[Patch, fortran] PR24862 IO for arrays of derived type.

[Tobias.Schlueter@Physik.Uni-Muenchen.DE]

Quoting Janne Blomqvist <jblomqvi@cc.hut.fi>:
> my transfer_array patch that was committed about two months ago didn't
> handle IO for arrays of derived type correctly. This patch fixes it.
>
> The control flow in the new version is a bit convoluted, so I'd be
> happy to recieve suggestions on how to improve it.

WRT this hunk:
+	  tmp = gfc_get_dtype (TREE_TYPE (se.expr));
+	  if (((TREE_INT_CST_LOW (tmp) & GFC_DTYPE_TYPE_MASK)
+	       >> GFC_DTYPE_TYPE_SHIFT) != GFC_DTYPE_DERIVED)
+	    {
+	      tmp = gfc_build_addr_expr (NULL, se.expr);
+	      transfer_array_desc (&se, &expr->ts, tmp);
+	      goto finish_block_label;
+	    }
+	  else
+	    {
+	      /* Cleanup the mess getting the descriptor caused.  */
+	      expr = code->expr;
+	      ss = gfc_walk_expr (expr);
+	      gfc_init_se (&se, NULL);
+	    }

You could do away with the call to gfc_get_dtype and the subsequent if by saying
the following, mimicing the code in gfc_get_dtype:
  if (TREE_TYPE (gfc_get_element_type (se.expr)) == RECORD_TYPE)

Since we usually abstract away the detailed middle-end types, this should
probably be made a predicate function, such as gfc_is_derived_type (tree). 
Apart from that the patch looks fine.

- Tobi