On Fri, Jul 31, 2020 at 11:41:59AM +0100, Richard Sandiford wrote:
> @@ -10135,11 +10176,7 @@ process_init_element (location_t loc, struct c_expr
> value, bool implicit,
> /* Otherwise, if we have come to a subaggregate,
> and we don't have an element of its type, push into it. */
> else if (value.value != NULL_TREE
> - && value.value != error_mark_node
> - && TYPE_MAIN_VARIANT (TREE_TYPE (value.value)) != fieldtype
> - && (fieldcode == RECORD_TYPE || fieldcode == ARRAY_TYPE
> - || fieldcode == UNION_TYPE
> - || gnu_vector_type_p (fieldtype)))
> + && initialize_elementwise_p (fieldtype, value.value))
I wonder if it wouldn't be better to do the value.value != NULL_TREE
checking inside of initialize_elementwise_p too, simply because it is done
the same way in all the 3 callers.
Other than that LGTM, but as I'm not a C FE maintainer, please give them a day
to express their opinions (though Joseph commented in the PR, so I assume he
is ok with the general idea).
> {
> push_init_level (loc, 1, braced_init_obstack);
> continue;
> @@ -10227,11 +10264,7 @@ process_init_element (location_t loc, struct c_expr
> value, bool implicit,
> /* Otherwise, if we have come to a subaggregate,
> and we don't have an element of its type, push into it. */
> else if (value.value != NULL_TREE
> - && value.value != error_mark_node
> - && TYPE_MAIN_VARIANT (TREE_TYPE (value.value)) != fieldtype
> - && (fieldcode == RECORD_TYPE || fieldcode == ARRAY_TYPE
> - || fieldcode == UNION_TYPE
> - || gnu_vector_type_p (fieldtype)))
> + && initialize_elementwise_p (fieldtype, value.value))
> {
> push_init_level (loc, 1, braced_init_obstack);
> continue;
Jakub
