Re: [wsbpel] Issue 29 - Proposal For Vote

[Assaf Arkin <arkin@intalio.com>]

Yaron Y. Goland wrote:

> The reference to XPATH 2.0 isn't relevant, XPATH 1.0 != XPATH 2.0.

For future reference, my e-mails are shorter and more to the point when 
I quote the XPath 2.0 and only the XPath 2.0 specification, since that 
version is much better organized and better articulated.

I assume that most members of this group will find the more 
readable/concise text easier to follow, and that the few members who are 
deeply interested in this subject can go and validate my statements 
against the XPath 1.0 specification.

I will always point out if there are differences between the two specs 
(see below).


>
> And at the end of the letter you point out yourself that you wouldn't 
> want to use empty node-sets as the context but that is exactly what 
> your proposal requires.

To further clarify, my proposal requires that you either pass an empty
root node (XPath 1.0) or an empty document node (XPath 2.0/XQuery 1.0)*
as the context node, in compliance with these specifications.

If it's still not clear, can you point out where I'm being confusing?

Assaf

* You should note the name difference between the two specs.


>
>     Yaron
>
> Assaf Arkin wrote:
>
>> Yaron Y. Goland wrote:
>>
>>  > I realize I'm being thick but I don't believe the sections you quote
>>  > solve the problem. The key issue is - is there such a thing as an
>>  > empty node in XPATH? I believe the answer to be no.
>>
>> I believe the answer is found in the XPath data model specification:
>>
>> "Document Nodes *must* satisfy the following constraints.
>>
>>    1.
>>
>>       The *children* *must* consist exclusively of Element, Processing
>>       Instruction, Comment, and Text Nodes if it is not empty.
>>       Attribute, namespace, and Document Nodes can never appear as 
>> children
>>
>>    2.
>>
>>       If a node /N/ is among the *children* of a Document Node /D/, then
>>       the *parent* of /N/ *must* be /D/.
>>
>>    3.
>>
>>       If a node /N/ has a *parent* Document Node /D/, then /N/ *must* be
>>       among the *children* of /D/."
>>
>> http://www.w3.org/TR/xpath-datamodel/#DocumentNode
>>
>> You may also want to look at XPath implementations to see what they
>> accept as context node as part of their formal API. I will be interested
>> in hearing of an implementation that doesn't accept an empty context
>> node.  For example, the Java 1.5 XPath library:
>>
>> http://java.sun.com/j2se/1.5.0/docs/api/javax/xml/xpath/package-summary.html 
>>
>>
>> (you can see the example at the bottom)
>>
>>  > An alternative argument is - is there such a thing as an empty
>>  > node-list in XPATH? There I think the answer is potentially yes. But
>>  > it isn't clear to me if an empty node-list can be legally used as a
>>  > context for an expression because all de-references would be illegal.
>>  > E.g. /foo would be an automatic fault but "/" would NOT. That worries
>>  > me alot. I think it's misleading to allow "/" to work but "/foo" not
>>  > to. I think it's better for users if we just ban "/".
>>
>> I don't know of many implementations that support a node-set as an
>> evaluation context, and I wouldn't want to rely on anything that's
>> implementation specific.
>>
>> Assaf
>>
>>  >
>>  >         Yaron
>>  >
>>  > Assaf Arkin wrote:
>>  >
>>  >> Yaron Y. Goland wrote:
>>  >>
>>  >>  > This proposal is only relevant if we don't decide to adopt a
>>  >> mechanism
>>  >>  > to map properties to XPATH variables.
>>  >>  >
>>  >>  > As for the desire to use XPATH 1.0 in a standard manner,
>>  >> unfortunately
>>  >>  > that is quite impossible given our semantics. It is explicitly
>>  >> illegal
>>  >>  > to have a null context node in XPATH 1.0. In section 1, the
>>  >>  > introduction to XPATH 1.0, a series of requirements are listed 
>> for
>>  >>  > defining the context in which a XPATH executes. List in there 
>> is a
>>  >>  > node (explicitly referred to as the context node). This is then
>>  >>  > followed by a requirement for a pair of non-zero positive 
>> integers
>>  >>  > identifying the context position and context size. If the 
>> context is
>>  >>  > empty or null then these two numbers would have to be 0 and 
>> that's
>>  >>  > explicitly prohibited. Hence why we already are forced into a
>>  >>  > situation where we have to require a BPEL specific XPATH 
>> processor.
>>  >>
>>  >> It is illegal to have a null context node, it is permissible to 
>> have an
>>  >> empty context node. My suggestion takes XPath into account so it 
>> doesn't
>>  >> violate any XPath rules. So my suggestion still stands, now let me
>>  >> explain why it works.
>>  >>
>>  >> The context size refers to the size of the node-set which 
>> contains the
>>  >> current context node, while the position refers to the position 
>> of the
>>  >> current context node in that node-set. When you begin evaluating an
>>  >> XPath expression - the initial values are 1 and 1, meaning one 
>> element
>>  >> in the node-set (the context node) in the first ordinal position. It
>>  >> matters not what the contents of the context node is, i.e. 
>> whether it's
>>  >> a document, an element, a text node, or an empty node.
>>  >>
>>  >> The XPath 1.0 specification is easy to misread. If you try to 
>> implement
>>  >> it, you will soon catch this subtelty when you have to implement
>>  >> functions like count() and position(), but without going deep 
>> it's easy
>>  >> to misread it. I suggest referring to the XPath 2.0 
>> specification, which
>>  >> is not fundamentally different on this issue, just more 
>> expressive and
>>  >> precise:
>>  >>
>>  >>     *
>>  >>
>>  >>       Definition: The *context item* is the item currently being
>>  >>       processed. An item is either an atomic value or a
>>  >>       node.][Definition: When the context item is a node, it can 
>> also be
>>  >>       referred to as the *context node*.] The context item is 
>> returned
>>  >>       by an expression consisting of a single dot (|.|). When an
>>  >>       expression |E1/E2| or |E1[E2]| is evaluated, each item in the
>>  >>       sequence obtained by evaluating |E1| becomes the context 
>> item in
>>  >>       the inner focus for an evaluation of |E2|.
>>  >>
>>  >>     *
>>  >>
>>  >>       [Definition: The *context position* is the position of the 
>> context
>>  >>       item within the sequence of items currently being 
>> processed.] It
>>  >>       changes whenever the context item changes. Its value is 
>> always an
>>  >>       integer greater than zero. The context position is returned 
>> by the
>>  >>       expression |fn:position()|. When an expression |E1/E2| or 
>> |E1[E2]|
>>  >>       is evaluated, the context position in the inner focus for an
>>  >>       evaluation of |E2| is the position of the context item in the
>>  >>       sequence obtained by evaluating |E1|. The position of the 
>> first
>>  >>       item in a sequence is always 1 (one). The context position is
>>  >>       always less than or equal to the context size.
>>  >>
>>  >>     *
>>  >>
>>  >>       [Definition: The *context size* is the number of items in the
>>  >>       sequence of items currently being processed.] Its value is 
>> always
>>  >>       an integer greater than zero. The context size is returned 
>> by the
>>  >>       expression |fn:last()|. When an expression |E1/E2| or 
>> |E1[E2]| is
>>  >>       evaluated, the context size in the inner focus for an 
>> evaluation
>>  >>       of |E2| is the number of items in the sequence obtained by
>>  >>       evaluating |E1|.
>>  >>
>>  >> Assaf
>>  >>
>>  >>  >
>>  >>  >     Yaron
>>  >>  >
>>  >>  > Assaf Arkin wrote:
>>  >>  >
>>  >>  >> -1
>>  >>  >>
>>  >>  >> If we decide to manifest properties as independent XPath 
>> variables,
>>  >>  >> then we have a syntax that is consistent in how variable 
>> values are
>>  >>  >> accessed, directly (the entire variable) or indirectly (a
>>  >> property of
>>  >>  >> the variable). I happen to be on the side that likes 
>> consistency and
>>  >>  >> abhors making changes to XPath implementations.
>>  >>  >>
>>  >>  >> "We have in the past altered XPATH in order to suit our 
>> needs. For
>>  >>  >> example, in some BPEL expressions we have banned the use of the
>>  >>  >> global context node, a node whose presence is actually 
>> mandated by
>>  >>  >> the XPATH specification. If we are willing to change XPATH 
>> that much
>>  >>  >> I see no reason not to change it here as well. In for a dime, 
>> in for
>>  >>  >> a dollar."
>>  >>  >>
>>  >>  >> Since we have nothing to pass in the context node, wouldn't 
>> it be
>>  >>  >> easier if we use the XPath specification to achieve that 
>> instead of
>>  >>  >> forcing implementations to deviate from the specification, write
>>  >>  >> custom XPath implementations (only for BPEL), and so forth. The
>>  >> XPath
>>  >>  >> specification mandates that you pass a context node, but the 
>> XPath
>>  >>  >> specification does not require the context node to have any 
>> content
>>  >>  >> in it. You can pass an empty root node (XPath 1.0) or an empty
>>  >>  >> document node (XPath 2.0) and be in full comformance with the 
>> XPath
>>  >>  >> specification and the BPEL model at the same time.
>>  >>  >>
>>  >>  >> Let's keep it simple (that's what this issue is all about). 
>> Get the
>>  >>  >> semantics we want for expressions, but without having to change
>>  >>  >> existing specifications or implementations.
>>  >>  >>
>>  >>  >> assaf
>>  >>  >>
>>  >>  >>
>>  >>  >> Yaron Y. Goland wrote:
>>  >>  >>
>>  >>  >>> Issue 29 - Simplification of XPath expressions
>>  >>  >>>
>>  >>  >>> Proposal: Require that all arguments to BPEL defined XPATH
>>  >> functions
>>  >>  >>> must be quoted strings that are statically defined within 
>> the XPATH
>>  >>  >>> expression.
>>  >>  >>>
>>  >>  >>> Note: This proposal won't be necessary if we decide to manifest
>>  >>  >>> properties as independent XPATH variables since this would 
>> let us
>>  >>  >>> get rid of getVariableProperty.
>>  >>  >>>
>>  >>  >>> Rationale:
>>  >>  >>>
>>  >> 
>> getVariableProperty(getVariableProperty(foo,bar),getVariableProperty(ick,bick)) 
>>
>>  >>
>>  >>  >>> is completely legal in XPATH. This is a problem because such an
>>  >>  >>> expression makes it impossible to statically analyze the 
>> expression
>>  >>  >>> that contains the function call and determine what variable is
>>  >> being
>>  >>  >>> referenced. This is problematic for static analysis of the 
>> process
>>  >>  >>> because it makes it impossible to determine what variable and
>>  >>  >>> property are being accessed so there is no way to check if that
>>  >>  >>> variable is accessible from that point in the BPEL process
>>  >>  >>> definition or if the variable has the referenced property
>>  >> defined on
>>  >>  >>> it. Even worse (in my mind anyway) is that the previous 
>> plays havoc
>>  >>  >>> with optimizations for compensation handling. If the previous
>>  >>  >>> function was contained within a compensation handler then there
>>  >>  >>> would be no way to know what variable was being accessed and 
>> so the
>>  >>  >>> BPEL process would have no choice but to persist all variables
>>  >>  >>> visible from the compensation handler, major yuck!
>>  >>  >>>     We have in the past altered XPATH in order to suit our 
>> needs.
>>  >>  >>> For example, in some BPEL expressions we have banned the use 
>> of the
>>  >>  >>> global context node, a node whose presence is actually 
>> mandated by
>>  >>  >>> the XPATH specification. If we are willing to change XPATH that
>>  >> much
>>  >>  >>> I see no reason not to change it here as well. In for a 
>> dime, in
>>  >> for
>>  >>  >>> a dollar.
>>  >>  >>>
>>  >>  >>> Section 9.1 -
>>  >>  >>>
>>  >>  >>> Add a new paragraph after the following paragraph "Any 
>> qualified
>>  >>  >>> names used within XPath expressions are resolved by using 
>> namespace
>>  >>  >>> declarations currently in scope in the WS-BPEL document at the
>>  >>  >>> location of the expression.":
>>  >>  >>>
>>  >>  >>> The arguments to all XPATH functions defined in this 
>> specification
>>  >>  >>> MUST be given as quoted strings. The previous requirement 
>> MUST be
>>  >>  >>> statically enforced. It is therefore illegal to pass into a 
>> BPEL
>>  >>  >>> XPATH function any XPATH variables, the output of XPATH
>>  >> functions, a
>>  >>  >>> XPATH location path or any other value that is not a quoted 
>> string.
>>  >>  >>> This means, for example, that 
>> getVariableProperty("varA","propB")
>>  >>  >>> meets the previous requirement while
>>  >>  >>>
>>  >> 
>> getVariableProperty($varA,string(getVariableProperty("varB","propB"))
>>  >>  >>> does not. Note that the previous requirement institutes a
>>  >>  >>> restriction which does not exist in the XPATH standard.
>>  >>  >>>
>>  >>  >>>
>>  >>  >>>
>>  >> 
>> ---------------------------------------------------------------------
>>  >>  >>> To unsubscribe, e-mail: wsbpel-unsubscribe@lists.oasis-open.org
>>  >>  >>> For additional commands, e-mail: 
>> wsbpel-help@lists.oasis-open.org
>>  >>  >>>
>>  >>  >>>
>>  >>  >>
>>  >>  >
>>  >>  >
>>  >>
>>  >
>>  >
>>
>
>


begin:vcard
fn:Assaf Arkin
n:Arkin;Assaf
org:Intalio
adr;dom:;;1000 Bridge Parkway Ste 210;Redwood City;CA;94065
email;internet:arkin@intalio.com
title:Chief Architect
tel;work:(650) 596-1800
x-mozilla-html:TRUE
url:http://www.intalio.com
version:2.1
end:vcard