[Patch, Fortran, OOP] PR 45438: [4.6 Regression] ICE with -fcheck=pointer

[Janus Weil]

2010/9/20 Janus Weil <janus@gcc.gnu.org>:
> Hi Mikael,
>
>>> Index: gcc/fortran/trans-expr.c
>>> ===================================================================
>>> --- gcc/fortran/trans-expr.c  (revision 164445)
>>> +++ gcc/fortran/trans-expr.c  (working copy)
>>> @@ -3169,27 +3169,30 @@ gfc_conv_procedure_call (gfc_se * se, gfc_symbol *
>>>
>>>        if (gfc_option.rtcheck & GFC_RTCHECK_POINTER && e != NULL)
>>>
>>>          {
>>>
>>> -       symbol_attribute *attr;
>>> +       symbol_attribute attr;
>>>
>>>         char *msg;
>>>         tree cond;
>>>
>>>         if (e->expr_type == EXPR_VARIABLE)
>>>
>>> -         attr = &e->symtree->n.sym->attr;
>>> +         attr = e->symtree->n.sym->attr;
>>>
>>>         else if (e->expr_type == EXPR_FUNCTION)
>>>
>>>           {
>>>
>>>             /* For intrinsic functions, the gfc_attr are not available.  */
>>>             if (e->symtree->n.sym->attr.generic && e->value.function.isym)
>>>
>>>               goto end_pointer_check;
>>>
>>> -           if (e->symtree->n.sym->attr.generic)
>>> -             attr = &e->value.function.esym->attr;
>>> +           if (e->ref)
>>> +             /* TBPs & PPCs.  */
>>> +             attr = gfc_expr_attr (e);
>> Would it be possible to use gfc_expr_attr in all cases ?
>
> Actually I'm not sure. But probably we should use it at least for
> EXPR_VARIABLE (otherwise POINTER components etc will not work). I'm
> now regtesting the attached variant and will commit if there are no
> failures.

Regtest successful. Committed as r164462.

Cheers,
Janus