View Updating and Relational Theory (Theory in Practice)

Q&A with C.J. Date, author of "View Updating and Relational Theory"

Q. Why did you write this book?

A. View updating has long been regarded as one of the great unsolved problems in the database world. It’s also something I’ve been interested in for just about as long as I can remember! In fact, my very first two published papers (done jointly with a colleague at the time, Paul Hopewell), which appeared in 1971, had to do with view updating. Now I think I have something useful and constructive to say regarding how to solve this problem.

Q. So what exactly is the problem?

A. Well, perhaps I should first say what views are. As you know, databases are made up of tables. And there are two main kinds of tables-- base tables, which are the “real” ones in a sense, and views, which are “virtual” tables, defined in terms of the real ones. For example, a personnel database might have two base tables, EMP and DEPT, containing information about employees and departments, respectively. Now imagine some specific user of that database who happens to be interested only in employees who are programmers and who work in a department located in Silicon Valley. We can make life easier for that user by defining a view over EMP and DEPT, derived by (a) joining EMP and DEPT together and then (b) discarding everything from that join that doesn’t concern Silicon Valley programmers. As far as that user is concerned, that view is the database-- he or she can operate on that view exactly as if it were a base table. At least, that’s the idea. So that’s what views are-- virtual tables that the user thinks are real ones.

But since views are virtual, in a sense they don’t exist! (You know the definition of “virtual”: You can see it but it isn’t there. It’s the opposite of “transparent,” which means it’s there but you can’t see it. But I digress.) So when the user performs some operation on a view, what the system has to do is convert that operation into operations on the real tables (that is, the base tables) in terms of which that view is defined. Now, it turns out that this conversion process is comparatively straightforward in the case of retrieval operations-- it works 100 percent of the time. But update operations are much harder! In fact, many people have thought for a long time that the problem of converting update operations properly is, in the final analysis, unsolvable-- that is, certain updates on certain views just can’t be done at all.

Q. But you don’t agree with that position-- right?

A. Right! Numerous and varied approaches to the view update problem have been investigated over the past 40 years or so, but those approaches-- at least, the ones I’m familiar with-- just didn’t solve the problem. That is, they showed how some views could be updated, but there were always others that couldn’t be. By contrast, what I’m proposing (and here let me give credit to my friend David McGoveran, who’s been working with me on this issue for several years and whose ideas have heavily influenced my thinking on this topic) in effect allows all updates on all views. Let me immediately add that it’s true that those updates will sometimes fail, of course-- to be specific, they’ll fail if they violate some integrity constraint-- but that’s true for updates on base tables as well.

Q. So why is the ability to update views important?

A. Because without it, we have to give up on the goal of data independence-- a goal that, as a matter of fact, was one of Ted Codd’s primary motivations for introducing the relational model in the first place, back in 1969. By the way, I’m assuming here that you do know what data independence is. If you don’t, then please read the book!

Q. Any last points you’d like to make?

A. Yes. I said my scheme “allows all updates on all views.” That’s true, but let me say also that some people will probably find parts of what I’m proposing a little controversial. In this connection, I’ve tried to be as honest as I can and point out in the book those aspects of my proposals where there might well be some debate. So I don’t claim that what I have to say is the last word on the subject. Rather, what I’ve tried to do is make a step forward-- I’d like to think, a big step forward-- and move the debate up to another level, as it were. But I’m certainly open to constructive suggestions regarding possible improvements to what I propose.