See also: IRC log
-> http://www.w3.org/XML/XProc/2006/04/06-agenda.html
Accepted.
-> http://lists.w3.org/Archives/Public/public-xml-processing-model-wg/2006Mar/0069.html
Accepted.
Any regrets?
No regrets given
Norm submitted the transition request and got approval.
Norm submitted the publication request dated 11 Apr 2006
-> http://www.w3.org/XML/XProc/docs/WD-xproc-requirements-20060411/
Norm proposes: who can commit to attending?
Rui: unsure; Norm: yes; Alessandro: unsure (also for Erik); Henry: yes; Mohamed: unsure, but probably; Andrew: no; Paul: yes; Henry says Alex, Michael, Jeni say yes; Richard: unsure
Tally: 7=yes; 5=unsure; 1=no
Proposed: we will meet in Toronto on the dates specified.
Accepted.
Henry points out that formally we can't decide to do this, all we can do is ask the CG to allow us. No one seriously expects the CG to say anything but "yes"
<scribe> ACTION: Norm to get this into the CG calendar [recorded in http://www.w3.org/2006/04/06-xproc-minutes.html#action01]
The chair asks if we have a volunteer editor.
Norm volunteers.
Norm warns that editing, chairing, and taking minutes may prove too much for one person. He suggests that Henry, Michael, and perhaps others may get called upon to take minutes.
Norm wonders if rotating the minute taking is the right thing.
-> http://www.cogsci.ed.ac.uk/~richard/pipeline.html
Norm wonders if it makes sense to consider Jeni's proposal of passing URIs instead of documents around.
Henry doesn't think passing filenames (URIs) around is really going to work.
Henry: Jeni's story could be read
either way (as passing filenames or using filenames)
... I'm still interested in thinking about this in terms of
giving local names to things (inputs/outputs) and using those
names to refer to them.
Norm: With respect to giving names to local things, did you mean inputs and outputs
Henry: Yes, but I also mean
static resources/secondary inputs/whatever you want to call
them.
... In the pure piped ontology which Richard offered at the
f2f, there is a qualitative difference between an XSLT
component that has one pipe coming in and one going out and a
parameter which is the name of a static stylesheet and an XSLT
component that has two pipes coming in and one going out.
Norm: My proposal, to unify
these, is to allow a shortcut for a "read-from-URI" component
that attaches to the stylesheet input pipe.
... Do you see any problem with that approach?
Henry: No, but I'm still swinging back and forth between thinking of the pipeline runtime as a resource manager and the pure dataflow model.
Norm: I tend to swing back and
forth as well. I hope that we don't have to pick one.
... The resource manager view has the problem of dealing with a
pipe that contains a sequence of otherwise anonymous
docments.
Henry: Local names are just
conveniences. They are single documents or doucment streams as
appropriate.
... The only interesting case is when you use a non-local
name.
... In MT pipe, the convention is if you use a #-ed name, it
just plugs together. If however it's a primary input/output
connection and you give it a non-#'d name, then you get the
single document if it's one or the last document if it's a
sequence.
... There's no clear answer to the question of what does the
label mean if a sequence of documents is addresed by a lable in
the resource-manager view.
Norm ponders the idea of a resource manager that handles a collection
Norm suggests the fragment question as a next reasonable point to discuss
Norm describes the situation as an interoperability issue (since some implementations might not even notice and others might fall over)
Henry: It's entirely reasonable for some implementations to fall over if you pass anything other than real XML "Document"s .
Norm expresses a view that either the pipeline author has to fix it, or the pipeline engine has to fix it.
Norm: I suggest for V1 that we say it's the pipeline author's problem
Henry: works for me
Norm wonders if the rest of the group agrees
Rui: I believe that just documents is good enough for me.
No one objects.
Proposal: only XML 1.x documents (proper Documents in the XML sense) pass between components; if you need to pass something else in your pipeline, the pipeline author has to wrap and unwrap as necessary.
Accepted.
<scribe> ACTION: Norm to begin summarizing the points of consensus [recorded in http://www.w3.org/2006/04/06-xproc-minutes.html#action02]
Norm: Richard proposed a single standard conditional that takes a document and an XPath, is that enough?
Norm wonders if the WG thinks that's all we need
Henry: If you need something
really complicated, you can write an arbitrarily complicated
computation that produces a document.
... And then switch on that conditional.
Proposal: The pipeline conditional component is XPath expression over document. If you need more, build a document and use that?
Accepted.
Murray: We could put the statement in terms of the available processor.
Norm suggests that won't work
Henry: Several
possibilities:
... 1. Use a convention for naming XPath expressions that are
the values of attributes; xpath=, xpath1=, or xpath2=
... A question we have to address is, are we going to subset
XPath in order to guarantee that it isn't hard to do
this.
... I think we should subset.
Henry proposes that the XPath expressions that you can use in conditionals to being ones that are streamable
<MoZ> ht, what about attribute minxpathversion="1.0" to parse
Henry: bearing in mind again that if you need the full power of XPath, then you can use an XSLT processor to build a document ove rwhich a streaming test will succeed.
Alessandro: We can say that the expression is going to be XPath 1 or 2 and the engine can analyze the expression and stream if it wants to. Otherwise, it can just run a full XPath engine
Norm asks if Alessandro is opposed to the subset
Alessandro: Yes
Henry: The problem I have is that it puts a huge burden on the implementor for functionality that we've already determined most users won't ever use
Norm: My concern is inventing the subset. Do you think the XML Schema schema subset is appropriate.
Henry: Not quite.
... The thing that's missing is [@foo]
Norm: So the tradeoff is inventing a subset or using an off the shelf processor. Or is the problem really the analysis for streaming?
Henry: The analysis is hard. Can
we float a trial balloon and examine the possibility of using
that in V1?
... Two issues for V.next are, should we accept any XPath or
should we require people to detect a certain class as
streamable
Norm: Why detect? Non streaming will always work.
Henry: One of the crucial things
about a viewport is that the XPath expression be streamable
because that's how you deal with documents that are too large
to read into memory.
... Conditionals are not the only place where XPath expressions
are going to turn up.
... It seems plausible to try to tell a consistent story.
Norm: I agree, I don't want XPaths on different components to have a different flavor.
Henry: Maybe the regex for detecting the streaming subset isn't too hard.
Norm ponders the plausibility of using a regex
Henry: I think a regex could detect the the Schema subset
Norm objects to the idea of *requiring* a processor to support streaming
Norm proposes that we take the XPath 1/2/subset question to email
Alessandro: Did we have a conversation about setting up bugzilla
Norm: Yes
-> http://www.w3.org/Bugs/Public/
<Alessandro> Excellent
Norm will make the XPath question an issue as an example