This is expected as the current compiler behaves like ETL3. The new standard is more flexible about that I believe, but our compiler does not handle this because we are trying to optimize expanded types and when used as an attribute it prevents that optimization.