On the completeness of visual operations for a semantic data model