[Patch, fortran] PR47592 - Multiple function invocation with ALLOCATE (SOURCE=REPEAT('x',bar()))

[Paul Richard Thomas]

Dear Tobias,

I thought that you were not doing gfortran this weekend :-)

> Note that I have removed the calls of gfc_start_block and replaced
> them with gfc_init_block, since the former does all sorts of strange
> things with declarations as the warning in trans.c indicates.
>
> I think there are several places in gfortran where an init_block would be
> better than a start_block.

Indeed - I have been burned more than once by using start, rather than init.

>
> Bootstraps and regtests on FC9/x86_64 - OK for trunk?
>
> OK, though you could consider shortening the code as indicated below. Thanks
> for the patch!
>
> +
> ! 	  ref = expr->ref;
> ! 	  /* Find the last reference in the chain.  */
> ! 	  while (ref && ref->next != NULL)
> ! 	    {
> ! 	      gcc_assert (ref->type != REF_ARRAY || ref->u.ar.type ==
> AR_ELEMENT);
> ! 	      ref = ref->next;
> ! 	    }
> ! 	  if (!ref)
> ! 	    allocatable = expr->symtree->n.sym->attr.allocatable;
> ! 	  else
> ! 	    allocatable = ref->u.c.component->attr.allocatable;
> !
> ! 	  if (allocatable)
>
>
> That's mostly unchanged from previous version, but couldn't one replace it
> by a simple "if (gfc_expr_attr (expr).allocatable)"? I think gfc_expr_attr
> does not always do the right thing (regarding when to ignore and not ignore
> array/substring refs), but I think here it does.

I'll give it a try.  Thanks for the review.  Now to module backend_decls.....

Cheers

Paul