[Date Prev] | [Thread Prev] | [Thread Next] | [Date Next] -- [Date Index] | [Thread Index] | [List Home]
Subject: RE: [wsbpel] Issue - 157 - Proposal For Vote
Alex, It is obvious that you have a strong viewpoint on this but there seems to be enough of a feeling that we can actually use XSLT to derive equivalent functiionality to copy that it would be fair to present two proposals to the TC with pros and cons attached, and let the TC make a choice. In my mind this is the most important open issue right now, followed by 6. I think we should take care of 11 using XSLT, as Yaron says. The question remains whether we can also deal with 157 with that same approach. I would be very happy to see a formal fleshed out XSLT based proposal for 157 which addresses or at least counters Alex's objections, but don't feel qualified to make one myself. Satish ________________________________ From: Alex Yiu [mailto:alex.yiu@oracle.com] Sent: Mon 6/6/2005 4:15 PM To: ygoland@bea.com Cc: Satish Thatte; Yuzo Fujishima; wsbpeltc; Assaf Arkin; Danny van der Rijn; Charlton Barreto; Alex Yiu Subject: Re: [wsbpel] Issue - 157 - Proposal For Vote Yaron, [A] This is NOT an implementation issue. But, this is a language semantics and specification issue. I add a reply to explain this with an example later. [B] BTW, can someone show me an XSLT implementation that will actually modify the source tree without generating a seperate result tree? (Please read my quotation from XSLT spec: "The result tree is separate from the source tree."). If that XSLT implementation does not violate the spec, I don't know what implementation will violate the spec. :-) [C] To clarify, I don't mind ADDING an optional "stylesheet" attribute to the from-spec. I think that is a good feature, as long as current <copy> functionalities are kept. BTW, I am curious who voted to close Issue 48 without changes. :-) If the group really likes the transform features, let us EITHER open a new issue similar which is somewhat equivalent to Issue 48 OR re-group the XSLT feature under Issue 11. Thanks! Regards, Alex Yiu Yaron Y. Goland wrote: I don't see any of Alex's issues as being show stoppers. Yes, XSLT is a replace. So what? A dump implementation will be expensive and a smart implementation will be cheap. What else is new? As for the vote, the group agreed to keep assign and copy semantics, I would argue that XSLT provides those copy semantics so I don't see this action as violating the word or spirit of the motion. Using XSLT would drastically simplify the specification, allow us to avoid re-inventing a whole stack of wheels, provide all the functionality we wanted from copy and resolve issue 11. Using XSLT seems like a winner to me. Yaron Alex Yiu wrote: Hi all, Yuzo, thank you for the sample example also. _*[A] Nature of XSLT*_ The same old reminder to people: XSLT and XQuery are used to _create *NEW* documents or XML fragments based on an existing piece of XML data_. Quotation--1: from the Introduction of XSLT 1.0 Spec: -------------------------------------- A transformation expressed in XSLT describes rules for _transforming a source tree into a result tree_. The transformation is achieved by associating patterns with templates. A pattern is matched against elements in the source tree. A template is instantiated to _*create* part of the result tree_. _The result tree is *separate* from the source tree_. The structure of the result tree can be completely different from the structure of the source tree. ... ... _A template is *instantiated* for a particular source element_ to _*create* part of the result tree_. A template can contain elements that specify literal result element structure. A template can also contain elements from the XSLT namespace that are _instructions for *creating* result tree_ fragments. When a template is instantiated, each instruction is executed and replaced by the result tree fragment that it creates. -------------------------------------- [Note: "replace" here does NOT mean replace the source-tree. It means: replace the instructions in the template with execution result. Think of any web-page-template engine. e.g.: ASP, JSP, PHP.] Typically, an XSLT has a default "catch-all-else" copy rule similar to below to do a copy any nodes from source tree to result tree without any transformation: ----------------------------- <xsl:template match="@*|node()"> <xsl:copy> <xsl:apply-templates select="@*|node()"/> </xsl:copy> </xsl:template> ----------------------------- Quotation--2: [ http://www.w3.org/TR/xslt#copying ] ------------------------------- The xsl:copy element provides an easy way of _*copying* the current node_. _Instantiating the xsl:copy element creates a copy of the current node_. ------------------------------- If the result tree was NOT a distinct copy from the source tree, the above <xsl:copy> would not be needed. In fact, if the above "catch-all-else" copy rule are _NOT used in XSLT_, most of nodes (if not all) in the _source tree will NOT be present in / copied to the result tree_. I hope the above quotations are good enough to prove that XSLT creates NEW documents but NOT replacing any existing docs. If not, we just need to cross reference to our W3C rep in XPath, XSLT, and XQuery WG - TRIO in W3C. Furthermore, XSLT is a non-procedural language. If executation of XSLT were allowed to modify any source or existing XML Document, its semantics can be extremely difficult to be defined. E.g. Will rules in XSLT be re-fired, if the source document got changed? (very hairy business there. Will it result in infinite loop?) In short, XSLT is used to create new documents. _When the base document is large and delta is small, the runtime efficiencies of such an XSLT will be inheritly low. _ _*[B] How to apply XSLT in BPEL *_ Yuzo's example illustrates one of the possible ways to interprete Assaf's original suggestion to apply XSLT in BPEL [i.e. (1b) below]. *(1)* _Using XSLT to transform a variable or a variable part:_ *(1a)* Refering to a variable or a variable part as the source document: -------------------------------- <assign> ... <transform variable="var1" part="part1" stylesheet="http://foo.com/my.xslt"; <http://foo.com/my.xslt> /> outputVariable="var2" outputPart="part1" /> ... </assign> -------------------------------- Or alternatively using a syntax closer to existing <copy>: -------------------------------- <assign> ... <copy> <from variable="var1" part="part1" stylesheet="http://foo.com/my.xslt"; <http://foo.com/my.xslt> /> <to variable="var2" part="part1" /> </copy> ... </assign> -------------------------------- *(1b)* Embedding XSLT into BPEL with variable bindings: Examples would be very similar to what Yuzo has provided. (1a) and (1b) are not that different. One allows multiple input variables, while the other allows one variable only. One uses embedded XSLT, one refers the XSLT by a URI. _*(2)* Still using 100% of current <assign>/<copy> syntax._ E.g.: --------------------------- <assign> <copy> <from>$var/p:abc/p:def</from> <to>$var/p:ghi/p:def</to> </copy> </assign> --------------------------- However, when one try to define its semantics of syntax, one may reuse fragments of definitions from XSLT (e.g. "xslt:copy", "xslt:copy-of", "xslt:value-of") _*KEY POINTS to analysis about XSLT facilities:*_ * The biggest difference between (1) and (2) is: the granularity of replacement: ==> _variable or msg-part *VS* any node within a variable or a msg-part _ * I would emphasize that the _replacement of any node is a big part of <assign>/<copy> logic in BPEL 1.1_. Since we passed to keep <assign>/<copy> functionalities, we MUST/SHOULD keep this part of functionality as well. And, not having the capabilities of a smaller granularity of replacement has a _BIG impact on efficiency of <assign>_. For example: in order to replace a small zip code field of a large PO documents (e.g. 100 line items), we would effectively copy all those 100 line items. That is NOT an implementation-dependent issue. The XSLT spec clearly shows its intention (see the quotations above). * And, leveraging XSLT can be _a totally orthogonal add-on feature to the current <assign>/<copy>_. For example: XSLT features in (1a) and (1b) can be _extended and blended with existing <copy> features_: -------------------------------- <assign> ... <copy> <from> _<transform>_ <!-- embedded XSLT code ... similar to Yuzo's example --> _<transform>_ </from> <to> *$var2.part1/p:lineItem[10]* </to> </copy> ... </assign> -------------------------------- -------------------------------- <assign> ... <copy> <from variable="var1" part="part1" _stylesheet="http://foo.com/my.xslt"; <http://foo.com/my.xslt> _ /> <to> *$var2.part1/p:lineItem[10]* </to> </copy> ... </assign> -------------------------------- * The "xslt:copy", "xslt:copy-of", "xslt:value-of" are used to _*append*_ the execution result of "xslt:*" into the current result tree, which is being created brand-new. It _has NO definition on how to do node replacement_ for an existing doc tree. Regardless whether we go (1a), (1b) or (2), we still need to define a table very similar to the table in my previous Issue 157 email. E.g.: Src\Dest EII AII/TII EII with Complex-Content RE F EII with Simple-Content RE RC AII/TII RC RC Because, we need to define the semantics on how to replace the value pointed by the to-spec withe result of the XSLT transformation. E.g. there should be a fault, if one tries to fill in a simple-typed variable with an element of complex content. (i.e. the Fault caseas denoted by "F" above). Even if somebody tries to cut the corner by removing the capabilties to replace a smaller granularity of variable tree, it will NOT make the definition simplier. Because, the copy/replacement behavior on Attribute (AII) and Text (TII - from simple-typed variable) are virtually the same. (See the table above) In short, _attempt to leverage XSLT concept to simplify Issue 157_ is just a _*"false hope of economy"*_. On the other hand, adding XSLT to BPEL (under Issue 48 or Issue 11) does have its merits in a sense. * To define an XSLT to just do simple-type manipulation is just awkard and anti-ease-of-use. Think of about the case that one needs to increase an counter value. With current <copy>, it is straight forward and easy: <assign> <copy> <from>$count + 1 </from> <to> $count </to> </copy> </assign> Try to do that with an XSLT. ;-) ... Possible but clumsy. * Currently, <assign>/<copy> and Issue 103 are _data-model independent_. That is, we have no bias to any expression language. If BPEL's only operation under <assign> has such a strong dependency on XSLT1.0, the BPEL's expression language independence would be destroyed. (If BPEL spec ever enforces any data model in future, we should enforce XPath 2.0 data model, because of its schemaful nature, instead of XPath1.0/XSLT1.0.) [Side note: Between (1a) and (1b), I prefer (1a), because I am afraid that we would go through another issue 103 exercise, even though we may be able to reuse a number of definitions from Issue 103. ] _*[C] Summary*_ Main goals for Issue 157: * _to clarify the semantics of <copy>_ o The typical and trivial cases in the table <copy> logic are quite clear already. That's why there are more than 3 companies coming up with interpretation in those cases. All we need to clarify those corner cases (e.g. when a standard runtime fault would happen). * _NOT to re-define <copy>_ into a completely different animal and causing all sorts of un-resolvable backward compatibility problems. o We have passed a binding resolution to keep <assign>/<copy>. _If we defined XSLT transform as the only operation under <assign>_, that is virtually getting rid of <copy>. I would argue that is a _direct violation to our binding resolution_ in the F2F in Palo Alto. * Let us focus finish the clarification of <copy> in Issue 157 and call it a day. :-) o I am happy to see a standardized XSLT transform construct also, if we want to _*add *XSLT stylesheet attribute on top of from-spec_. But, _let's put it into Issue 48 or Issue 11_. Thank you for reading such a long email! Regards, Alex Yiu Satish Thatte wrote: Yuzo, Thanks for the example. I will study it. It seems to me that you have used XPath variable binding in the template, which assumes that our current Xpath variable binding mechanism would carry over without additional work. Is that your understanding? Satish ________________________________ From: Yuzo Fujishima [mailto:fujishima@bc.jp.nec.com] Sent: Sat 6/4/2005 8:11 AM To: Satish Thatte; wsbpeltc Cc: Assaf Arkin; Danny van der Rijn; Charlton Barreto Subject: Re: [wsbpel] Issue - 157 - Proposal For Vote I don't quite see what are the problems. I've written an example and attached the files used. (Apply assign.xsl to var1.xml. Then you'll get newVar1.xml.) Suppose we want to assign from var2 and var3 to var1. In that case, we write something similar to assign.xsl (excluding the variable binding part) within an assign activity. Perhaps we should rename the activity to "transform". The activity should look like below: <transform variable="var1"> <xsl:stylesheet version="1.0" xmlns:xsl="http://www.w3.org/1999/XSL/Transform"; <http://www.w3.org/1999/XSL/Transform> > <xsl:template match="orderNumber"> <xsl:element name="orderNumber"> <xsl:value-of select="$var2/purchaseOrder/orderNumber"/> </xsl:element> </xsl:template> <xsl:template match="billingAddress"> <xsl:element name="billingAddress"> <xsl:copy-of select="$var3/customers/customer[ @id=$var2/purchaseOrder/buyerId]/address/*"/> </xsl:element> </xsl:template> <xsl:template match="@*|node()"> <xsl:copy> <xsl:apply-templates select="@*|node()"/> </xsl:copy> </xsl:template> </xsl:stylesheet> </transform> Those parts of var1 that need not be modified are copied by <xsl:template match="@*|node()"> template. Pretty simple and idiomatic. I noticed that the use of XSLT may sometimes result in more efficient (rather than inefficient) processing because multiple assigns can be bunched together to form one transformation. How do you like above? Yuzo Fujishima NEC Corporation Satish Thatte wrote: Alex has pointed out various technical difficulties he sees with using XSLT to resolve 157, privately to some of us at the F2F. The principal one being that XSLT templates costruct new XML documents rather than update existing ones. According to Alex, doing anything else requires non-standard usage of XSLT. I am no XSLT expert so I am looking for confirmation or otherwise of Alex's position. If he is right then we are stuck with doing 157 as far as I can see. I would much prefer not to add features to BPEL for any XML data manipulation, but we have assignment/copy today and we have voted not to remove them so the only other option is to find a way to use some existing XML data manipulation spec as normative with the usual extensibility. XPath is what we have used in copy but that seems to require the sort of xII rules that the current 157 proposals specify. If someone has a simpler way to resolve this I am all ears. Satish ________________________________ From: Assaf Arkin [mailto:arkin@intalio.com] Sent: Fri 6/3/2005 9:28 AM To: Danny van der Rijn Cc: Satish Thatte; Charlton Barreto; wsbpeltc Subject: Re: [wsbpel] Issue - 157 - Proposal For Vote I wouldn't rule out that we closed issue 48 because the scope was a bit more generic, and from the issue list, apparently because 13 solves the problem. So given issue 13 you can incorporate XSLT into BPEL as a proprietary extension. The problem we're facing is how to handle simple assignments as a normalized part of the spec, which 48/13 do not cover. I, for one, am not interested in re-inventing the wheel, so in this particular context I have no problem revisiting issue 48 or resolving this as part of issue 157, whichever is more convenient. Assaf Danny van der Rijn wrote: What you're proposing, Satish, sounds like Issue 48, which was rejected in April of 2004. I'd be happy to revisit it, but I'm not sure how others feel. Danny Satish Thatte wrote: Wonderful. I didn't see this mail before sending mine. +1 ________________________________ From: Charlton Barreto [mailto:cbarreto@adobe.com] Sent: Thu 6/2/2005 9:29 AM To: Assaf Arkin Cc: wsbpeltc Subject: Re: [wsbpel] Issue - 157 - Proposal For Vote I second this proposal. I feel that we should delegate the problem scoped by issue 157 by leveraging another spec which is actually focussed on it, rather than attempting to invest the considerable amount of time and energy necessary to address it in BPEL. On 01/06/2005, at 16:08, Assaf Arkin wrote: I want to propose an alternative way to solve issue 157, without introducing significant complexity to the language. The XSLT specification has been dealing with the exact same problem of evaluating an expression and placing the results in a target context. XSLT provides two separate mechanisms: copy-of and value-of. I would propose using these two XSLT elements as a normative part of the BPEL specification, instead of inventing yet another copy mechanism. http://www.w3.org/TR/xslt#copy-of http://www.w3.org/TR/xslt#value-of Assaf || <arkin.vcf> Charlton Barreto P 1.408.536.4496 cbarreto@adobe.com www.adobe.com <http://www.adobe.com/> <http://www.adobe.com/> logo.gif --------------------------------------------------------------------- To unsubscribe from this mail list, you must leave the OASIS TC that generates this mail. You may a link to this group and all your TCs in OASIS at: https://www.oasis-open.org/apps/org/workgroup/portal/my_workgroups.php --------------------------------------------------------------------- To unsubscribe from this mail list, you must leave the OASIS TC that generates this mail. You may a link to this group and all your TCs in OASIS at: https://www.oasis-open.org/apps/org/workgroup/portal/my_workgroups.php --------------------------------------------------------------------- To unsubscribe from this mail list, you must leave the OASIS TC that generates this mail. You may a link to this group and all your TCs in OASIS at: https://www.oasis-open.org/apps/org/workgroup/portal/my_workgroups.php
[Date Prev] | [Thread Prev] | [Thread Next] | [Date Next] -- [Date Index] | [Thread Index] | [List Home]