Re: Attribute uniqueness test: a radical proposal

>However, it is a very wise architect who can forsee all uses of a
>general technology so well in advance so as to be able to determine
>what sort of features are unwise for all future applications.

At this point, I think I have to remind folks that XML originated very
specifically as a _reduction_ from SGML, based on the 90/10 rule. ("Handle
the 90% most common use cases which can be addressed with 10% of the coding
effort; ignore the remaining 10% which require the other 90% of the
effort.")

XML --- even the portion we consider "core XML" -- has been growing rapidly
since then. It's already become hard to believe the claim that a student
can implement a functionally complete (if inefficient) XML processor as a
final project. I'm afraid we may be on the verge of losing our focus and
turning back into SGML bloatware.

Hence, I strongly suggest that we _NOT_ attempt to allow for every possible
scenario. If there are important use cases that must be addressed, and
making the Namespace Declaration accept relative URI Reference syntax is
clearly the best way to address them, that's one thing. If we're just
guessing about the future, I submit that this can be dealt with in a future
revision of XML, or in a more specific tool derived from XML, once we have
a better sense of what problems we're actually trying to solve and what
approaches to solving them might be appropriate.

There used to be a PL/I coding example which went something like:
     IF IF IF = THEN THEN THEN = ELSE ELSE ELSE = IF THEN ...
(Apologies for not remembering the exact syntax, but you get the idea.)
PL/I's ability to use keywords as variable names and statements as
expressions allowed it to parse this correctly ... but I think everyone
would agree that anyone who actually wrote that code should be buried
alive, preferably face down and nine edge forward. It _IS_ possible to make
a tool excessively general, and doing so burns cycles and engenders user
confusion.

If folks really want a universal and excessively flexible metalanguage and
are willing to pay for it, let's point them back at SGML. That wasn't, and
shoudln't be, XML's design point.


______________________________________
Joe Kesselman  / IBM Research

Received on Wednesday, 7 June 2000 16:01:13 UTC