Pre-relational database models' influence on "theoreticians"
Quote from AntC on August 4, 2019, 7:49 am
The application's data model had 'meaningless' integral PRIMARY KEYs on every row -- including on 'bridging tables' whose only purpose was to link other 'meaningless' IDs.
Of course that's silly. But it's just a case of Good Rules Gone Bad, like American English copy editors trying to convert all restrictive which to that, or going from "Starting all your sentences with And is boring" to "Never start a sentence with And."
When the subject matter of a relation is some Real World object, synthetic primary keys are pretty much a necessity. I know of no property of human beings, for example, that is both unique and stable (not even "abstract" properties like national identification numbers). We don't even know what identity across time means for physical objects, despite attempts made going back to Heraclitus (the ship of Theseus) and doubtless well beyond. Yet as long as we insist on using nouns, we are going to have to pretend that we know when something is "the same thing" that we've seen before, and when it isn't. Babies are pretty good at it after about six months.
Chosen points on the geoid (an abstraction of the Earth's surface), on the other hand, need a primary key of {latitude, longitude}, truncated to a tractable precision. Synthetic primary keys are folly there.
But the application's meaningless IDs were already persistent, with all the changing dimensions non-key.
Another instance of GRGB, from "don't trust what other people provide without checking" to "never trust anything that anyone else provides." Five minutes' reflection would show that if other people are unreliable, so are you. See also Jante Law.
I was told I was being old-fashioned for worrying about disk space. No that wasn't my worry: it was the bottleneck on the GUID-allocator. Did I mention these IDs had to be globally unique, not just unique within their table?
While certainly valid in its time, I think that's obsolete, given the availability of strong random number generators, which are greatly superior to incrementing-integer keys "from every conceivable point of view" (Paul Pedersen).
Then can we say GUIDs (database-global) are a more reliable indicator of OO thinking?
Quote from AntC on August 3, 2019, 6:53 amThere's somebody I would describe as a RM troll just materialised on StackOverflow. I don't think it's Fabien under an alias(?)
No, PerformanceDBA on SO is I-don't-remember-his-name but it's the guy who managed to get kicked out of the old discussion list.
I'm rather surprised he's surfacing again.
Quote from Erwin on August 4, 2019, 6:18 pmQuote from AntC on August 3, 2019, 6:53 amThere's somebody I would describe as a RM troll just materialised on StackOverflow. I don't think it's Fabien under an alias(?)
No, PerformanceDBA on SO is I-don't-remember-his-name but it's the guy who managed to get kicked out of the old discussion list.
I'm rather surprised he's surfacing again.
The guy who kept moaning that if people would just use IDEFIX modeling they would never even run into any of the problems TTM claimed to solve.
Quote from Erwin on August 4, 2019, 6:24 pmQuote from Erwin on August 4, 2019, 6:18 pmQuote from AntC on August 3, 2019, 6:53 amThere's somebody I would describe as a RM troll just materialised on StackOverflow. I don't think it's Fabien under an alias(?)
No, PerformanceDBA on SO is I-don't-remember-his-name but it's the guy who managed to get kicked out of the old discussion list.
I'm rather surprised he's surfacing again.
The guy who kept moaning that if people would just use IDEFIX modeling they would never even run into any of the problems TTM claimed to solve.
Derek A.
Quote from Erwin on August 4, 2019, 6:33 pmQuote from Erwin on August 4, 2019, 6:24 pmQuote from Erwin on August 4, 2019, 6:18 pmQuote from AntC on August 3, 2019, 6:53 amThere's somebody I would describe as a RM troll just materialised on StackOverflow. I don't think it's Fabien under an alias(?)
No, PerformanceDBA on SO is I-don't-remember-his-name but it's the guy who managed to get kicked out of the old discussion list.
I'm rather surprised he's surfacing again.
The guy who kept moaning that if people would just use IDEFIX modeling they would never even run into any of the problems TTM claimed to solve.
Derek A.
I intentionally avoided mentioning his name in post #2 in this thread, lest he google himself (he's the sort who probably does that frequently), find he's mentioned here and make himself annoying (he's the sort who probably does that frequently) to me or others. Hopefully, your reference is sufficiently oblique that he won't find himself.
Yes, he thought IDEFIX was the solution to everything, Sybase was the ultimate DBMS, Codd was infallible (and perhaps omniscient), and his post series on transactions was pure gold (hint: all you need is a transaction number, apparently.)
But I didn't ban him for any of those things. I banned him because he turned out to be a blatant racist.
Quote from Dave Voorhis on August 4, 2019, 7:44 pm(hint: all you need is a transaction number, apparently.)
Well, it is all you need — if it has enough bits in it.
Quote from johnwcowan on August 4, 2019, 9:27 pmQuote from Dave Voorhis on August 4, 2019, 7:44 pm(hint: all you need is a transaction number, apparently.)
Well, it is all you need — if it has enough bits in it.
I'm fairly sure that's not what he meant.
Quote from Dave Voorhis on August 4, 2019, 9:29 pm
I'm fairly sure that's not what he meant.
Well, yes. In any case, IDEFIX modeling is French, and therefore uninteresting. Here in the English-speaking lands, all modelers have switched to DOGMATIX.
(Sorry, it's Sunday.)
A more serious question: What would be a reasonable name for the network/DBTG/Codasyl data model that is neither actively misleading (like the first) nor utterly indecipherable (like the last two)? (No wisecracks, please; I can make up my own.) It should not involve "set", because Codasyl sets are neither unordered nor have unique elements.
Quote from johnwcowan on August 4, 2019, 9:36 pmA more serious question: What would be a reasonable name for the network/DBTG/Codasyl data model that is neither actively misleading (like the first) nor utterly indecipherable (like the last two)? (No wisecracks, please; I can make up my own.) It should not involve "set", because Codasyl sets are neither unordered nor have unique elements.
The "network/DBTG/Codasyl data model"?
That's not really a wisecrack, but as you pointed out, it is Sunday.
Quote from Erwin on August 4, 2019, 6:33 pmQuote from Erwin on August 4, 2019, 6:24 pmQuote from Erwin on August 4, 2019, 6:18 pmQuote from AntC on August 3, 2019, 6:53 amThere's somebody I would describe as a RM troll just materialised on StackOverflow. I don't think it's Fabien under an alias(?)
No, PerformanceDBA on SO is I-don't-remember-his-name but it's the guy who managed to get kicked out of the old discussion list.
I'm rather surprised he's surfacing again.
The guy who kept moaning that if people would just use IDEFIX modeling they would never even run into any of the problems TTM claimed to solve.
Derek A.
391457b if anyone really wants to know. [Feel free to delete.]