[
Lists Home |
Date Index |
Thread Index
]
Is the reason that we accept schemas with non-semantic required-
ordering just a matter of the tools we have to use? DTDs make "and next"
easy to declare but "and also" difficult. So there is
(a, b, c, d)
but no
(a && b && c && d)
Similarly there is
(a | b | c | d)
but no
(a || b || c || d)
(meaning you can have one of these, anywhere)
I used to think this was just intrinsic to grammars (because we need to
minimise combinatorial explosions for the automaton) but it seems the
derivation method used in some RELAX NG (brought up by Joe English)
has overcome this.
This kind of question came up during WXS 1.0 discussions. One opinion that
came up was that it should be treated as an information-theoretic
problem: the more order we enforce, the more compressable, indexable
etc the data is. I tend to think that is mumbo jumbo, myself: for the
sake of flexibility, the less we encourage people build non-semantic
constraint into their processing software the better. Of course,
projects with niche requirements (data serialized from legacy databases,
one-off projects with no maintenence, or high-volume systems) may
have a different trade-off.
Cheers
Rick Jelliffe
|
