See also: IRC log
Mohamed gives regrets.
Murray reviewed it, but suggested that someone with more implementation experience should also take a look.
<scribe> ACTION: Alex to review before the face-to-face. [recorded in http://www.w3.org/2007/10/04-xproc-minutes.html#action01]
Paul: Discounted hotel reservations for the face-to-face expired yesterday.
Norm: Yes, and hotel rooms are
ahrd to come by that week; good luck!
... Who's coming?
Henry, Norm, Paul, Alex, Richard is tentative
Richard: Comments to the comments list, but many of them turn into long discussions.
Norm: I think it's ok if we only have admin on the WG list. New last call comments should go on the ocmments list, I think.
Mohamed: Where are we on the test suite?
Norm: I'm generating new ones as I work through my impl; I encourage others to send them in using the test format.
Some discussion of the test suite.
<scribe> ACTION: Norm to arrange for the tests on tests.xproc.org to be in Subversion so that anyone can update them. [recorded in http://www.w3.org/2007/10/04-xproc-minutes.html#action02]
Norm describes the issue. Is symphathetic.
Henry: I'm not sympathetic. I
wish we could undo some of the architecture changes that XSLT
has *already* imposed on us.
... I think parameters have ended up swamping what was otherwise a fairly clean design.
Henry: Basically, my view is that
this problem is the thin edge of a long and complicated issue
dealing with inputs of an unknown arity.
... I can think of at least three ways to solve it and we'll need to do it in V.next. I'd rather not do it at the last minute for XSLT when we'll need to do it later.
... I'm just skeptical of the whole thing; the cost-benefit ratio is hugely out of whack.
Alex: What XSLT really needs is a
way to associate a name with a arbitrary input.
... We've removed arbitrary inputs so we'd have to invent some much more heavyweight mechanism to do that.
... I'm with Henry in some ways; I wonder if we should consider the ability to bind some number of documents with some number of names.
Henry: I think we can do it in
V.next in any one of at least three ways as I outlined in
... It'll take time to think them through.
Norm and Henry discuss how some implementors allow for XSLT parameters to be bound to files.
Norm: You can work around this with a generated stylesheet, of course.
Henry: I don't think now is the time to design the general solution.
Alex: It allows documents in under the covers.
Norm: I don't know what you mean by under the covers.
Richard: XSLT says that its
implementation defined how parameters get passed in from
outside. AFAICT, there's no requirement accept any kind of
... Mine only allows you to pass in strings. For things that you run from the command line, there's no obvious way to pass in anything else.
... If we say this is going to work, that means you're saying your XSLT implementation has to be able to accept things that aren't strings.
... Otherwise you'll get interoperability problems.
<Zakim> ht, you wanted to ask about the api issue
Henry: At the moment if I write
parameter name=foo select=.//pset[@foo=baz] I get an element
node which gets stringed and I get a string.
... How will I tell if there's an attempt to pass something other than a string.
Richard: We'd have to change the language, saying that things selected don't get turned into strings so you have to use string(). There are several possibilities.
Straw poll: Do you support allowing non-string parameters?
Results: Y 4, N 4, C: 1
Norm: I don't see a consensus to make a change to the spec.
Propose: Not in V1.
<scribe> ACTION: Henry to reply to the commenter. [recorded in http://www.w3.org/2007/10/04-xproc-minutes.html#action03]
Scribe notes that the numbers got messed up.
Richard: Like Henry, I'd like
very simple pipelines to be simple. So I don't want to have to
put in input and output statements in those cases.
... I constructed a hard case, but I don't really want to throw out the mechanism.
Norm: I can see that, but I want a solution that doesn't require the depth of analysis that Richard's example would require.
There's some question of whether this only applies to pipelines in libraries.
Richard: Henry suggested that you might only be allowed to do this in the top-level pipeline. But that doesn't really help.
Alex: Couldn't we just outlaw circularity?
Richard: Yes, we could say that
the only circumstance where you can't do it is the one where
you can't figure it out.
... But simple recursion is allowed.
Alex: We could say that circularities must be explicitly broken. It's not straightforward to detect, but we coudl have a blanket statement about it.
Henry: The way it ends up working
is that we revise the statement about how pipeline outputs
default to say that they default to the output of the last
contained step provided that that can be uniquivocally
... Expanded slightly if necessary to make it clear.
Richard: If the last step isn't a call to a pipeline that doesn't have explicit outputs.
Henry: I don't think that stating that carefully will take more than two or three sentences.
Richard: I'm not sure that this will turn out to be the case.
Alex: Yeah, I'm not sure it can be done locally.
Henry: My proposal is fairly radical: maybe it'll be confusing but, I would say: you have to have explicit not-calling-other-pipeline steps in every place where you have to look.
Richard: It mustn't be a pipeline.
Henry: Yes. That's the minimum to declare victory.
Alex: I'm confused. Every step
has a declaration...
... Think about a processor that's building step declarations. You'd have to flag the ones that come from pipelines.
Richard: But you haven't determined the declaration of another pipeline
Alex: In this case, yes, but ...
<scribe> ACTION: Henry to craft the prose to cover this case. [recorded in http://www.w3.org/2007/10/04-xproc-minutes.html#action04]
Norm: In principle, do we want to do this?
Richard: This is something who's effect is completel system dependent.
<Zakim> ht, you wanted to talk about / conversion
Henry: I think there are two or
three details here, but I'm happy to wait until we have a
... What about serialization?
... What about the working directory?
... We need to do / conversion in file paths
<scribe> ACTION: Norm and Alex to craft a proposal. [recorded in http://www.w3.org/2007/10/04-xproc-minutes.html#action05]
<richard> this is an optional step, right?
Norm summarizes the thread.
Henry: I think the defaulting semantics is a little tricky to specify clearly. But I do absolutely want to have pipelines with fixed inputs.
Richard: That's quite different from defaulting input then you don't need any syntax at all. You just make the first step read from a constant source.
Henry: That inclines me even
further towards the position that this is more confusion that
it's worth. We don't have any use cases for it.
... I'm particuarly unhappy with Norm's answer to Richard's question about how it works when you call a named pipeline.
Norm isn't sure if he was clear
Alex: I can see this going either
... Do we say what happens in the pipeline case. You can have a step that has an unbound input. But pipelines can have bindings that go with them. When you use a named pipeline, it can use that binding.
Richard: If you put a call to a pipeline in, and you don't ...
Norm: Maybe it's too confusing.
Richard: The only trace where this even arises is in the place where it says that pipeline inputs can be defaulted.
Norm: I suggest we take this one to email and leave it for a week.
Henry: I encourage Alex to write up what he saw as the use for that.
Alex: Now that I think about it some more, I don't think we need this at all.
Norm: Do you change an absolute xml:base to relative?
Anyone think we should change xml:base attributes?
Some discussion of whether or not a step with an xml:base attribute can have a different base URI.
Alex: I think this step should make this explicit.
<scribe> ACTION: Henry to attempt to improve the prose for p:add-xml-base [recorded in http://www.w3.org/2007/10/04-xproc-minutes.html#action06]