XML Processing Model WG

Meeting 69, 24 May 2007


See also: IRC log


Norm, Alessandro, Henry, Paul, Mohamed, Richard, Rui, Alex, Andrew


Accept this agenda?

-> http://www.w3.org/XML/XProc/2007/05/24-agenda.html


Accept minutes from the previous meeting?

-> http://www.w3.org/XML/XProc/2007/05/17-minutes.html


Next meeting: telcon 31 May 2007

Norm, Henry give regrets

Alex to chair

Open action items

Henry completed his action on p:wrap

Alex completed his actions wrt to steps in the latest draft

Parameter proposals

Norm: Henry and I discussed it this morning

Henry: In response to your mail, Norm where you asked how to feed a parameter to only one XSLT step.
... Wouldn't you have to change the pipeline?

Norm: Yes

Some discussion of how to make it work for arbitrary parameters

Richard: I think you could do it from the command line, but it might be messy

Norm attempts to summarize the differences between Henry's proposal and the alternate draft.

Henry: The reason I made another proposal is that there appears to be a lot more mechanism and new rules than appeared to be necessary in the 90% use case of the new mechanism.
... If we wanted grouped parameters to send to separate steps, then let's do just that.

Alex: One of the things you can do with Norm's proposal is avoid shadowing so, for examples, you can avoid having *all* parameters go to http-request as headers

Norm: Right.

Henry: That's true, there's no way to do that.
... There are two levels: first, that it would be reasonable to spec http-request to say that it could read parameters as headers by explicitly naming them.
... But we still use QNames, right?
... That's what namespaces are for. If you need to avoid name collisions.

Alex: We could augment Henry's proposal with some sort of exclude parameter.

Some discussion of how bindings would also work with namespaces and multiple steps of the same type

Henry: If, after suitable experience, we feel that we need some way of throwing away inherited parameters, we could add that.

Alex: I think if you want to write a rock-solid pipeline, you're going to wind up trying to deal with parameters by moving them all up to top-level optoins
... I'm worried that our current flexibility might allow user-specified parameters to cause XProc to go off the rails.

Henry: I'm somewhat sympathetic. The fact that we have as yet not said anything about parameters on the command line means we should give it careful thought.
... Even if it's implementation defined if they can be set on the command line.
... With respect to options, the situation seems a bit more clear. But for parameters it seems less clear.
... We're working on parameters against a background of assumptions that I think it was right for Alex to point out.

Norm: I don't think we can try to reach closure on this today, given how recently new proposals have surfaced.
... But I also don't think we can push out a new draft until we have resolved it, so that'll just have to wait a week or two.
... Anything else on parameters this week?

Nope; let's take it back to email.

p:position() vs position()

Norm attempts to summarize the current state of play, conceding that he's been convinced by Jeni

Henry: I convinced myself that it has to be a function, but I don't think it's the XPath position() function.
... It has to be a function because in the case where you have a step that accepts a sequence, the pipeline processor is only going to invoke the step once.
... It follows that there can't be a variable that has the binding because there's only one invocation so there's only one option to pass in the binding.
... Inside a step, the way an expression can tell you your position must be a function.

<Zakim> MoZ, you wanted to add some argument *againt* position()

Mohamed: The use of position() requires that we define a context and I don't want to have to define a context.

Norm: I don't think that makes sense, the XPath 1.0 spec says the context exists and has five things in it.

<alexmilowski> function vs variable really doesn't matter for establishing the XPath context... you'll still have to be clever if you want a clean API for your step implementors.... but I'm not going to argue against functions.

Henry: All we can do when mandating what components do when they evaluate XPaths is tell them what we make available in the context.

Richard: I'm now confused about how this whole thing is supposed to work. Henry is assuming that the step gets to go off and evaluate XPaths. I wasn't aware that that was how it was supposed to work.
... There's no way my steps can evaluate anything in the context of XProc.

Henry: There are certain options to certain of our steps who's static value is an XPath expression.
... these are expressions that the step must evaluate.

Richard: I was expecting the delete step to get run once for each of the matches
... Why does it matter if its a variable or a function?
... If you have a step that has a bunch of documents coming into it then in order to know inside the step what the position is implies that the implementation must be exactly synchronized with the input to the pipeline
... I agree that we're getting on dangerous ground, but I think we have to.

Some discussion of p:matching-documents

Richard: The case were it seems to be difficult is where you expect independent step writers to be able to determine which position in the sequence a document is.

<Zakim> Norm, you wanted to point out how p:position() doesn't work any better than position()

Norm: The p:position() function is essentially the same as the context position.

Norm: For a step like matching-documents -- for the first doc. in the sequence the context will be a context node: the document; a context position: 1; a set of variable bindings, function bindings, etc.

... for the next doc, change the first two to 'second document' and '2'

... I missed 'context size' -- it's 1, 2, 3, . . .

Mohamed: that approach to context size is why I don't like this proposal

... It means 'last()' is wrong

Norm: But you have to give _something_ , so it will be wrong one way or another

... the only alternative is to say you can't stream

Alex: You can't have a set of documents in XPath 1.0

Norm, Richard: Yes you can -- a node set of document nodes, e.g. <xsl:for-each select="document(a)|document(b)">....

Alex Err... I didn't say you can't have a set of documents in XPath 1.0

Henry: we're going to say that conformant step implementations of steps which take document sequences as input must use this context.
... We're not saying that they're evaluating against a set.
... When a step implementation of a step which takes sequence input evaluates an XPath against a context node which is one of the nodes in that sequence, then the rest of the context must look like this.

<ht> ... where 'this' is what Norm said above

Richard: Am I right in thinking that although you have this context you can't get out to other documents.

<richard> context size should be NaN

<alexmilowski> "Uncle!" +1 for context position and the postion() funciton.

Henry: Given the choice between saying you can't stream or saying last() will be wrong, I much prefer the latter

Alex: Is there anything bad that can happen by using position()?

Norm: I don't think so.

Alex: I'm fine with it then.

Mohamed: I still think that defining the context is a bad idea. Too many things will go against what people will find obvious.
... Node-set is very different from sequence.

<alexmilowski> straw poll ?

Mohamed: I think it's bad information to give to users to say that we use position() and the last() will be wrong.
... Ok, I think I'm beginning to understand.

Richard: What did we agree to for determining if you're the last document?

Norm: We didn't.

Richard: So we could set the context size to the largest possible integer.

<ht> ... or 1+plus the context position()

<richard> or +inf - who can tell whether it's an integer?

Henry: Straw poll: Can live with using context position: Richard, Alex, Alessandro, Henry

<ht> Rui, Norm can live with

<ht> Paul and Andrew pass

<ht> Mohamed against

Richard: The point is to ensure that position() never equals last()

Back to email, I guess.

Any other business



Summary of Action Items

[End of minutes]

Minutes formatted by David Booth's scribe.perl version 1.128 (CVS log)
$Date: 2007/06/07 16:10:52 $