Re: numeric-sequence predicates

> That's an interesting suggestion. How feasible it is depends on:
>
> (a) how you define "concisely"
> (b) whether the level of "conciseness" needed for truely
>     interoperable implementations, doesn't require "drilling down"
>     as deeply as the formal semantics does anyway.
>
> Currently the XQuery documents are organized such that
> understanding
>
> (1) XQuery 1.0 and Functions and Operators (as user reference)
> should suffice to /use/ XQuery
> Of course XQuery 1.0 should do as much as possible to explain XQuery
> concisely enough, such that users aren't surprised (and I *think*
> the current WD achieves this already pretty well).
>
> (2) XQuery Formal Semantics and (to a certain extent XQuery
> datamodel) should facilitate truely interoperable implementations.
> Users normally should not be required to dig deeply into it.
> The reward for implementors digging deeply enough will most
> likely be that they arrive more easily at conformant implementation.
>
I think all these formalism and semantics is great if they are equally
convincing; like you mention, it will definitely need more prose
to stop the reader from making guesses and assumptions.

True interoperable implementations will have to be achieved over
time, loads of deployments and feedback. I agree that FS is
one of the factor that will help achieve interoperability but its
not the only factor. We need to short the feedback loop; this I think
we can achieve by ensuring XQuery 1.0 gets out on time;
FS can still track and trail till up to 2.0. Every serious implementer
will look into FS details but they have to be convinced that what's
in FS makes sense, and realistically, their accurate interpretation
and implementation will take time to realize even if they are
absolutely perfect.

> Granted, the formal machinery of the formal semantics document
> (type inference, value inference, and mapping rules) may not be
> everybody's game, very much like it is not everybody's game to
> read (and deploy and further develop) source-code. In fact, it
> IS very much like an open source implementation (with the notable
> exception that it can be used for own implementations without
> any restrictions).
Agree, chances are, one would arrive at the same conclusions in
FS from the surface syntax but what if the surface syntax has to
change ? And chances are they will, I am not sure how
much, but they surely will after deployment. I mean, look at what
is happening to XPath ? Is the FS layer operating at such an
atomic level that it will be resilient to changes from the above
syntax ? Will implementing FS in version 1.0 facilitate or
impede subsequent changes ?

I find it somewhat puzzling that it is often suggested that
formalism is daunting; IMO if the various conceptual frameworks
and data models are crystal clear, it will actually be fairly
straight forward to follow the semantics. Indeed, it will be
great if there is a set of FS that we can totally rely upon to
implement blind !

>
> What are the alternatives?
>
> Well, one can restrict oneself to prose,
> or quasi-prose (lot's of tables, if-then-else-"english", etc.). There
> are
> specs out there, which have done that and they've taken some heat
> for being "too complicated". In addition, I have some doubts whether
> pure prose really scales to more subtle aspects of XQuery such as type
> inference,
> type subsumption, variable scoping, context-binding, namespace handling,
> whitespace-handling in element construction, such that implementations
> are interoperable. As a borderline case consider the infamous
> "valid derivation by restriction" in XML-Schema, which to my knowledge
> hasn't really led to fully interoperable implementations so far.
>
It remains a fact that
XML Schema: Formal Description WD is dated 25 September 2001
XML Schema Part 0,1,2 reached recommendation in May 2001

We need to balance, given the limited time and talented pool.
FS has its value but lets put it into realistic perspective.
Interoperability
will have to be achieved over time and over many feedback
processes and not just FS alone. Sure, if FS can be done
accurately and in time and with all the specs in sync, by all means
do it, that would be ideal, but for now, I do not think its a
realistic assessment.

> There are other specs out there, which use seemingly simple prose,
> are short (not to be mixed up with "concise"), but have generated
> lot's of confusion, uninteroperable implementations, etc.
> So a bit of formal machinery does have value. Nevertheless, the XQuery
> 1.0 FS
> could of course use a bit more prose at places.
Agree and concepts must be concisely clear first. If the editors
can still breath properly under all these stresses, then I think they
know where to strike the balance between seemingly prose and
conciseness.

Unless XQuery has a very clean and elegant model, just like
RELAX NG, (may not be fair to draw the comparison though),
then I think the FS approach may be the right one despite the limitation
of time.

> The observation that the formal semantics currently isn't fully
> synchronized with XQuery 1.0 WD, is to a large extent due
> to the general difficulty of synchronizing an implementation
> with a spec; and we're certainly working hard to better synchronize.
> In addition, working out the formal semantics has more than once
> generated design revisions for XQuery 1.0, typically filling holes,
> and generally leading to a more simple, more robust language
> (very much like implementations are the ultimate reality check
> for specs).
Sure, and XQuery may have multiple conformance levels, we
have not dealt with DTD and non-validated XML, and we can
press a sync-button during each iteration and to reach recommendation
by summer (my own speculation) ? Like I mention previously,
I think you have an NP problem on hand to solve; in a few more
iterations in relatively short successions. This might be intriguingly
challenging but is it worth the effort ?

Can all specs reach recommendation by summer ? Prioritise and
I believe WG can get a robust and solid XQuery 1.0 out
with good user acceptance and XQuery will be a success.
Too ambitious and I think WG can potentially loose control
of the outcome.

"Abstract XQuery 1.0 from FS"

> (speaking for myself)
me too :-)

Received on Wednesday, 13 February 2002 22:00:38 UTC