Minutes F2F Meeting, London, 4-6 August 2004
Agenda
-----------
http://lists.w3.org/Archives/Member/member-ws-chor/2004Jun/att-0020/Agenda08042004_0.txt
Attendees
--------------
Martin Chapman
Steve Ross-Talbot
Carine Bournez
Yves Lafon
Nickolas Kavantzas
Peter Furniss
Monica Martin
Jeff Mischkinsky
Charlton Barreto
Bijan Parsia
Robin Milner
Kohei Honda
Nobuko Yoshida
IRC Logs at:
http://www.w3.org/2004/08/04-ws-chor-irc
http://www.w3.org/2004/08/05-ws-chor-irc
Scribe
---------
Meeting scribes were: Charlton Barreto, Monica Martin, and Jeff Mischkinsky
Past Minutes
------------------
Minutes of previous meeting http://lists.w3.org/Archives/Member/member-ws-chor/2004Jun/att-0015/Minutes06152004_0.txt
Steve recalls that exception handling issues were discussed in last meeting but can't find a record in the minutes.
Steve proposes to adopt minutes but remove his notes.
Minutes passed subject to SRT taking out his notes
[ed note: what minutes are being referred to as minutes from 20th july have not been distributed, we also have not approved 6 and 13 july minutes:
6 July 2004 http://lists.w3.org/Archives/Member/member-ws-chor/2004Jul/att-0004/Minutes07062004_0.txt
13 July 2004 http://lists.w3.org/Archives/Member/member-ws-chor/2004Jul/att-0008/Minutes07132004_0.txt
]
Review of Agenda and Meeting Goals
-----------------------------------------------------
Three goals for the F2F:
1. Reduce issues by 50%, Most issues fairly small in scope and easy to address
2. Endpoint projection
3. Behavioural type system that is lock-free and leak free
2. Rely on Kohei, et al for pi-calculus theory
Approach for F2F
Focus on endpoint generation - language long enough for us to think about this
Focus on model checking
Focus on ease of use
only five proposals at this stage
(in nick's absence, XInclude was incorporated)
1. Bi-directional Interaction
2. Composition
3. State Alignment
4. Repetition
5. WSDL 1.1 support
Clarification
1. Level of abstraction
2. QoS features
3. Relationships between other standards (where we fit in the scheme of things)
When we come to endpoint generation, there are certain SLAs that need to be adhered to for certain protocols (TWIST, BPEL, etc.)
David not available at this F2F - we will need to decide upon Templates in CDL
Nick - any comments on the Editors' Draft; Steve - it just came out last week
Agenda changes - move endpoint generation to Wed pm; make room for editors draft discussion
robin will be here today; kohei and yoshida will be here tomorrow
endpoint projections
Editors Draft Discussion
---------------------------------
http://lists.w3.org/Archives/Public/www-archive/2004Jul/att-0032/cdl_v1-editors-jul24-2004.pdf
http://lists.w3.org/Archives/Public/www-archive/2004Jul/att-0031/cdl_v1-editors-jul24-2004-with-revisions.pdf
Editors' Draft - non-language construct references to 'participant' were changed to 'parties'
Martin - should we then refer to the language construct as 'party', then?
participant - this is type information, not instance information
explain difference between participant and party
one participant for one process
contraint on process compositions
role is 1:n WSDL interfaces
can have many channels between parties in an interaction
participant - e.g. type of buyer and seller
Nick - one participant with one role - points to one role; one participant to many roles, channels point to role instances
Martin - my point is that participant/party is a type
Nick - then role is a type and we need to decide upon that, file an issue, and make the changes
Steve - proposal: in all places in CDL where we are dealing with an entity which is a type, the word 'type' s/b appended to the entity name - e.g. roleType, participantType
Nick - since I converted this from 'fooType' to 'foo', I'm not going to change it back. :-)
Martin - editors to sort out; is there a separate thing to change 'participant' to 'party'
participant/party types in an interaction are both located at the same domain of influence
Nick - 'perform' - super optimized form of 'interact' at the end of the day
Steve - represent parties in an interaction as a single particpant - e.g. buyer and seller are pointing to a single location, in essence
Steve - hint or guide for implementors to provide this 'aliasing'
Nick - perform - connect things together - these endpoints have to be free/available - if cannot make the connection, there should be some sort of warning, etc.
Steve - in order for two parties to be aliased, one must be able to 'perform' the aliased action
'these two things are logically colocated'
Nick - aliasing the performing choreography to the performed choreography
Nick - variables are associated with roles
it exists at the location, at the role
Steve - say we have a participant type, with three roles; with a perform, we can map alpha->beta between A & B, and gamma->delta in B & C
where A, B and C are roles
Martin proposed simplification in specifying variable-role association
Individual variable unless all the roles are in the same participant
Nick this won't work because we won't be able to provide the nested visibility
(correct me if I have this wrong -C.)
(correct my earlier entry - nesting and visibility -C.)
Nick - inconsistencies found in the draft - bet. declare and define
Steve - chors name the relationships over which the chor will operate over - association of roles; can variable definitions refer to any roles, or can they refer to roles in the relationship
NEW ACTION: Martin to do an UML model
Steve - what visibility to variables on roles have?
Is this a model checking issue?
Nick - what is happening with the Primer? Martin - won't talk about it until closer to the end of the process.
SRT - roles for variable definitions can only be taken from the relationships that are taken from the chor
SRT - when you're doing an alias in a perform, is it going to reference a chor that it is in? It can be defined in a nested chor? What variables can appear in an alias in a perform - one of them m/b defined in the chor in which the perform is called, while the other does not
SRT - variables in a perform - dominant (one 'doing' the perform) and submissive (the one being performed) - the dominant m/b in the same chor as the 'perform'
SRT - definition of 'free' in this instance is the variable that is defined as submissive
Martin - change the element name 'free' to something more semantically meaningful
Martin - i propose changing 'free' to 'alias'
SRT - combination of the role attribute and the variable are what provide this aliasing
Martin - why are we restricting in how we perform aliasing - we should be able to alias other elements
SRT - doesn't like 'The role attribute aliases the roles..." sentence in the spec - we're not aliasing the Roles
SRT- closer to say something closer to the following sentence
SRT - the reason to have the role there is to function as a key; this and free are identified by variable and role; this/free is aliasing from/to; free is used b/c it has to be a free variable in the performed chor
Nick - if it is not free, we cannot do perform; to Martin, aliasing can encompass non-free variables as well; SRT - I think 'bind' is an infinitely better term to use
SRT - If they're not in the same participant, we can't do this anyway
"The roles within a single alias element must be carried out by the same participant " This makes the alias a tau-based interaction which because it is within the same participant is hidden - hence the tau.
Martin - the variable defined as free must be defined as a free variable in the performed chor
Nick - we should add action item to add this rule to the definition in the spec
Nick - m/b we can use word 'shared', but agree that we should not use 'free' because it c/b confusing in this context
SRT - let's change 'alias' to 'bind', leave 'free' as is, and add the rule per Nick's action
NEW ACTION: Editors to update perfrom as follows: 1) alias -> bind, 2) rule added to perform: the free variable must have free set to true in its definition, 3) rule added to variable definitions: variables with free set to true are required to be be bound in a perform in before they can be used.
(sorry if I'm adding redundant actions here)
SRT: I propose that we adopt the editors' draft we have before use as the current working draft
SRT - This forces the group and others to raise issues against this draft and not another; I would rather this was done before this meeting concludes.
SRT - We will review this before the end of the F2F and make a decision then.
Monica - Does our scope extend beyond Web Services?
Martin - It is a web service language, but we're designing it so that it can extend beyond web services (David writes the drafts as completely orthogonal to web services)
Nick - the CDL model is a generic model - it s/n be bound to web services
Yves - no model per se in web services - it is not SOA
Monica - 1.2: interoperability and interactions required to ensure conformance/compatibility between services within one business entity.
NEW ACTION: Editors to change last bullet before section 1.4: change "Compatibility with other Specifications", "Specification Composability"
Monica - because this says "one business entity" - we had another discussion about the use of "business entity"
Martin - I don't think this affects anything, but we can change the language to ensure that we're not missing that this can address services across business entities
Martin - choreography tells us how interactions must conform, as a spec
Monica, Nick - let's take this sentence out and reverse the change
NEW ACTION: Editors change last section in section 1.2 - reverse change in latest editor's draft - change "interoperability and interactions
required to ensure conformance between services within one business entity" to ""interoperability and interactions between services within
one business entity"
Monica - "In addition a record MAY be defined..." - what would you record if you had no exchange?
If this is a busn entity beyond the context of CDL, this makes sense, but only in the context of CDL....
Interacts
-----------
Nick - in an interaction, and exchange is optional - this means an exchange is still taking place, but not where would declare what variables are being used; exchange indicates use this variable here - there is an exchange of some information - Nick had this with exchange having a '+' (but it didn't make the draft)
If we do have WSDL - we look at the operation and not doing anything special - exchange is optional, although we still do record - exchange is '*'
(this implies a 'vanilla' WSDL operation)
Martin - exchange breaks down an interact into its constituent messages
Martin - In a one-way WSDL operation, why do we need the addl exchange? Makes sense for request-response.
Nick - is there a mech that an interact request has been sent (and not yet received) - is there a way to say this with an assign?
Martin - synchronicity/handshaking depends on alignment
Nick - internal proposal - use preferences instead of priorities - how to basically figure out how to override the state - convenience for users to code things, but doesn't remove race conditions issues - is there any way (given that we can only manage race conditions) to use preferences to address this
SRT - priorities are more effective because they are simpler, and in every example given w.r.t. race conditions, it has been about what to select and what to discard
Nick - one way to resolve dist choice is preferences/priorities
Nick - what if someone wants to code the protocol himself and not use this convenience mechanism
Nick - whether we can do an assignment inside the interaction, atomically?
BREAK for lunch
After lunch - complete interactions, continue on to issue resolution
channel resolves to WSDL through the role
client - from Role - has no WSDL - WSDL is at the to Role
Martin - syntactically, we don't have to repeat the from and to Roles
Martin - WSDL models the server side; right now we have the channel representing the server side
Nick - not necessarily - it the endpoint projection was solely WSDL, I would agree; however, we have a WSDL file which solicits a request from service - a WSDL which is a requestor - it is a sort of client
SRT - the Role for the channel is what truly represents the behaviour - it is the service
SRT - so an interaction requires that the to Role has an interface
Martin - channels become nothing more than web service addresses - with more description around it
Robin: In pi-calculus, we can name processes but do not carry around these; at runtime we avoid having to distinguish between names
Martin: we have message sets - with names - which you can receive
Martin: endpoint - collection of message types - this is a role
Nick: bookstore is a role - channels w/b Barnes & Noble, Amazon, etc. - only one role - bookstore, but these services are instantiations of the bookstore role
Martin - e.g. Role: Books (a wsdl) with operations browse, buy, sell, comment; Role: purchase (another wsdl) with operation pay
Martin - until now (WSDL 2.0), we had no inheritance in WSDL
Martin - people want WSDL fragments to be reusable in different cases; Roles represent this aggregation facility
Martin - input channel, output channel, fault channel - reply-to more of a binding level item
Nick: what if I have an exception on a response?
Martin: there's no fault on a response; if fault occurs, have retry
from WSDL 2.0 perspective this does not matter
WSDL 1.1 silent on fault destinations, etc.
reply-tos are an example of where we have to break down into exchanges
Martin - WSDL allows us to define: operation(in) - one way msg, operation(in, out) - two way msg; HTTP: synchronous; Reply-to: addresses asynchronous scenarios
Can write WSDL broken down into one-way messages
Break down into: WSDL X:operationA(in), X:operationB(in); Y:operationB(out)
People generally write WSDL the first way, not the second
In WSDL 1.1, no normative way of passing these channels around, in WSDL 2.0, it is more normative with destinations, etc.
Nick - doesn't matter which language we use, it is all handled under the covers for CDL;
SRT - To an interact then is both a normative and a broken down, the exchange construct in the interact handles this; without the abstraction, interact becomes a mini-choreography in itself
SRT - ability of a choreography to model message exchange patterns; do we need in interaction the same expressive power as we need outside it?
Jeff & SRT - we've taken WSDL 2.0 as our message exchange pattern 'base'; all else in this domain are seen as something to be addressed with choreography.
Nick - why do we still have client characteristics in WSDL? They are never used....
Jeff and Bijan - there is a history with these where they are still desired
Nick - unless we have interacts from the other side, from Role always has empty interfaces; when do we have non-empty interfaces in from Role?
Martin - interacts don't (need to) have two WSDLs
Nick - if so why do we have solicited request
Martin - to support output only messages
SRT - is it a subscribe then (solicited request)?
Nick - so it is a notification?
SRT - something has to tell us what to generate
Nick - spec is very clear on how we get channelVariable, etc.; Martin - it isn't very clear on how to send messages - how does A get B's address from - config issue of where we fill in these details
Jeff - there's nothing special about outs/notifications - they are supported
SRT - from CDL perspective, we don't care if have operationA(in), operationB(in, out) or X:operationA(in), X:operationB(in); Y:operationB(out)
If have no constraints - what do we generate
Nick - why do we need to have the client side perspective - e.g. Books Role - newBook(out)
SRT - if the WSDL doesn't exist at all - you have not constraints - need rules to determine what to generate
Nick - spec only supports two MEPs - just like BPEL
Jeff - WSDL 2.0 defines the MEPs - you can define your own as an extensibility point
Bijan - need to define more complex MEP so that we don't have to define own extensibility points
Martin - CDL is constrained - interaction has one input message and can have one output message - not a sequence, etc.
Bijan - symmetric complaint - WSDL and CDL should be more in sync - either constrain them both or allow custom extensibility points
Robin - can i have a server being a client for another interaction - hierarchical procedure calls
Robin - I don't get the need to limit the exchange
NEW ACTION: SRT to raise MEP extensibility problem with the WS-CG
steve responds - wsdl2.0 promised to end last call after the next (i.e. Sept) choreo F2F and this was minuted at the CG.
http://lists.w3.org/Archives/Member/chairs/2004JulSep/0048.html
WSDL Last Call period ends 4 October 2004 and we need to comment as a group ( see later)
steve again: So the next CG is early Sept and the next choreo F2F is late Sept.
SRT - to summarize the interactions discussion today
1) we have broad agmt that we need InputChannelVariable, OutputChannelVariable, FaultChannelVariable
2) We will support existing MEPs only - WSDL 2.0 + WSDL 1.1 - an interact itself m/b able to have any support any MEP supported by WSDL 2.0 + 1.1 - with ISSUE on SolicitResponse/Notification
3) Tell WSDL WG that extensible MEPs will not be supported in WS-Choreography
We make certain assumptions about MEP consumption patterns that we can support (e.g. one input message, one output message)
4) Requirements for rules governing endpoint generation in the absence of WSDL (there are occasions to generate this way and we need constraints); rules for generation in presence of WSDL
NEW ACTION: We need InputChannelVariable, OutputChannelVariable, FaultChannelVariable; we're looking for volunteers for proposals for this; we will log this as an issue
NEW ACTION: We need proposals for requirements for rules governing endpoint generation both in the absence and presence of WSDL
Left to cover for Day 2: Import review; Concrete Bindings/QoS (Charlton) (proposal); Interact (Record and State Alignment and Priority); Repeat (proposal); Getting rid of state (proposal); Perform (proposal)
(correction:
Removal of state aligment proposal
------------------------------------------------
http://lists.w3.org/Archives/Public/public-ws-chor/2004Jul/att-0070/Removal_of_State_Alignment.pdf
Nick - 1) What exactly do we mean by 'state', 2) Alignment is more than state, it is ensuring that both participants are in sync - aligning exchange information in one such example
SRT - only one example in the proposal - direct from the CDL spec itself
SRT - example - records the state of the interactions
Nick - no difference between state and variable from the model POV
Martin - why a difference in the language
Nick - no difference whatsoever
Martin - state variables are discussed in interaction (sec. 2.5)
Nick - 2.4.2 page 19 of editors' draft
SRT - are there additional semantics that we can add to interactions that eliminate the need for state alignment
Martin - we need to differentiate between state variables and state alignment
Nick - 2.5.2.1 can be deleted - if not deleted, need to properly define it per requirements
Martin - if we delete this section, we remove the state changes from interact, and we use record statements in the interact to manage state changes; let's proceed and delete section 2.5.2.1
NEW ACTION: Editors to remove text for 2.5.2.1 and clean up spec to remove references to state variables (just call them variables)
Type Systems for CDL
------------------------------
Kohei Honda & Nobuko Yoshida presented work on a type system for cdl
pi-calculus - very fine grained representation; very small subset of Java
Kohei - example of how to represent boolean in pi-calculus
[Yoshida Nobuko - use cases - interactions, exceptions, other CDL constructs using pi-calculus]
[Yoshida Nobuko - use cases - deadlock detection - modality + pattern matching to check for locking scenario within compositions (and thus provide serialized access to resources)]
[Honda Kohei - pi-calculus not complete isomorphic]
(correction - ... not completely isomorphic ... -C.)
[Honda Kohei & Yoshida Nobuko - fine granularity of pi-calculus enables one to easily map from CDL to lower level languages and back]
Robin - important not to make assumptions about isomorphism
Peter - timeouts do not change possibilities but probabilities
Robin - we need a way to represent relationships between timers
[language - primitive] in comparison to [pi-calculus - discipline]
Robin - choice construction in CDL is a session?
Nick - what if I only have a send and receive - what do I have to structure this as a session?
Nick - choice and interact essentially provide session discipline for you in CDL
SRT - session types - given that a session is some well defined fragment in CDL, and type checking - we would take all fragments of CDL and perform causality checking
Nobuko - check session type is correct (pattern-matching type), or use causality based check (i.e. check all things once, or divide)
Robin - type systems in general - can think of a type, e.g. T ::== (some channel name followed by Truth construct) - regard types of predicates for running a program - expect to have logical conjuntion of types - does it make sense to have logical conjunction in pi-calculus? Where does pi-calculus sit with this, will logical connectives be added to pi-calculus type system?
Nobuko - w.r.t. intersections, it is already in there, but with respect to certain other, more complex connectives, it may be beyond the scope of pi-calculus; we can describe certain more complex connectives using behavioural types
Bi-directional Interactions Proposal
-------------------------------------------------
http://lists.w3.org/Archives/Public/public-ws-chor/2004Jul/att-0070/Bi-Directional_Interactions.pdf
SRT - prioritization of interactions within choice w/b based on lexical order - therefore no changes w/b required to CDL notation
Peter - too tightly focussed - reason for these collisions being the result of one another - is because the protocol is driven by the global model - have collision case and side X is allowed to do A - what is happening is that there is a state change on one side or the other - and need to make one collision win over another is because we're not allowing an action to go back - can cancel an order which has not been performed (buyer/seller scenario) - seller
Peter - two party protocol is four party protocol because there is someone driving it from each side
SRT - in a protocol where there is a crossover or a race condition, someone has to decide from a modeling perspective what to do about it
SRT - we need to prioritize - need explicit notion of how to make priority in such cases
Nick - looking at this the way it is proposed - are we saying that we need a protocol that implements distributed choice?
Martin - no; Nick - how do we implement this; Martin - this is embodied in choice; Peter - need this for multi-party case in addition to two-party protocol; Nick - we need an agreement protocol; Peter - this is what choice provides
Peter - choice handles this for 2-party case, but how will we handle multi-party choice?
SRT - can't adopt proposal unless address multi-party cases...?
Kohei - should we not use priority and use more explicit approach such that multi-party case can be addressed?
Kohei - mixing priorities and exception cases makes things more confusing
Martin - need determinism to address this
Nick - we need to look carefully at crafting something useful that can address each of the cases - simple race condition cases experienced in TWIST, FIX, etc. - we're not there yet in terms of addressing this in a simple and flexible manner
SRT - send it back to editors, to address how to manage multi-party cases; need to find a shorthand which provides clarity
Monica - how do we handle external prioritization?
SRT - need to figure out best way to handle coding of such situations within CDL - it really cannot be external
SRT - the question becomes 'do we support what essentially what is a bad habit or not?'
Martin - should we actually model how we manage such cases in the human way, or should we reapproach it from the automated way? This is our challenge.
Peter - [describes FIX cases and how they address these]
SRT - I propose to send this back to the authors - this will open up a number of other discussion - how to we want to allow people to build bad protocols, or provide them guidance on how to do this in a well-mannered way
SRT - need some threads/discussions going on the issue of distributed choice
Martin - either we enforce unidirectional interactions between two parties or attempt address these race condition issues
SRT - m/b good for type system to tell us that a race condition exists
SRT - having some sort of validation for choreography for such issues - e.g. deadlocks may occur in this choreography
Martin - w.r.t. this proposal - the challenge is still open
Robin - it has to be deterministic on both sides - unique decision - we need rules for this, they will be pretty grotty, but there are necessary
SRT - financial markets represent this well - "word is my bond" decision by a trader to make a transaction - it is done once that decision is made, and all other decisions branch from this
Robin - but not every case is as clear - what about cases where there isn't such a deterministic fashion of approaching this?
Kohei - impression is that it may be good to make this explicit - types may make this suggestion - whether it is a case of 'judge'(?) or priority, we will need to determine
Martin - challenge put to the mailing list - one response so far - challenge is still open - need to provide proposals so that we have something to consider to address the issue(s)
NEW ACTION: Steve to reissue the challenge to remind the group to respond; SRT will formally reissue it to Jean-Jacques
Repetition Proposal
----------------------------
http://lists.w3.org/Archives/Public/public-ws-chor/2004Jul/att-0070/Repetition.pdf
Martin - example essentially indicates "repeat bar until foo happens"
Martin - second syntax - essentially a form of guard - allowing a sequence of things "any number of foos followed by a bar"
Nick - why not use a choice around the work unit? isn't this essentially a choice
SRT - can look at this that way
SRT - we need a composite way to represent this
SRT - want something more declarative and powerful to address this
SRT - how do we set conditions - to handle when this is a combination
Martin - shall we separate out repetitions from work units?
Martin - separate as into a separate construct?
Robin: strange not to have the repeat condition inside the repeat clause!
NEW ACTION: We agree that we should separate repetitions into a separate construct. We need to sent this back to the authors and
explore as a group how to approach this, including expressing conditions on both variables and interactions.
WSDL 1.1 vs WSDL 2.0 (aka Concrete Bindings and QoS)
----------------------------------------------------------------------------------
Proposal: http://lists.w3.org/Archives/Public/public-ws-chor/2004Aug/att-0001/wsdl11v20.pdf
Martin - Critical sucess factor - work with WSDL 1.1, BPEL 1.1. WSDL 2.0 going into last call
Jeff - Requirement for solicit response. If not, if we bind to WSDL 1.1, it will be to BP versions of those specs
Many of BP requirements have put us closer to WSDL 2.0; only reason not to do it - solicit response
Nick - areas of the specification where we have to specify message parts (1.1, not 2.0) may reflect difference. But we should do this 1.1.
Martin - would anyone object if our spec allows for Basic Profile 1.0/1.1 to be supported in addition to WSDL 2.0.
Nick - how much work to do both.
Conditionally accept given due diligence.
NEW ACTION: Charlton to look at impact on our language of supporting BP (WS-I Basic Profile) 1.1,
SSBP (Simple SOAP Binding Profile) 1.0, AP (Attachments Profile) 1.0
Duration Proposal
-------------------------
http://lists.w3.org/Archives/Public/public-ws-chor/2004Aug/att-0002/Deadline.pdf
Nick - A duration based timeout when the interaction has started.
Jeff - No global view since we have no global clock.
Martin and Jeff - absolute timeouts are straightforward enough assuming that the global clock (or simulation thereof -C.); but how can we define a relative timeout?
SRT - we have two issues w.r.t. timeouts
1) scope of the clock
assume a maximum skew
propagation of the timeout
how to handle the window of opportunity presented by skew and propagation
2) Exception handling
3) Propagation of Exceptions
Nick - we need forms of agreement - we have the same time, expire together, etc.
4) How do you guarantee that the information that the propagated exception carries is available to the system
(Alignment, in other words)
5) properties of mechanism to ensure alignment is achieved
Martin - we will describe the properties of the protocol, but not define how it is to be implemented
SRT - in order to achieve exception propagation, we would need to define an alignment policy for interactions for state and exceptions; and this exception alignment mech w/h the same requirements as state alignment in interacts.
[ed note the following was in the irc log but I dont understand the context - seems to be covered above
Bindings proposal - Charlton Barreto
16:25:36 [scribe]
WSDL v1.1/2.0 differences and direction
16:25:36 [scribe]
Differences in MEP
16:25:36 [scribe]
Reference: http://lists.w3.org/Archives/Member/member-ws-chor/2004Jul/0014.html
16:25:38 [scribe]
http://lists.w3.org/Archives/Public/public-ws-chor/2004Aug/0001.html (WSDL)
16:25:39 [scribe]
http://lists.w3.org/Archives/Public/public-ws-chor/2004Aug/0002.html (deadline)
16:25:41 [scribe]
Reasonable to include v1.1 support in addition to v2.0.
16:25:43 [scribe]
This supports BPEL which creates an issue for CDL.
16:25:45 [scribe]
I don't see it as onerous to support both.
16:25:47 [scribe]
Proposal provides examples on differences in either.
16:25:49 [scribe]
Chapman: What in the CDL language does this affect?
16:25:51 [scribe]
Barreto: It doesn't affect interacts or our level of MEP.
16:25:53 [scribe]
Chapman: We have a charter (v2.0) and a goal to support BPEL (v1.1).
16:25:55 [scribe]
WSDL v2.0 is going into last call.
16:25:57 [scribe]
May have an impact on generation however. When we get to binding
end of ed note]
Issues list/categories
-----------------------------
Martin: We need to have the list of issues - editorial (closed) and other
issues along with each ED draft.
Import
------
All the import issues were all resolved at the last f2f. Apart from
checking the editorial correctness, there are no more technical issues wrt
import outstanding.
Interacts
---------
Issue 627 - resolution
No support for WSDL 2.0 MEP extensibility point.
Support for MEPs supported in BP 1.1
Support all the varieties of WSDL 2.0 in, in/out.
NEW ACTION: Open new issue to investigate the solicit-response and notification in
WSDL 1.0 and WDSL 2.0
Issue 666 - why specify message type in interact?
The intent was to exploit the fact that channel is a first class concept,so
that the CDL level should express when only a channel is being
exchanged. The current syntax does not capture this. You have to do an
type "investigation" to figure out that the type being used is a channel
type and not another "user-defined" type.
Proposal is to make the messageContentType be a choice between a (sub)type
of Channel or something else. - UNAN
NEW ACTION: editors to define the specific syntax for a choice on messagecontenttype
Discussion about reorganizing the type hierarchy to make the info type be a
generic of either a channel type or "another" qname.
Scheduling
---------------
Next f2f is in nyc, hosted by Greg/Novell, 28-30 sep
NEW ACTION: everyone -- discuss on next con call: how far are we from LC "at the
5000 foot level" - i.e. are there major features that we are missing, are
there major features that cannot be cut. If so, what are they?
Charleton: don't think we should be dropping anything major. Not sure if we
have quite enough time.
Yves: should ensure that things like formal proof are possible. Need to
ensure that we can do end point projections.
Bijan: Since this represents new work that needs experience, should focus
on getting it out with current feature set.
Martin: BPEL projection and demonstrate use/applicability in a vertical.
Steve: wants to do a reassessment of current state of issues list based
upon the discussion at this meeting. There are lots of issues that could be
closed.
There will be an "informal" call next tuesday 10 aug to review the issues
list.
NEW ACTION: Yves to categorize the issues in bugzilla to add in our categories -
editorial, interact, etc.
Next formal call on Aug 31.
Editors can use the time slot for the "off" weeks. Issues processing call set for next tuesday (10th Aug)
NEW ACTION: Editors to produce and publicize the list of changes made in the latest
ED draft.
NEW ACTION: Editors to evolve it in the next few weeks with goal of enabling the WG
to adopt a new WD at the next formal meeting 31 Aug.
NEW ACTIONI: Review and produce LC comments on the WSDL 2.0 docs by the end of the
next f2f. Need some to coordinate the activity. Chairs to put on agenda
for 31 aug.
What are the CR requirements?
Not yet defined. Possible levels of testing
1. CDL doc validator
2. Generate WSDL
3. Generate implementation (some execution language) 3. Generate WSDLs
4. Interop between generates implementations
AI: Chairs to verify planned LC date with Team
Thanks to Steve for arranging/hosting the meeting.
Summary of New Actions
-----------------------------------
NEW ACTION: Martin to do an UML model
NEW ACTION: Editors to update perfrom as follows: 1) alias -> bind, 2) rule added to perform: the free variable must have free set to true in its definition, 3) rule added to variable definitions: variables with free set to true are required to be be bound in a perform in before they can be used.
NEW ACTION: Editors to change last bullet before section 1.4: change "Compatibility with other Specifications", "Specification Composability"
NEW ACTION: Editors change last section in section 1.2 - reverse change in latest editor's draft - change "interoperability and interactions
required to ensure conformance between services within one business entity" to ""interoperability and interactions between services within
one business entity"
NEW ACTION: SRT to raise MEP extensibility problem with the WS-CG
NEW ACTION: We need InputChannelVariable, OutputChannelVariable, FaultChannelVariable; we're looking for volunteers for proposals for this; we will log this as an issue
NEW ACTION: We need proposals for requirements for rules governing endpoint generation both in the absence and presence of WSDL
NEW ACTION: Editors to remove text for 2.5.2.1 and clean up spec to remove references to state variables (just call them variables)
NEW ACTION: Steve to reissue the challenge to remind the group to respond; SRT will formally reissue it to Jean-Jacques
NEW ACTION: We agree that we should separate repetitions into a separate construct. We need to sent this back to the authors and
explore as a group how to approach this, including expressing conditions on both variables and interactions.
NEW ACTION: Charlton to look at impact on our language of supporting BP (WS-I Basic Profile) 1.1,
SSBP (Simple SOAP Binding Profile) 1.0, AP (Attachments Profile) 1.0
NEW ACTION: Open new issue to investigate the solicit-response and notification in
WSDL 1.0 and WDSL 2.0
NEW ACTION: editors to define the specific syntax for a choice on messagecontenttype
NEW ACTION: everyone -- discuss on next con call: how far are we from LC "at the
5000 foot level" - i.e. are there major features that we are missing, are
there major features that cannot be cut. If so, what are they?
NEW ACTION: Yves to categorize the issues in bugzilla to add in our categories -
editorial, interact, etc. [Already done by Martin]
NEW ACTION: Editors to produce and publicize the list of changes made in the latest
ED draft.
NEW ACTION: Editors to evolve it in the next few weeks with goal of enabling the WG
to adopt a new WD at the next formal meeting 31 Aug.
NEW ACTIONI: Review and produce LC comments on the WSDL 2.0 docs by the end of the
next f2f. Need some to coordinate the activity. Chairs to put on agenda
for 31 aug.