I think that I would be satisfied with something less ambitious: If the source stream for the INSERT involves a table function, then always attempt a bulk insert--naturally, the bulk insert optimization would be applied only if the target table was empty. This wouldn't
+1 to improved performance with no non-standard language changes! +1 also to the suggestion of Knut Anders to enable a standards-compliant TRUNCATE TABLE and then gain import speed that way. The entire discussion has been interesting and instructive, thank you for helping us understand the issues, even though the rest of us aren't commenting :) thanks, bryan
