See also: IRC log
Face-to-face in 2007?
Henry: I'll check if the W3C is ok with having a WG that runs for 18 months without a f2f meeting.
Richard: The rescheduled Tech Plenary is partly responsible.
Paul: We had a f2f in August, and I thought we didn't plan to have a lot of f2f meetings.
Norm: If we had a meeting, I think I would propose meeting in Europe, perhaps colocated with XTech in Paris or around the XML Summer School in Oxford in July?
Henry: Edinburgh would be happy to host.
Norm: Let's leave the question open for today, but we should decide soon.
Another public WD in February?
Norm: If review of last week's draft is generally favorable, I'd like to propose a new public WD in the next week or so then seriously consideer what obstacles prevent us from doing a Last Call in March.
a. Is the defaulting story right?
Henry: Not quite: I can't find
where we talk about the situations in which we write p:output
with a binding.
... That's where there ought to be something about defaulting.
... You ought to not have to have an output.
Norm: No, I didn't consider the defaulting case where you don't specify an output at all.
Henry: I think we should.
... In particular, in 2.4, it's easy to read this as if it only worries about inputs.
Norm: I'm going to take it as an editorial suggestion that 2.4 needs to say more about output bindings.
Henry: Consider also 4.2.7: "The result of the p:for-each" ...
Norm: And the defaulting story is...
Henry: A container instance that has one of these inward/output facing outputs...
Norm: If the container doesn't
declare any output bindings, then it's output is what would
have been input to a putative step that came after what is in
fact the last step.
... Henry's email is: http://lists.w3.org/Archives/Public/public-xml-processing-model-wg/2007Feb/0014.html
Henry: Consider figure 3
... Note that this pipeline has no p:output
Richard: Is this what we want? Don't we want to be able to refer to the outputs.
Alex: I didn't think we were going to default outputs
Richard: I was assuming you would have defaults for outputs, but you'd have to declare the outputs.
Alex: Where is this helpful? On viewport, for-each, etc.
Henry: Can we look at figure 4
... It should be clear where the input from the XSLT step comes from
Richard: How do you declare a choose that has no outputs then?
Henry: I'd be sorry if that obscure case required me to put outputs in every p:when
Richard: I'd been assuming that you'd have to declare them, you just wouldn't have to bind them.
Henry: I think that would just irritate users every time they had to do it.
Richard: We can consider whether or not you have to declare them and whether or not you have to bind them separately.
Norm: You could use a DevNull component to make a component that has no outputs.
Henry: I really want to write
... The flow is completely defaulted.
... I think the 90% case is where people will want to default the entire flow
Richard: In order to get the
abbreviated syntax you want, we're now having to change the
... It used to be clear that if you didn't have any outputs, there weren't any.
Alex: So what's the name of the output of the choose?
Norm: There isn't one. There's just an anonymous output that's only accessible as the default input to some following component.
Henry: In the non-defaulted
syntax, I think I'd be a little happier if I did have an
explicit indication that the component didn't have any
... In particular, consider: unconnected outputs are not a bug, that's just fine. So I'm considering the case where someone writes a choose and, ignoring the defaulting discussion, does not put any outputs on any of the whens.
... But the last step does have an output port.
... That seems to me to be a slightly odd situation. Any container that ends with a step that produces output in a container which doesn't have any outputs, I'd expect that to at least be a warning situation.
... That leads me to feel that I'd be perfectly happy to require components to explicitly state that they don't produce any outputs.
Richard: I had not considered this question.
Norm: Ok, let's leave this for a week and come back to it next week.
Henry: In updating the DTD, I
realized that there's an interesting difference between XPath
context and input.
... Both have a binding, but only input allows a select attribute.
Henry: My first thought was that
that was ok, but then I thought about it a little longer and
the following example occurred to me:
... I ended up thinking I'd rather write it that way.
Norm: It seems reasonable to me.
Some discussion of what happens when sequences are identified
Alex points out the difference between the way select is proposed to work on xpath-context and the way it works on input.
Henry: The paragraph that
finishes "set of nodes is wrapped in a document" needs
... If we accept my proposal, I think I do want select on xpath-context to work differently.
... Maybe we need to think about this for a while too
Norm: Yes, probably.
c. Do we want to do something similar about for-each/viewport?
Norm tries to outline why the single input to viewport/for-each could or perhaps should also be anonymous.
Henry: This is the thing which
drives the process but isn't something that gets referred to
... It's perfectly reasonable to have a different input. I think the simplest thing to do would be have an input with no port name.
Alex: In the case of viewport, isn't the input often going to be defaulted?
Henry: I agree with Norm's analysis that giving it a name makes it available in which is odd.
Alex: But you can't read from it inside.
Richard: You could bind to it, just as you can bind to the inputs of a pipeline.
Alex: Is that how this works?
Henry: If not, then it voilates the principle of least suprise at the very least
Norm: Taken together I think
these three issues stand in the way of a public draft.
... Let's please resolve them next week
Henry: Can you add my updated examples and the DTD to the draft?