New ICE -Wall?

[Jerry DeLisle]

Paul Thomas wrote:
> There is something horribly wrong with my ISP tonight: I have seen this 
> on the list but haven't had the forward yet - hence not noticing that 
> the patch is not attached.
>> This is embarrassing - I forgot about derived type components *sigh*
>>
>> Since the code->expr has been resolved, the expr character length is 
>> as available as it ever will be, if it is constant.  Thus the patch 
>> uses that, rather than the symbol's.
>>
>> I notice that substring references could be resolved but are not - see 
>> below.
>>
>> I can do something about this at the weekend - if anybody wants to get 
>> this out of the door now, please do.  Note that it is not regtested.
>>
>> Paul
>>
>> module linebufmod
>> implicit none
>>
>> type linebuf
>>  character(len=80), dimension(:), pointer :: data=>NULL()
>> end type linebuf
>>
>> type linebuf_short
>>  character(len=2), dimension(:), pointer :: data=>NULL()
>> end type linebuf_short
>>
>> contains
>>
>> subroutine linebuf_add (buf, buf_short)
>>  type(linebuf), intent(inout) :: buf
>>  type(linebuf_short), intent(inout) :: buf_short
>>  character(4) :: chr
>>
>>  buf%data(1) = "blah"
>>  buf%data(1)(1:2) = "blah"   ! misses this
>>  buf_short%data(1) = "blah"  ! { dg-warning "will be truncated" }
>>  chr(1:2) = "blahblah"       ! misses this too
>> end subroutine linebuf_add
>>
>> end module linebufmod
>>
> 
> 
> ------------------------------------------------------------------------
> 
> Index: gcc/fortran/resolve.c
> ===================================================================
> *** gcc/fortran/resolve.c	(revision 120520)
> --- gcc/fortran/resolve.c	(working copy)
> *************** resolve_code (gfc_code * code, gfc_names
> *** 5088,5097 ****
>   		&& gfc_option.warn_character_truncation)
>   	    {
>   	      int llen = 0, rlen = 0;
> ! 	      gfc_symbol *sym;
> ! 	      sym = code->expr->symtree->n.sym;
> ! 	      if (sym->ts.cl->length->expr_type == EXPR_CONSTANT)
> ! 		llen = mpz_get_si (sym->ts.cl->length->value.integer);
>   
>   	      if (code->expr2->expr_type == EXPR_CONSTANT)
>   		rlen = code->expr2->value.character.length;
> --- 5088,5098 ----
>   		&& gfc_option.warn_character_truncation)
>   	    {
>   	      int llen = 0, rlen = 0;
> ! 
> ! 	      if (code->expr->ts.cl != NULL
> ! 		    && code->expr->ts.cl->length != NULL
> ! 		    && code->expr->ts.cl->length->expr_type == EXPR_CONSTANT)
> ! 		llen = mpz_get_si (code->expr->ts.cl->length->value.integer);
>   
>   	      if (code->expr2->expr_type == EXPR_CONSTANT)
>   		rlen = code->expr2->value.character.length;

I will regression test this tonight and if all goes well,  will prepare a commit 
and report back tomorrow night.

Jerry