Yes, I think this (possibly combined with anonymous struct types) would be better than structural records.
This came up in