Lists Home |
Date Index |
From: Rick Jelliffe [mailto:email@example.com]
>But XML is broken for that anyway: because standalone="yes"
>is voluntary and probably ignored by recipients anyway, you cannot
>guarantee that if you need to take advantage of XML's expressive
>power (e.g. use entities, attribute defaulting, external subsets, etc)
>the receiver will end up with the same infoset.
I agree. This puts XML back to where SGML was during the
>The result is this cart-before-the-horseism that vomits itself up
>on XML-DEV every few microseconds: "I want to get rid of
>the parts of XML I don't use". Rather than "Lets fix the parts
>that are broken", which is constructive and congenial.
I mostly agree. Some assert the need to "bless" subsets
they are already applying by formal specification (soap)
The danger asserted is that "subsets growing in the wild" will
lead to broken interoperability. The counter-argument is
that attempting to get all of the requirements for different
groups into one subset or even a few is as likely to cause
interoperability problems. The third position is, Works
As Designed. Don't touch. I think the first position is
suspect or at least unsupported. The second position
depends on how processors are required to recognize
instances that the sender assumes rely on the subset
features at the receiving node. The third position isn't
hard to support, status quo, because it translates to
use the specification as designed and ensure the receivers
do the appropriate filtering.
> 1) Deprecate Well-Formedness for public specs, documents and
>data interchange (it is a niche thing only suitable for editors);
After five years of the XML core telling the public that well-formedness
is the sine qua non of XML, that will be a hard sell however
rational it may be. They don't usually admit they are wrong.
> 2) Give a formal name for a headless XML, in which a DOCTYPE
> declaration is an error (and make it a little simpler: don't check
> naming rules or normalization) SOAP can use this.
> 3) Give a formal name for an unvalidated, guaranteed-infoset XML:
> all parameter entities fetched, defaults applied, entities defined,
> IDs typed, but content models not checked. XHMTL can use this.
Agreed but as you point out, the trick is the guaranteed infoset.
>Solutions like "get rid of attributes/entities and everything will be well"
>look like armchair speculation to me: the trick is not to come up with
>a tidy idea, but to come up with a practical one that moves us forward.
I agree with that wholeheartedly, but I am old enough to remember systemts
that tried that approach to SGML. They became dust buckets.