This might be excellent: fully implemented MERGE in other RDBMS seemed to be highly unreliable. Given Postgres' focus on quality, not on number of features, not

This might be excellent: fully implemented MERGE in other RDBMS seemed to be highly unreliable. Given Postgres' focus on quality, not on number of features, not implementing MERGE fully might be a great choice. Just imagine how many permutations we'd need to unit test the implementation of full MERGE. Sounds scary. Better develop a frequently used subset, the UPSERT, and stop at that.