See also: IRC log
Norm gives regrets, Henry to chair.
Henry completed ACTION-2008-08-14-01, the XProc charter is extended until 31 Dec 2008.
There are only a few left, and few are coming in. Yay!
Norm: p:choose currently requires p:xpath-context followed by p:variable*
Henry: Right, I think we should
allow them in any order. I could, for example, want to bind a
variable that I then want to refer to in the
... It'll be clearer if the variable can come first.
Mohamed: How about just swapping p:variable* and p:xpath-context?
Henry: That's sufficient for my use case, but I actually think that the order in the spec is the one that folks are usually going to want to do.
Norm: I don't see any compelling reason not to do what Henry suggests.
(SCRIBE MISSED MOHAMED's COMMENT)
Vojtech: It's already the case that p:for-each and p:viewport require the context to come first.
Mohamed: Variable can be very useful, but they make the spec very, very unclear.
Norm: We could relax this restriction in the future, since it wouldn't make any pipelines invalid.
Henry: The thing that I didn't
realize before is that p:xpath-context unlike p:input doesn't
have a select attribute.
... I guess I withdraw my suggestion.
Proposal: The suggestion is withdrawn, close with no action
Norm: This is encryption and
decryption, I don't think we've made any progress here.
... I've agreed to talk to the XML Security WG on Tuesday, 16 September.
<scribe> ACTION: Norm to ask the Security WG to invite the rest of the WG (anyone who's interested) [recorded in http://www.w3.org/2008/09/04-xproc-minutes.html#action01]
Norm: Let's hold off on trying to deal with encrypt and decrypt until at least after that meeting
Norm attempts to summarize the p:pipeinfo vs. allowing non-step namespaced elements everywhere.
Norm: Anyone want to reopen this discussion.
Michael: Allowing foreign
namespaces anywhere is fine if the basic rule is that it's
there in the document and what you can do with that document is
what you can do with it.
... if I put stuff in a foreign namespace somewhere in a pipeline, can anyone access it?
Norm: You can always do introspection on the pipeline docuement, but I think most extra stuff is either for implementation-defined purposes or for other applications, not for the pipeline itself.
Proposal: Reject the request, leave the status quo.
Henry: Yes, I tripped over this too when revising the DTD. It's not a substantive change, it's just an editorial correction and I think we should do it.
Norm: Yes, I think so to.
Proposal: Ask the editor to make the editorial improvements necessary
Henry: The short answer is, we
considered and rejected this. I think the better answer is, you
can just leave out the whole input from your identity
... The wG has decided that when you write p:pipe, you have to say everything. Anything else is just too confusing.
Norm: Allowing different defaults just seems too confusing.
Henry: Missing attributes on pipe having one convention and missing input having another is just too likely to cause confusion.
Proposal: Reject this proposal, stay with the status quo.
Henry: Looking back at the
original problem Mohamed observed, I think there are three
... 1. It's an error; we can say that either p:option declarations must specify a default or be required (or we could say the same at runtime)
... 2. The implementation can't access a value for the option.
... In this case, there will be an error; I was thinking of atomic steps first.
Mohamed: So I don't think we need a @required.
Henry: I don't agree, there are lots of cases where the option isn't going to be used at all.
Mohamed: But I think we need a special function to test if the variable is bound or not.
Henry: 3. The default for default
is the empty string.
... But I really don't like that.
... It follows naturally for XSLT, because the default comes from teh content, but we don't do that.
Norm: I'm inclined along the lines of Henry's #2. I think it's ok if options have no value. In the case that Mohamed points out, if it wasn't a test case, we'd suggest putting required='true' on that option.
Henry: I think I agree. I'd like
to preserve the distinction between defaulted and not
... If we go this way, the prose about the XPath context will have to be clear that the in-scope options do not contain options that were optional and not supplied.
... It's worth noting that the phrase "in-scope options" isn't defined.
<ht> Yes, there is a definition of in-scope bindings. . .
Henry: I'll look to see if there's a place to say that processors may/must allow atomic step implementations to distinguish between the two cases.
<ht> Ah, we have a notion of "specified options"
Mohamed: The difference between Henry's and Jeni's proposal is that Jeni's was "fail fast": you must specify a default or required; Henry's is "fail later".
Norm: Yes, I prefer Henry's form as well. So it's a dynamic error if a non-defaulted, non-required option has no binding and a reference is made to it.
Henry: Yes, but the right way to say it is that the variable bindings in the XPath context are determined by the *specified options*.
Norm: For all the atomic steps that we define, that have options that are neither required nor have default values, we should say what it means if they aren't provided!
Vojtech: The serialization options for p:store, for example.
Henry: The defaulting of
serialization options is a specified in the QT serialization
... I think this also means that we must say somewhere that implementors of atomic steps must be given a way of determining whether optional, non-defaulted options where specified or not.
<scribe> ACTION: Henry to review the optional options and suggest what we should say about each of them [recorded in http://www.w3.org/2008/09/04-xproc-minutes.html#action02]
Proposal: Allow optional options, it's a runtime error if you attempt to refer to an unspecified optional option.
Norm: Anyone going to the plenary, remember to register and book your hotel
No other business heard.