),PARAMETER :: TT = T(1)
        CONTAINS
          SUBROUTINE sub1(a)
            TYPE(T), OPTIONAL, INTENT(OUT) :: a
            WRITE(*,*) 'foo bar'
          END SUBROUTINE sub1
        SUBROUTINE sub2
          CALL sub1()
        END SUBROUTINE sub2
       END MODULE foo


Due to my limited knowledge of these matters (derived types), I'm not  
sure this is allowed, although I think it is. Could you comment on  
this? If this code is indeed legal, I think simply changing the  
condition into "if (e && ...)" would also be correct there. I'm in no  
position to regtest this last change right now, because my tree is in  
an awful mess as I'm finalizing my patch for intrinsics as actual  
arguments.

Thanks,
FX