[
Lists Home |
Date Index |
Thread Index
]
> >>>From: "Sean McGrath" <sean.mcgrath@propylon.com>
> >>> Schemas have entropy - a tendancy towards disorder - disorder the grows
> >>> over time as the schema is modified....
> >>[Rick Jelliffe]
> >>I wonder how much is this actually a problem with
>
> >>1) Using content models, rather than systems that allow partial orderings
> >>to be declared (i.e., "bib should come before bub and to hell wit the
> rest")?
[Michael Sperberq-McQueen]
>I'm confused here: you seem to be saying that you don't think
>content models, or more generally formal grammars as defined by
>Chomsky, can express the constraint "bib should come before bub
>and anything else can occur in any order". Surely you can't
>believe that.
Here is what happens in my experience:
Constraint: "bib must occur before bub and then anything can occur in any
order"
Developer: "This is manageable in a number of grammar based schema languages.
For example,...yada, yada..."
Time moves on, the constraint changes:
Constraint: "bib must occur before bub and then anything can occur in any
order unless there
if a bab in there somewhere."
Developer: "Oh dear. If it can be done at all, it will make the model a lot
more complex....
How about this? "(bib|bub|...)*"
Simply put, my experience has been that as soon as grammar based
constraints start to
stretch a grammar based schema language, the content models embark on a
slippery slope towards
"everything optional, anything in any order" In SGML, advanced cases of
this phenomenon
took the form
<!ELEMENT foo (A|B|C...)* + (i1,i2,i3...)>
(Note to non-SGML people: the weird bit on the end is called "inclusion
exceptions". Buy an
old-timer a beer for more details:-)
In XML DTDs, its a bit simpler but still pretty horrid:
<!ELEMENT foo (A|B|C...)*>
I think this is one of the reasons why WF XML ended up fitting some SGML
systems
in industry quite well. Over time, the schemas had become less and less potent
anyway to the point what all they were really doing was checking that all
elements used were declared in the "tag bag" DTD :-/
I think Rick's thinking (and I agree with it) is that rule based validation
constraints
that concentrate on constraining aspects of the structure rather than the
whole structure, do not "degrade" as extra constraints need to be added
into the validation.
The theoretical expressibility of constraints in either grammer or rule
based systems in not the central concern here. The concern is how
robust the two approaches are in the face of (inevitable) real world
change.
Sean
http://seanmcgrath.blogspot.com
|
