List:Internals« Previous MessageNext Message »
From:Peter Gulutzan Date:July 10 2009 6:26pm
Subject:Re: ANSI Isolation Level vs. InnoDB consistent read implementation
View as plain text  

Ken Jacobs wrote:

>As a former member (7.5 years) of the SQL committte, including when
>this part of the Standard was written , I guess I should comment.
>Also, I was involved in the design of the Oracle serializable mode,
>and we do believe we comply with the Standard.
>The "normative" part of the Standard is descriptive, but not
>necessarily precise.    Only the syntax and general rules in the body
>of the standard can be used as a yardstick.  You can't interpret the
>English text in any way that differs from the detailed specifications.
>So, as long as a product prevents a transaction from seeing prohibited
>phenomena, it conforms.  Therefore it is correct to say that the
>Standard defines serializable in a way that is less strict than a
>mathematical definition.
>Hope this helps.
>Sent from my iPhone

I suspect you meant the descriptive part of the standard document
is less normative, that is, less compulsory than syntax or general
rules. We could say that about notes, annexes, and (loosely)
SQL/Framework Concepts. But the "serializable" definition
in SQL/Foundation is normative. It is also "normal", that is, it
is a strict definition that a serial execution must be possible.

In the article that Mr Fisk quoted,
I gave a reason for suggesting that Oracle conforms to the standard.
However, as Mr Golubchik said, the standard definition of serializable
is, well, serializable.

Peter Gulutzan
Sun Microsystems / MySQL