See also: IRC log
Mohamed gives regrets.
Norm: Schedule for the first week of November in Santa Clara.
Norm: Is anyone else planning to attend?
Alex: I'm planning to be there, it's local for me.
Mohamed: I'm on the fence.
Paul: I can't make it.
Vojtech: I'm still unsure. It depends on the status of my membership.
Henry: Try real hard, we'd like to have you.
Vojtech: I've made good progress on getting membership restarted. Now waiting on the final step.
Henry: I think there's been no feedback on this revised version.
Norm: Is the 4 Aug version linked above the revised version, or the original?
Henry: That's the most recent one.
Alex: The two sentences "given a
pipeline library document..." and "given a top-level pipeline
... I believe you mean the visited set.
... where you say "singleton set"
Henry: What I understand Alex to be saying is "Given a pipeline ... it is an error if ... against the background of a visited set being a singleton set containing DU."
Alex: Right. However you want to phrase that.
Norm: Ok, I think this would be fine, though I'm not sure I like having teh defn of bag-merger in a footnote.
<scribe> ACTION: Henry to make one more pass over the prose and insert it into the spec as a revised App G. [recorded in http://www.w3.org/2009/09/10-xproc-minutes.html#action01]
Norm: I don't feel strongly, but I think it should preserve the parameters as p:http-request does.
Alex: I agree.
Mohamed: They're incorrect in UTF-8, but they aren't incorrect in Unicode.
Alex: Is there a valid mapping from the code points to Unicode?
General agreement that there is.
Vojtech: My question was about encodings in general, not that specific one.
Norm: I heard some agreement that we preserve the values.
Proposal: The charset value (and other parameters) are preserved.
Resolved, see above.
Vojtech: What happens if you
specify an xpath context in p:when but you don't specify a
... In our implementation, the p:when is using the default readable port, not the binding from p:xpath-context from p:choose.
Norm: I think this is an edge
case that we didn't think of, so we just need to say what the
... Why did we allow the binding inside xpath-context to be optional?
Vojtech: I think we might have done it to preserve the default.
Alex: So this one uses the default context?
Norm: I think there are two
possible interpretations, an empty p:xpath-context either goes
back to the default readable port of the p:choose or it goes
back to the default on p:choose.
... Or we make it illegal by requiring a binding inside p:xpath-context.
Vojtech: Right now the spec says it works just like p:input, so it would get connected to the default readable port.
Norm: Making an empty
xpath-context go back to the choose would be redundant.
... So I think that boils down to two reasonable intepretations: the default readable port or we make it an error.
... For the 1 in 999,000 case when someone might use this, I guess that would be ok.
Mohamed: I think it's a bad idea,
when a user uses xpath-context in the choose, then I think we
should make the user be explicit in any p:when where they want
a different binding.
... I think we should forbid having an empty p:xpath-context.
Vojtech: I think I agree with Mohamed on this one.
Norm: Ok by me.
Proposal: Make it an error to leave the p:xpath-context empty.
Norm: I'll change the content model so that it's required, we don't need a new error code.
Norm: The JSON RFC doesn't define an XML encoding, it just defines JSON
Alex: The c:query step is for XQuery, not random queries.
Vojtech: Perhaps he meant that if the content-type on p:data was applicatin/json then it would be turned into XML.
Mohamed: I don't think the purpose of this spec is to convert all tree-like structures into XML.
Henry: I think this is an area where it's perfectly reasonable for implementors to compete. When we were doing the markup pipeline, it ended up being the case that it was appropriate to add a command line switch to upconvert STDIN from some format (like SGML) to XML.
Norm: Yes, and I think, I'd have
to go back and read carefully, that an impl could recognize
application/json and turn it into XML.
... Nope, I was wrong.
Alex: There's nothing in this message that seems to imply we're supposed to translate JSON into XML. We've already got ways to represent JSON in a pipeline, using c:data.
Proposal: Reply that you already can include JSON as text using c:data. If you want conversion to XML, you'd need an extension step for that.
Some additional discussion of Henry's use case.
Some discussion. General agreement that making the error dynamic rather than static would be very painful for implementors.
Alex: Changing the definition of a fundamental step is a bad idea
Norm: I think the rules we have are fine, the consequence of the rules is that for some changes, we'll introduce a new namespace or change the step name.
Alex: So that just means he has to rearrange the choose, right?
Alex: So the end result would be just a slightly different pipeline.
Proposal: Reject making the error dynamic, point out that the constraints are on future versions of steps with the same names, not future functionality.
Vojtech: I think there are two more questions. What happens if the schema changes so that some elements can contain new elements that weren't supported in V1.
Norm: Oh, so we add a p:xyz child of steps.
Vojtech: Not just steps, but also
in p:serialization, for example.
... or in p:document we add a new child.
Norm: I guess we could say that those are ignored. I have some reservations, but I can't articulate them.
Vojtech: I can imagine cases where this could cause problems. What if we wanted to add a new kind of instruction like p:choose or p:try. If you ignore it then the pipeline might not make any sense anymore.
Norm: Right so if we add p:map-reduce ignoring it would be all you could do but it wouldn't be the right thing.
Mohamed: I think it has to fail.
Norm: If we add new language elements then you can't write backwards compatible pipelines that use them.
Vojtech: If you introduce a new
builtin step then you could wrap it in a choose and use step
... No, that won't work because you have to know the signature.
Mohamed: The problem we have is that we have to compute a new dependency graph. Adding new builtin constructs just makes it not backwards compatible.
scribe: I don't find it too restrictive, because when we provide a new instruction perhaps we can provide a wrapper step for it.
Proposal: No, we're not going to ignore unknown elements.
Norm: And the last one is covered by the fact taht you're not allowed to declare steps in the p: namespace unless the URI begins with the right prefix.