From minutes of 17th April 2012. 145 Provide a 'dispatch' way of discriminating a choice for better performance of the envelope/payload use case (Steve, Mike, Suman) 12/7: See minutes. Need to choose a proposal and flesh out. 19/07: Waiting for proposals 26/07: Waiting for proposals 16/08: Waiting for proposals. Suman added to action. ... 1/11: Steve to send a proposal ... 21/03: Steve has sent a proposal. Mike has sent a counter proposal. Steve to respond. 28/03: Steve has sent a revised proposal. Review for discussion next week. Ensure proposal handles Mike's scenario where tag value to branch mapping is not 1-1. 05/04: Discussed Mike's review comments and Suman's concerns. Agreed that name should be elementID, should be a single DFDL String Literal value, and that matching of choiceBranchRef expression result should only be against elementID to avoid QName v String confusion. Steve to recirculate with a schema example. 17/04: Closed. Discussion on whether the choiceBranchRef expression should retiurn xs:string or something else. Agreed on xs:string. Discussed whether elemenID should be a pure xs:string or a DFDL String Literal. For consistency with other DFDL properties it should be a string literal, but raw byte entities and character classes should be disallowed to avoid complications. Discussed scope of uniqueness of elementIDs. Agreed that uniqueness is both local to a choice, and across all global elements in
Discussed this some more on the WG call. The original proposal from Steve envisaged a single property, choiceBranchRef, which was an expression that returned a QName. The attraction of this is that no property is needed on the element. It supports the use case where the 'tag' in the data corresponds closely to the element name, which is the case for something like SWIFT. Further, it is easy to extend this idea to xs:any wildcards in a future DFDL release. However if the 'tag' is very different from the element name then the expression can become a big 'if' statement and some of the performance benefit is lost. Hence the current proposal for choiceBranchRef expression returning a simple string, and the need for a property on elements. This gets us into a problem with global elements and uniqueness. What is needed for DFDL 1.0 is a mechanism that gives good performance for the known use cases, and is local to a choice, and does not preclude the provision of a QName based solution in a future DFDL release. What we have now is a halfway house. So the following is proposed: Change the name choiceBranchRef to choiceKey (or choiceTag or TBD). Note that the use of 'Ref' is dropped as that implies QName in all other properties that use 'Ref'. Change the name elementID to branchKey (or branchTag or TBD). Disallow branchKey on a global element, so it is allowed only on local elements and element refs. This removes the need for the special rule about choiceBranchKey on element ref overriding that on a global element. The mechanism is now entirely local to a choice, and will not clash with any future QName based scheme. Regards Steve Hanson Architect, IBM Data Format Description Language (DFDL) Co-Chair, OGF DFDL Working Group IBM SWG, Hursley, UK smh@uk.ibm.com tel:+44-1962-815848 From: Steve Hanson/UK/IBM To: dfdl-wg@ogf.org, Date: 19/08/2013 15:07 Subject: Fw: [DFDL-WG] Direct dispatch choice clarifications OK, I can go with changing elementID to something less suggestive of XSDL. I've spoken with Suman and changing it is not a big deal in our model. Regards Steve Hanson Architect, IBM Data Format Description Language (DFDL) Co-Chair, OGF DFDL Working Group IBM SWG, Hursley, UK smh@uk.ibm.com tel:+44-1962-815848 ----- Forwarded by Steve Hanson/UK/IBM on 19/08/2013 14:35 ----- From: Tim Kimber/UK/IBM@IBMGB To: dfdl-wg@ogf.org, Date: 17/08/2013 11:51 Subject: Re: [DFDL-WG] Direct dispatch choice clarifications Sent by: dfdl-wg-bounces@ogf.org I'm in general agreement with that. Just to avoid any possible confusion: I think the important thing is to stop thinking of elementID as any sort of XSD/DFDL language ID. It is not. It is a DFDL String Literal, meaning but it must describe a simple string value and it's value is matched against the return value of the choiceBranchRef expression. data. regards, Tim Kimber, DFDL Team, Hursley, UK Internet: kimbert@uk.ibm.com Tel. 01962-816742 Internal tel. 37246742 From: Mike Beckerle <mbeckerle.dfdl@gmail.com> To: Steve Hanson/UK/IBM@IBMGB, Cc: "dfdl-wg@ogf.org" <dfdl-wg@ogf.org>, "dfdl-wg-bounces@ogf.org" <dfdl-wg-bounces@ogf.org>, Tim Kimber/UK/IBM@IBMGB Date: 16/08/2013 18:54 Subject: Re: [DFDL-WG] Direct dispatch choice clarifications I am ok with leaving the name elementID as is purely due to our schedule. However, I really do think it would be good to change it now. I think it is that name which is creating all the confusion, and we really ought to be more careful. I think the important thing is to stop thinking of elementID as any sort of XSD/DFDL language ID. It is not. It is a DFDL String Literal, meaning it's value is matched against data. For example: It could be dfdl:elementID="%NUL;%NUL;%NUL;%NUL;" meaning 4 null chars. So that when the choiceBranchRef="{ ../hdr/tag }" evaluates to a 4 character string, then if they are all nul the branch is taken by direct dispatch. This character NUL isn't allowed in any namespaced identifiers in XSD or DFDL. It isn't even allowed in XML. If I had dfdl:termnator="%NUL;" you wouldn't ask "what namespace is that %NUL; in?" So I think elementID is not, in anyway, a namespace-qualified identifier any more than a delimiter is. So I think uniqueness of the elementID within the alternatives of a choice should be the only requirement. This stuff about unique within a namespace if on a global element should be dropped. The other argument for changing the name from elementID to something more choice-dispatch specific is that this workgroup seems to be speculating about using elementID for some sort of future wildcard-oriented feature. I think we should reserve names we want for that future by NOT using them now. If we use them now, we lock down part of the semantics in a way which may just not work properly with some future addition to DFDL. Do we really want elementID which is about direct dispatch choices but is not an identifier, and then in the future have dfdl:wildcardElementID which is about wildcards and IS an identifier. That's a mess. If you think elementID is a really good name for some future wildcard feature, then you should be advocating to NOT use it now for direct dispatch choices. That is, unless you have a complete design in mind for the wildcard stuff and are confident that elementID can be overloaded in a backward compatible way. I've seen nothing like this articulated. ...mike On Fri, Aug 16, 2013 at 12:32 PM, Steve Hanson <smh@uk.ibm.com> wrote: IBM has started to implement choiceBranchRef, in so far as the XML schemas for DFDL have been updated to include elementID and we have regenerated all our model code. I'd like to stick with elementID please. Regards Steve Hanson Architect, IBM Data Format Description Language (DFDL) Co-Chair, OGF DFDL Working Group IBM SWG, Hursley, UK smh@uk.ibm.com tel:+44-1962-815848 From: Mike Beckerle <mbeckerle.dfdl@gmail.com> To: Steve Hanson/UK/IBM@IBMGB, Cc: Tim Kimber/UK/IBM@IBMGB, "dfdl-wg@ogf.org" <dfdl-wg@ogf.org>, " dfdl-wg-bounces@ogf.org" <dfdl-wg-bounces@ogf.org> Date: 16/08/2013 17:09 Subject: Re: [DFDL-WG] Direct dispatch choice clarifications I think we need to stay out of identifiers and namespaces and qnames for this feature. The elementID should have to be locally unique within the alternatives of the choice that is dispatching to it. If that elementID lives on a global element declaration, that means nothing at all alone. In the context of a choice that has an element ref to that global element, then it has to be unique within the arms of that choice. This means it is possible to have two global element decls which have elementID="X", and there is no conflict unless they are both used via element refs from the same choice. I think this handles every use case I know of. The "namespace" requirement for elementID is only "unique within alternatives of a choice". Since nobody has implemented this feature yet, I would posit that we should change elementID to something less suggestive of an XSD identifier/QName-ish thing. Such as choiceDispatchTag. ...mike On Fri, Aug 16, 2013 at 5:18 AM, Steve Hanson <smh@uk.ibm.com> wrote: You are right about the xs:QName constructor, it takes a string which is prefix plus name. If we supplied a dfdl:QName constructor that took a URI and a name, that would simplify things. For example, the choiceBranchRef expression for SWIFT would be as below. (The element name is always Document and the namespace is used to distinguish the different messages). {dfdl:QName(fn:concat(fn:concat('urn:swift:xsd:fin.', /FinMessage/Block2/MessageType), ".2011"), 'Document')} If we stick with elementID for DFDL 1.0, then I agree with your 3 bullets, but not the defaulting to element name as it is setting a behaviour that may not be where we want to go for 2.0. Regards Steve Hanson Architect, IBM Data Format Description Language (DFDL) Co-Chair, OGF DFDL Working Group IBM SWG, Hursley, UK smh@uk.ibm.com tel:+44-1962-815848 From: Tim Kimber/UK/IBM@IBMGB To: dfdl-wg@ogf.org, Date: 15/08/2013 20:34 Subject: Re: [DFDL-WG] Direct dispatch choice clarifications Sent by: dfdl-wg-bounces@ogf.org The other problem with using a QName is that it involves using namespace prefixes. That means that there needs to be a mapping between prefixes and namespace URIs. I can see that getting very problematic if the choice group is located in a different xsd from the global elements that it references. I think we should - keep the elementID as a simple string - insist that all branches of a choice have different elementIDs - remove the global uniqueness constraint, for the reasons explained in this email chain I think it would be easier for modellers if the elementID defaulted to the local name of the element. I understand that name clashes can, in principle, occur. If users want to avoid that then they can be explicit about elementIDs and they could even define a naming convention for their elementIDs to make them look very much like QNames. Sounds like a lot of work, but DFDL models that are complex enough to need that approach will often be code-genned anyway. regards, Tim Kimber, DFDL Team, Hursley, UK Internet: kimbert@uk.ibm.com Tel. 01962-816742 Internal tel. 37246742 From: Steve Hanson/UK/IBM@IBMGB To: dfdl-wg@ogf.org, Date: 15/08/2013 18:37 Subject: Re: [DFDL-WG] Direct dispatch choice clarifications Sent by: dfdl-wg-bounces@ogf.org The more I think it through, the more I see the use of a string elementID (or local element name) causing problems when/if we extend to support xs:any in the future. In my original proposal for direct dispatch choice I proposed that choiceBranchRef returned a QName which therefore automatically selected the element, and coped with any namespace issues. The problem with QName though is that the expression to build it can become a big case statement negating some of the performance gain, if there is no automap way of getting from the 'tag' to the QName. Hence why we introduced elementID. Regards Steve Hanson Architect, IBM Data Format Description Language (DFDL) Co-Chair, OGF DFDL Working Group IBM SWG, Hursley, UK smh@uk.ibm.com tel:+44-1962-815848 From: Steve Hanson/UK/IBM To: Suman Kalia <kalia@ca.ibm.com>, Cc: dfdl-wg@ogf.org, dfdl-wg-bounces@ogf.org, Tim Kimber/UK/IBM@IBMGB Date: 15/08/2013 17:20 Subject: Re: [DFDL-WG] Direct dispatch choice clarifications Suman, comments to yours in pink Regards Steve Hanson Architect, IBM Data Format Description Language (DFDL) Co-Chair, OGF DFDL Working Group IBM SWG, Hursley, UK smh@uk.ibm.com tel:+44-1962-815848 From: Suman Kalia <kalia@ca.ibm.com> To: Tim Kimber/UK/IBM@IBMGB, Cc: dfdl-wg@ogf.org, dfdl-wg-bounces@ogf.org Date: 15/08/2013 15:31 Subject: Re: [DFDL-WG] Direct dispatch choice clarifications Sent by: dfdl-wg-bounces@ogf.org comments in green Suman Kalia IBM Canada Lab WMB Toolkit Architect and Development Lead Tel: 905-413-3923 T/L 313-3923 Email: kalia@ca.ibm.com For info on Message broker http://www.ibm.com/developerworks/websphere/zones/businessintegration/wmb.ht... From: Tim Kimber <KIMBERT@uk.ibm.com> To: dfdl-wg@ogf.org, Date: 08/15/2013 09:58 AM Subject: Re: [DFDL-WG] Direct dispatch choice clarifications Sent by: dfdl-wg-bounces@ogf.org See comment in <TK> tags. regards, Tim Kimber, DFDL Team, Hursley, UK Internet: kimbert@uk.ibm.com Tel. 01962-816742 Internal tel. 37246742 From: Steve Hanson/UK/IBM@IBMGB To: dfdl-wg@ogf.org, Date: 15/08/2013 12:57 Subject: [DFDL-WG] Direct dispatch choice clarifications Sent by: dfdl-wg-bounces@ogf.org Looking at this in more detail prior to writing up behaviour for section 9, there are a couple of things missing from the spec or that need clarification: 1) Description of elementID property should say that empty string is not allowed (this was in the erratum). 2) Should say that an elementID on an elementRef overrides any elementID on the global element (this was in the erratum). 3) Section 15.1.2 says that is a schema definition error if the elementId values of global elements are not unique within a given namespace. I don't see where namespace comes into this, the elementID is just a string so surely it needs to be unique across namespaces? (Strictly elementID needs only to be unique across the global elements involved in each specific choice, but it was minuted that global uniqueness was desirable to allow future xs:any support). <TK> In XML Schema, an xs:any does not, in general, match all global elements. The 'namespace' attribute can narrow the set to elements from a specified list of namespaces. There is no way in XML Schema 1.0 to further narrow the xs:any, So the rule is designed to ensure that future usage of xs:any when a single namespace is specified and processContents!='skip' does not throw up schema definition errors. However...I note that XML Schema 1.1 allows a new way to narrow the scope of an xs:any ( by specifying a list of not-included QNames ). My feeling is that the unique-within-namespace check is fragile. </TK> <SK> As per my recollection, we put the uniqueness rules across namespaces to accommodate chameleon namespaces. Consider a global element E1 in notarget namespace having elementID E1_ID and is included in 2 schemas with different target namespaces say TNS1 and TNS2.. Consider a choice containing element references TNS1:E1 and TNS2:E1, in order to disambiguate these elements in the context choice, the element ID has to be unique in the context of namespace. This is somewhat an edge case but can come more prevalent when the support for xs:any is provided. </SK> SMH: In the choice, the element refs to TNS1:E1 and TNS2:E1 both have an elementID string 'E1_ID' from the original E1 global element. In the choice, this is an error because the elementID is not unique in the choice (we match the result of the choiceBranchRef expression, which returns a string not a QName). The only way round this is to override the elementID on one of the element refs (see 2 above) and set a value that is unique. That then works. But that does not help the (future) xs:any scenario, where there is no element ref to carry the override. I think the chameleon namespace scenario will always cause a problem with xs:any because our elementIDs are strings not QNames. I think we should leave a global element uniqueness check out of DFDL 1.0. It doesn't actually future proof anything, as once I use xs:any the whole nature of the xsd changes. 4) Spec does not explicitly say that when choiceBranchRef is present each branch of the choice must have an elementID. This must be the case, as otherwise a choice branch will never be accessible. 5) Tim has suggested that if an element was silent about elementID, the local name of the element could be used instead. So conceptually an element would have an 'effective elementID'. This makes modelling easier if the 'tag' in the data is the same as the element name. <TK>...or if the element name is derivable from the 'tag' using a simple XPath expression</TK> SMH: True. The validation checks would need to ensure that the set of 'effective elementIDs' was unique; for the global element check as currently specified (see 3) this would mean that all global elements must have unique local names, unless an elementID is carried - I think this is too limiting. SMH: While defaulting to the local name sounds attractive, I can't convince myself that it won't cause problems if we add xs:any in DFDL 2.0 and multiple/chameleon namespaces are involved. SMH: Conclusion: For DFDL 1.0 we take the conservative position and say that you must specify an elementID on an element that is used in a choice with choiceBranchRef and it must be unique in the context of the choice only. No global uniqueness check is made. the same namespace (the latter is not strictly needed right now but accommodates any future addition of xs:any). Agreed that elementID should be on global element, local element, and element ref (in which case it overrides any elementId on the global element, which is ok as the property does not follow the usual scoping rules). Errata taken. Regards Steve Hanson Architect, IBM Data Format Description Language (DFDL) Co-Chair, OGF DFDL Working Group IBM SWG, Hursley, UK smh@uk.ibm.com tel:+44-1962-815848 Unless stated otherwise above: IBM United Kingdom Limited - Registered in England and Wales with number 741598. Registered office: PO Box 41, North Harbour, Portsmouth, Hampshire PO6 3AU -- dfdl-wg mailing list dfdl-wg@ogf.org https://www.ogf.org/mailman/listinfo/dfdl-wg Unless stated otherwise above: IBM United Kingdom Limited - Registered in England and Wales with number 741598. Registered office: PO Box 41, North Harbour, Portsmouth, Hampshire PO6 3AU -- dfdl-wg mailing list dfdl-wg@ogf.org https://www.ogf.org/mailman/listinfo/dfdl-wg -- dfdl-wg mailing list dfdl-wg@ogf.org https://www.ogf.org/mailman/listinfo/dfdl-wg Unless stated otherwise above: IBM United Kingdom Limited - Registered in England and Wales with number 741598. Registered office: PO Box 41, North Harbour, Portsmouth, Hampshire PO6 3AU Unless stated otherwise above: IBM United Kingdom Limited - Registered in England and Wales with number 741598. Registered office: PO Box 41, North Harbour, Portsmouth, Hampshire PO6 3AU -- dfdl-wg mailing list dfdl-wg@ogf.org https://www.ogf.org/mailman/listinfo/dfdl-wg Unless stated otherwise above: IBM United Kingdom Limited - Registered in England and Wales with number 741598. Registered office: PO Box 41, North Harbour, Portsmouth, Hampshire PO6 3AU -- dfdl-wg mailing list dfdl-wg@ogf.org https://www.ogf.org/mailman/listinfo/dfdl-wg Unless stated otherwise above: IBM United Kingdom Limited - Registered in England and Wales with number 741598. Registered office: PO Box 41, North Harbour, Portsmouth, Hampshire PO6 3AU -- dfdl-wg mailing list dfdl-wg@ogf.org https://www.ogf.org/mailman/listinfo/dfdl-wg -- Mike Beckerle | OGF DFDL Workgroup Co-Chair | Tresys Technology | www.tresys.com Unless stated otherwise above: IBM United Kingdom Limited - Registered in England and Wales with number 741598. Registered office: PO Box 41, North Harbour, Portsmouth, Hampshire PO6 3AU -- Mike Beckerle | OGF DFDL Workgroup Co-Chair | Tresys Technology | www.tresys.com Unless stated otherwise above: IBM United Kingdom Limited - Registered in England and Wales with number 741598. Registered office: PO Box 41, North Harbour, Portsmouth, Hampshire PO6 3AU -- dfdl-wg mailing list dfdl-wg@ogf.org https://www.ogf.org/mailman/listinfo/dfdl-wg Unless stated otherwise above: IBM United Kingdom Limited - Registered in England and Wales with number 741598. Registered office: PO Box 41, North Harbour, Portsmouth, Hampshire PO6 3AU Unless stated otherwise above: IBM United Kingdom Limited - Registered in England and Wales with number 741598. Registered office: PO Box 41, North Harbour, Portsmouth, Hampshire PO6 3AU