Re: Failure Modes and Ranges of Operation

From: paul c <toledobythesea_at_oohay.ac>
Date: Sat, 03 Feb 2007 19:07:38 GMT
Message-ID: <_35xh.872088$1T2.70464_at_pd7urf2no>


Bob Badour wrote:
> ...
> You've lost me, Paul. What is the problem with updating a union view?
> ...

There were some arguments on dbdebunk about it but I forget exactly what the titles were, only that Date said he had changed his mind about what he and McGoveran wrote back in 1994 or so where they basically said, if I remember right, that it is logically okay to insert to both A and B when an insert to the view A OR B is presented.

Maybe a better one is what Darwen said in his TTM group (I'm sure he wouldn't mind me quoting him as the quote itself suggests that he's said it lots of times):

(quote)
I question I have asked, of people who propose to support insert-through-union by inserting into both operands (or delete-through-join by deleting from both) is this:

Let x and y be Boolean variables and let z be defined as the view "x OR y". Would you support the assignment z := TRUE? If not, how do you square this with your position on insert-through-union? A similar question applies if z is "x AND y" and the assignment is z := FALSE. (end of quote)

I got into some to and fro' about this, saying that I thought it was okay to assign TRUE to both x and y, if an engine allows a user to insert to such a view. I must confess that I didn't fully understand the opposing arguments which were very technical, to do with imperative languages and determinism, if I recall right. My probably naive attitude remains that it's okay, say, to change y from FALSE to TRUE, even if whoever defined y in the first place never expected it to be used in a view, assuming that all relational ops have a single relation result. However, in the face of opposition by some pretty big db guns, I have to feel nervous about this.

>

>> If I think of all devices as having inputs and outputs (even a bridge 
>> that doesn't appear to move), such as air and fuel mixed in some 
>> proportion, it seems reasonable to not worry much about the times when 
>> somebody supplies an unacceptable input.

>
>
> I suspect the pilot who flew beyond the fuel supply might disagree with
> you on that one.
> ...

Yes, heh-heh, if I had been in charge of Pearl Harbour in 1941, probably I too would have fired shortly after the attack. I can see how "reasonable" has different meanings in wartime or around nuclear reactors - I suspect nobody was fired when they screwed up metric litres and US gallons forcing the AC pilot had to land a bit short in Gimli, personally I think somebody should have been.

>
> It's the combinations of
>

>> inputs that produce unexpected outputs that then can't be used as 
>> inputs    that puzzles me.

>
>
> You've lost me again. Exactly what outputs are unexpected?

I guess I was thinking of something like the insert-through-union problem above, where I take it that many people think 'not possible' is the intermediate output, an unacceptable 'operating range', if you will (not so much of arithmetic results like 'not a number').

p Received on Sat Feb 03 2007 - 20:07:38 CET

Original text of this message