Neil Graham wrote:
- At what stage during the processing of a document should XInclude
processing occur? Should it occur before DTD validation, namespace
processing, or after schema validation (if a document is schema-validated).
The REC itself is deliberately neutral.
IMO it's called for a more modular architecture: have a parser, an
XInclude processor, a validator, an PSVI constructor, an XSLT processor
and a processing pipeline. Unfortunately the APIs have yet to be hammered
out, so I'd think as a parser feature XInclude has to be happen before
validation. If someone wants to validate before XInclude, he can add
a separate XInclude processor at a later stage.
Abstractly, one could ask whether XInclude operates on a document's infoset
or on its PSVI. Does the spec say anything about this point, or does it
leave the question open, to be dealt with when a spec describing a
canonical XML processing model is created? By using SAX, one is at least
in some measure biasing oneself to a position on this: SAX provides a
version of the infoset, no PSVI.
The problem with "infoset" and also to some extent with "PSVI" is
that the standards are fuzzy, several other standards refer to it
with actually different stuff in mind, and everybody else uses
its own version of "infoset" anyway. Let's stick with SAX, we have
a formal documentation about the API and how it's supposed to work.
And I don't think a small project should start with designing a
PSVI pipelining API. Let's wait for DOM3, if it ever makes it into
software without causing a global network meltdown.
- In terms of the suitability of xml-commons as a home for a project like
this, can one imagine such a component being readily used as a utility,
independent of a parser?
Yes.
- You also wrote "It supports text and XML parsing"; I can't tell whether
the word "parsing" distributes over the conjunction.
This comes from the spec: there is an parse attribute which can take
the values "xml" and "text". In the first case the included stuff
must be XML and is formed into a tree before merging, in the latter
case just one text node is inserted.
What would that mean in terms of the XInclude spec?
These *are* terms of the XInclude spec!
BTW the XInclude spec is one of the more sick of the canon.
J.Pietschmann
