Re: Ideas for World Hierarchy Example
Date: Mon, 15 Jan 2007 09:38:29 GMT
Message-ID: <pYHqh.657217$1T2.72647_at_pd7urf2no>
Bob Badour wrote:
> Marshall wrote: >
>> On Jan 14, 10:22 am, Bernard Peek <b..._at_shrdlu.com> wrote:
>>
>>> How about turning the collective brainpower of the group on a
>>> constructive task: design a workable computer-based EAV system. The
>>> librarians built theirs 200 years ago, and it has been largely unchanged
>>> since Mr Dewey used Mr Aristotle's classification methods to create it.
>>> Can we do better in the 21st century?
>>
>>
>> Been working on this for the last two years with a decent sized team.
>> (I argued against this approach early on and lost.) The problems with
>> EAV are fundamental, so I don't see any path to doing "better."
>>
>> EAV is a cheapo way to capture some data when your requirements
>> are fuzzy, or when you don't understand your domain well enough
>> to model it. It lacks structure, so you necessarily do not get the
>> benefits of structure, semantic or otherwise.
>>
>> Sometime I mean to sit down and write up some queries in SQL
>> against a fully modeled schema and against an EAV schema and
>> compare.
> > > What do you consider the potential for using the EAV approach as a > physical storage medium from which to deliver relational results at the > logical level?
If one's starting point is Codd's theory, this comes closer than the usual EAV talk to hitting the nail on the head because it hints at a physical question. I think Codd pre-supposed that 1) a proper system would anticipate the questions users could ask of a given range of types and values and because of this would 2) stipulate how the range could be augmented. (He also had in mind parts of long-standing set and logic theory to tell us why we should believe the answers from such a system but as for multiple applications he left the opportunities for what he called "assymetrical exploitation" undescribed, merely pointing out that with his approach, two apps could share the same data bank which was a big deal at the time.) The EAV proponents I've met always missed these two points, preferring to think that any system could anticipate any data and any question without preparation or proof.
As Marshall hints at, EAV as it seems to be practised is an informal way to pretend there are no requirements for an application and then to change hats and stamp it paid. To date, I believe its proponents have also failed to acknowledge that a physical implementation might also have requirements and they have lazily assumed that the relational ops are appropriate. It seems to me that they might to take a clue from the main motive of normal form theory - what forms or structures are equivalent? It seems bass-ackwards to plant EAV on top of SQL, that acronym having come to be a porridge word that means nothing more general than a barometer for an implementer's narrow vision.
Personally, I think there is a place for a physical structure formality that might be analogous to the logical normal forms but the EAV proponents I've met tended to be one-tool wonders and so were at a loss to see physical opportunities. At least this might make the purported relational implementions a little more coherent and a little less idiosyncratic. Apart from that, something else seems to me to be missing, for example, it is known that certain kinds of recursion, eg., relations defined on rva types, can't be expressed without rva's unless information is added to them, ie., additional attributes. No offence to Codd's memory, but I'm guessing he saw the information principle as a way to wash our hands of these.
p Received on Mon Jan 15 2007 - 10:38:29 CET