See also: IRC log
Regrets from Henry
<scribe> ScribeNick: MSM
AM: have made a number of
formatting and stylistic changes.
... Table is compacted, I hope it's easier to read.
... The use cases now have identifiers which can remain constant across reorderings.
NDW: I'm comfortable with publishing this, although I still have questions about the table.
AM: well, the table should be either completed or dropped. it's valuable, but it's not worth delaying publication over.
NDW: recall that the use cases and requirements are not fixed and immutable from this point on.
MM: let's make it invisible rather than deleting it.
clarification: it's the table at the beginning of sec 5 we are talking about
<Norm> ScribeNick: Norm
MSM: I'd rather leave the table in place with a note to say that it's incomplete
Murray: Is that an offer to complete it?
<scribe> ACTION: MSM to draft a complete table; ETA: 21 Apr 2006 [recorded in http://www.w3.org/2006/03/23-xproc-minutes.html#action01]
We'll publish the document with the table, with a note that it's incomplete
<scribe> ACTION: Alex: add a note to the effect that the table is incomplete [recorded in http://www.w3.org/2006/03/23-xproc-minutes.html#action02]
Proposal: The WG has consensus to publish the Use Cases and Requirements document with minor editorial changes including a reordering of the Use Cases.
<scribe> ACTION: Norm to request first-public-working draft status [recorded in http://www.w3.org/2006/03/23-xproc-minutes.html#action03]
Publication date: 4 Apr 2006
<scribe> ACTION: Alex to produce a draft with reordered use cases by 29 Mar 2006 [recorded in http://www.w3.org/2006/03/23-xproc-minutes.html#action04]
Norm observes that last week Henry tried to outline Richard's position.
Norm asks Richard if he has anything to add
Richard: On reflection, the work
we did at the face-to-face was too complex.
... Having components with unread inputs or unwritten outputs spreads the conditional across the whole pipeline
... I'm not suggesting a simpler approach where the inputs/outputs/parameters of each side of the branch must be the same
... I'm still suggesting that the test be controlled by an XPath over a separate documents
... What if you want the two branches to have different inputs?
... You can put the inputs inside the conditionals.
... I've described this at the URL: http://www.cogsci.ed.ac.uk/~richard/pipeline.html
Alex: This is a lot like what I do now.
Richard: Unlike Alex, I
distinguish control structure from things that happen in
... I wanted to make the control structure explicitly part of the language.
... My aim is to find the smallest amount of control structure necessary.
... I think viewports, for example, don't have to be control structures.
Alex: Is this if/else or choose?
Richard: I've deliberately not answered that question. You could in principle compile from a choose down to an if/else as I've described here.
Norm: I think the question of
whether conditionals are part of the language or just
components is central.
... I'm leaning towards the idea that control structures should be in the language
Richard: I wanted to avoid the case where a component took two pipelines and a condition. This would be something like 'eval' and that makes things less transparent.
<Zakim> ht, you wanted to argue in favour of conditional-in-the-language position
HT: I want to endorse the
inclination that we should have conditionals in the
... Exposing conditionals as a language construct is important in order to allow the analysis of pipelines to be a sensible thing to undertake.
... Knowning that there are exclusive "OR"s about is a potentially important factor in analysis.
<MSM> Why not just say that the language does not guarantee either that every component will run, nor that components in the false branch won't run?
Norm: Anyone uncomfortable with conditionals as components-in-the language
MSM: Alex has argued in the past for the view that control structures shouldn't be part of the language so that you can add new control structures by adding new components
<ht> HST doesn't hear roll-your-own-control being ruled _out_, just exposing conditionals as part of the language _in_
Richard: That raises the question of side effects.
MSM: There are programming languages that say "under these conditions whether the side-effect is observable or not is undefined"
Richard: They say that about their basic conditional?
MSM: In particular, boolean conditionals may be evaluated in some indeterminate order, short circuited or not.
Some discussion of whether that's the same as the possible implementation of the else branch
Alex: I have a bunch of
developers using pipelines. I make no distinction between steps
and language constructs like choose and the viewports. My
impression is that they find this confusing.
... Implementation-wise, they're all components. But having conditionals and viewports not part of the "higher language" constructs is confusing to users.
... In the end, my choice to implement choose as a component is just an implementation choice, no more.
... I make sure the "else" branches don't run.
<MSM> If what is being proposed is "let us define semantics for conditional constructs as part of our spec", then I think I am OK with it.
I think that's what we're saying MSM
<MSM> As long as we're not saying "no component is allowed to do anything but read input and write output" ...
HT: I don't propose to rule out components that perform conditionals, just the positive aspect of making some conditionals part of the language.
MSM: I would have expected this
to be a question of what's defined in our spec, but we seem to
be going a little further than just putting it in the
... I want it to be a language construct vs. I want conditionals to be defined in our spec seems to be the issue.
Richard: Suppose that you didn't
have an "if" and you wanted to implement it as a
... This component would have to have inputs and etc., but also a description of the "if" and "else" branches.
... So if we said that's how it worked, we'd have to make it possible for pipelines to be parameters to components.
... This gives components a lot more power than we need to do the bare minimum.
... I don't want the basic, minimal language to have to support that complexity.
MSM: I guess I see some tension between what should be the minimum and what should be possible.
<ht> Richard contrasts fexprs and macros in Lisp
Richard: I'm quite happy with something that works like a macro, something that turns into some control structure using that. What I'm not happy with is something that effectively "evals" its arguments to perform a computation
Alex: I allow you to embed chunks of a pipeline in your components.
<MSM> Richard, can you type in the Lisp keyword / technical term you were pronouncing?
Alex: An analogy here is XSLT instructions. If you want to write your own instruction, you need to be able to embed other instructions inside it.
<richard> it's one the kinds of EXPR (expression)
Alex: In XSLT, that's completely implementation defined. It's an issue if we need to go there or not.
<richard> you won't find it in scheme!
Alex: That's one way to give extensibility without having to reinvent the world. You can invent your own crazy conditional with side effects or what have you. But it does put additional requirements on implementors.
<MSM> Ah, that may be why it sounded unfamiliar
<richard> scheme takes the approach i'm advocating, it has macros but no eval-like mechanism
<richard> And my point about viewports is that they can be implemented in a macro-like manner without any "magic" components
<scribe> ACTION: MSM to setup bugzilla for us [recorded in http://www.w3.org/2006/03/23-xproc-minutes.html#action05]
Norm: The ability to define and reuse a sub-pipeline in the same document is still an open quesiton
Richard: If we accept the
separation of the language into a conceptual part and a syntax
for expressing that, then this issue falls into the second
... I think we're all agreed that conceptually you should be able to put a sequence of steps anywhere that a component can occur
Norm: Do we all agree that conceptually you can put a sub-pipeline anywhere you can put a component?
Some "yes"es, some "not sure"s
<MSM> it sounds right, but I keep worrying about a catch I'm not seeing.
<alexmilowski> (sorry... stepped off...) yes
Norm: Raising the question to the conceptual level seems to reduce the urgency of answering the question
HT: But it does raise the urgency
on the question of whether we're talking about an abstract
syntax with a nearly-isomorphic concrete syntax or if we're
designing an virtual machine with a concrete syntax that might
be quite removed
... The time is coming when we're going ot have to make our minds up
Richard: I think you're right,
but the answer will be neither of the two extremes
... The abstract primitives aren't a virtual machine, but they're designed to make programming easier.
... They're more like the primitives of lisp that will be extended by writing functions and macros on top of them
Alex: Exact time is
... But I'm willing.
Norm: Anyone uncomfortable with having another f2f this soon?
<MSM> which week in August?
Richard: I'm not sure I can make it, but the date doesn't have much bearing.
HT: I think it's a good idea, I will try to come.
Norm: Please respond in email if you prefer 2-4 Aug or 14-17 Aug