See also: IRC log
Jim: I'm finding Thursday challenging.
Norm: Time of day is to suit west
coast USA and Europe.
... I don't like Monday's for meetings. I could do Tue, Wed, or, reluctantly, Friday.
Jim: Maybe Cornelia would be able to make it if we moved the meeting time?
Alex: Henry has a complicated schedule too.
<scribe> ACTION: Norm to send email polling for a different/better meeting time. [recorded in http://www.w3.org/2013/01/17-xproc-minutes.html#action01]
Norm: Alex, Henry, Norm, please report on your actions in email; suggest which ones are overtaken by events.
<jfuller_> zip and unzip ... still draft
Alex: What's the status on the Processing Model document?
Norm: The status is we're in Last Call and have comments from reviewers that we haven't addressed.
Alex: Does it make sense to try to bring this to closure as fast as possible, even if other stuff is more interesting.
Norm: Yes, that's the case.
Alex: I'll try to look at this after the call today.
<scribe> ACTION: Norm to make XML Processor Profiles document an explicit agenda item [recorded in http://www.w3.org/2013/01/17-xproc-minutes.html#action02]
Norm: Jim, I think you said you'd take a stab at it.
Jim: I worked on the zip/unzip
step instead. I did do some work, but it's not uploaded
... The link above has notes from our f2f meeting. I'll try to get a revised document online by Tuesday for next week.
Norm: I think we were talking about Vojtech's inheritence proposal.
Vojtech: The main idea was that
maybe we could get rid of parameters altogether by inventing
some sort of inheritence mechanism.
... Because we don't want to lose the "magical" behavior that they inherit from the pipeline.
... So options could possibly inherit from the calling context or maybe the other way around.
... This option "propagates down" if it's used. From a nested scope, for example.
<jfuller_> initial Norm link on fixing parameters - http://lists.w3.org/Archives/Public/public-xml-processing-model-wg/2012Sep/0014.html
Vojtech: It would be different to
our current parameters but it would allow for things like
having an XSLT step that has an option called parameters that's
declared that it inherits from above. And by some kind of
naming convention you could actually still use this magic that
you're used to.
... The main idea would be to get rid of parameters concept alltogether and replace it with options.
... And with maps, the "parameters" option could be a map and those key/values could be inherited.
Alex: Presumably we'd still need the with-param kind of construct.
<jfuller_> last meeting minute param discussion - http://www.w3.org/XML/XProc/2012/11/01-minutes.html#item04
Vojtech: That's a question. Or some way to extend or override the map options.
Alex: I think you could make that
work in a number of ways. With-param could just be on the XSLT
... You could have a default parameter option and with-param could just modify the map associated with the default parameter option. We could support an option name.
Vojtech: We'd have to be really
careful with with-param and with-option and the order that we
evaluate these things in. You could have with-option
dynamically building the map and with-param updating it.
... That whole inheritence mechanism is simpler but also introduces some complexity, some side-effects that could be surprising if you don't really know what you're doing.
Alex: Are you imagining the inheritence goes by the data-flow or by the structure?
Vojtech: On the declare-option
you'd say inherit=true and then when you use the XSLT step, if
the parameters option is not set, it would look in the
upper-contexts to find an option or variable with the same name
and use that value.
... You could also do it the other way around, on a pipeline you can declare an option called parameters and inherit=true and it could propogate down.
Norm: That's the same functionality, it's just a case of where you put the label, right?
Vojtech: Well, yes, perhaps, but I think it's safer in some sense to propogate down because the author has more control. The other way around, using the steps could introduce surprises.
Alex: Minus a whole bunch of
details, I like this idea. I never liked the distinction
between options and parameters. I didn't want to do it, but we
didn't have an alternative.
... So I like this approach, but we need to make sure we get the details right.
Vojtech: In my idea, it's all
based on the option names.
... Any option name could be inherited.
Alex: We need to decide if inheritence is a special thing that you turn on, or it's the default and you have a way to turn it off.
Jim: I'd go for always inherited.
Vojtech: If it's always turned on, it could help with unbound optional options.
Alex: I think I'm in favor of always inheriting too.
Norm: It only applies to steps inside compound steps so it's not too dangerous, I guess.
Vojtech: If you really want something to be unbound, you should be able to make it explicit.
Norm: But we don't have a syntax for that.
Alex: But we'd need one, I
... We need to look at the details.
Vojtech: I'm willing to dive a little deeper.
Norm: I think that would be great.
Henry: I'd like to see the same examples, if at all possible.
Vojtech: The same?
Henry: The same ones that Norm used in his proposal.
Henry: I'm confident that either one of these or almost anything else would be an improvement.
Norm attempts to summarize.
Henry: We had this in the old
markup pipeline engine, and the motivation in that case was
output type. You could allow internal XSLT steps to set output
type HTML and have that actually work at the end of the
pipeline because it got passed along with the metadata.
... The bad news is that you have to think through the semantics about what kind of steps preserve this and what kind don't. It's not at all obvious, for example, that an XSLT step preserves the metadata. It may not make sense for the metadata to propagate.
... That may just be too hard a question to answer.
Alex: It may also be the case that the pipeline author wants control over that.
Henry: But we don't want them to have to think about that if they don't want to.
Alex: But like inheritence, it
might be the case that there needs to be a default and a way to
specify the alternative.
... Output content type is a good example.
... I'm a little confused, because we had at least two proposals and then the f2f, and I'm not sure where we've gotten to.
Norm attempts to explain again: we started with Vojtech's proposal for binary which required passing along the MIME type and we generalized that to document metadata.
Alex: Right, but my proposal didn't require that.
The chair expresses some confusion about the state of the binary proposals.
<scribe> ACTION: Norm to put review of the binary proposals on the agenda for next week [recorded in http://www.w3.org/2013/01/17-xproc-minutes.html#action03]
Alex: Two things in my proposal are that XML is the only thing that flows through the pipeline. The binary is identified by reference. Everything is treated uniformly. It doesn't matter if the resource manager has a magic URI or if you point off into the ether of the universe
Norm: Ok, my bad.
Alex: I'd like to see some good use cases for document metadata.
Norm: Fair enough.
Henry: The other one that I recall that we recall that we used this for is preserving the character encoding. Which I can't remember if you can get from the infoset or not.
Alex: There's a whole bunch of things that can be grouped into: how I got the original data vs. parameters needed for serialization. There's a whole bunch of metadata in there that's related. People might want to preserve how they read the data in how they write it and metadata is a good way to transport that form one place to another.
Henry: The reason we did this for character encoding is that we invented an encoding that allowed us to preserve entity references.
Vojtech: I have a question: all the metadata values or attributes that you mentioned look like magic metadata fields. But applications can also set things, like timestamps and labels, yes?
Henry: Oh, absolutely, pipeline authors could come up with their own metadata.
Jim: For non-XML it allows us to add metadata without changing the documents. It's like implementing a state machine without having the state in the documents.
<scribe> ACTION: Norm to make sure document metadata stays on the agenda [recorded in http://www.w3.org/2013/01/17-xproc-minutes.html#action04]
Alex: Who's going to XML Prague
Jim: Everyone but Henry
<scribe> ACTION: Jim/WG to discuss XProcathon at next telcon [recorded in http://www.w3.org/2013/01/17-xproc-minutes.html#action05]