Re: Single-statement 'write consistency' on read committed. Oh, really?

> 1. As I mentioned earlier, why bother with the restart at all since the READ
> COMMITTED IL
> promise would not be broken, restart or no restart. What's the rationale for
> the restarts ?

Because the alternative is to process rows that no longer meet the criteria.

> 2. Oracle's behaviour in scenarios 2 and 3, as well as in your example, is
> not intuitive. For example, in scenario 3, the second predicate
> ( ' (pending * 0) = 0' ) will always evaluate to 'true' assuming 'pending
> not null'. If the CBO were smart enough, it would exclude the predicate
> and then the re-scan would not happen thus producing a different result.

The CBO is not smart. It is dumb. It's a computer: all computers are dumb. The CBO implements heuristics. What you are suggesting is that the CBO ought to recognise patterns, and under certain circumstances disregard those rules. Well, that sounds like a lot of complicated programming to me, a certain amount of overhead and a whole lot of grief.

Besides, if Session had set PENDING=NULL then (pending * 0) = 0' would not evaluate to TRUE.

> 3. The behaviour is not described anywhere in Oracle documentation.
>

You got me there.

Cheers, APC Received on Fri Dec 05 2003 - 03:53:30 CST