[QUOTE]Originally posted by julian
When you borrow a book from the library do you read the book or the library :-)
Sigh...

I said: that is what I have heard and I asked if it is true. Read carefully!
Your actual argument was:
Originally posted by julian

I thought that Oracle has made the change in 8i so that if you do not touch the real column, i.e., if you go for just pseudo-columns, then block is not read.
Not "that's what I've heard", but "that's what I think", isn't it? But it realy doesn't matter if you "think" or if you "have heard", my point was very clear: If you only realy give it a second thought it is more than self-evident that number of records can only be gotten from database blocks themself! Oracle can make no trics there... And no need for any "reliable URLs" to prove that .

My bottom line: The discussion of COUNT(rowid) vs. COUNT(1) vs. COUNT(*) is pointless, it is a no-isue. It is known that there is absolutely no difference among them. The discussion about COUNT() only leeds away from the real isue about this topic "Using ROWID in SQL queries " is somewhere else - READ CONSISTENCY! My basic claim was and still is: Wherever Oracle guarantees read consistency it is safe to use ROWIDs! I guess I should have entiteled this topic differently: "Using ROWID in SQL queries in the context of read consistency".