First off, thanks to Dieter for the examples
in his reply. They are spot on. Second some replies to Mike inline.
On 11/6/2009 2:21 AM, Mike Edwards wrote:
OFC7EB8A82.1243F002-ON80257666.00378E3E-80257666.0038B23A@uk.ibm.com"
type="cite">
Danny,
Comments inline.
Yours, Mike.
Strategist - Emerging Technologies, SCA & SDO.
Co Chair OASIS SCA Assembly TC.
IBM Hursley Park, Mail Point 146, Winchester, SO21 2JN, Great Britain.
Phone & FAX: +44-1962-818014 Mobile: +44-7802-467431
Email: mike_edwards@uk.ibm.com
From:
|
Danny van der Rijn
<dannyv@tibco.com>
|
To:
|
Mike Edwards/UK/IBM@IBMGB
|
Cc:
|
OASIS BPEL
<sca-bpel@lists.oasis-open.org>
|
Date:
|
05/11/2009 19:14
|
Subject:
|
Re: [sca-bpel] [Issue 53]
BPEL Process:
References with Multiplicity 0..1 - how are they supposed to work? |
So reading back over this, and swapping as much of it
as I can grasp back in, I'm thinking that we can't support 0..1 with a
variable, as I was leaning towards on the call.
IF we define 0..1 as in this email thread, which I think is reasonable,
then we only have to deal with Mike's 2 issues as below.
Issue 1
I'm not sure I understand the issue, to be frank. Or if it's still
relevant? Let's reopen discussion of this.
<mje>
The issue is a simple one, I believe.
If you read the WS-BPEL spec and the
SCA
BPEL spec, then at the moment, if you have a partnerLink that qualifies
as a 0..1 reference,
there is nothing that requires this
partnerLink
to be initialized.
1) I believe that the SCA BPEL
specification
should be modified to add a statement that a partnerLink introspected
as
a 0..1 reference
MUST be initialized if the partnerLink
has
a reference target. This places a requirement on any SCA BPEL runtime
to do something,
which is otherwise not stated.
<vdR>
I think I'm getting closer to understanding where you're coming
from, and I think I have to disagree with this. We actually don't
fully respect @bpel:initializePartnerRole, and perhaps we need to
examine the matrix of SBPEL2010-2013 with the 3 possible values of
@bpel:initializePartnerRole ("yes", "no", and no value.) But we MUST
respect the following statement from BPEL (6.2) whether we agree with
it or not:
If the initializePartnerRole
attribute is set to "no" then the WS-BPEL processor MUST NOT initialize
the EPR of the partnerRole before that EPR is first utilized by the
WS-BPEL process.
</vdR>
OFC7EB8A82.1243F002-ON80257666.00378E3E-80257666.0038B23A@uk.ibm.com"
type="cite">
2) For the case where there is a
partnerLink
introspected as a 0..1 reference that has no reference target (the 0
case...)
then what needs
to be done depends on how we solve
Issue
2. If there is some kind of test (as you propose) that permits the
BPEL process to work out
that there is no reference target and
thus
not to call that partnerLink, then I think that is OK.
</mje>
Issue 2
Again, I don't have a problem with dealing with exceptions. But I'm
not unsympathetic to the problem that Mike has.
<mje>
The problem I have with exceptions is
that
exceptions are fine when dealing with things that are truly exceptional
- ie not really supposed to
happen. But in BPEL, just as much as
in Java, I think it is simply painful using exceptions to handle things
that are expected - something that
you'd really expect to deal with via an
"if....then...else"
kind of construct. And worse, I think that BPEL is harder to write
for the exception
handling than Java (I'll wait to be
howled
down by the BPEL experts for that comment!) - there is nothing quite as
simple as the Java try...catch
construct.
So I am in favour of some new syntax
such
as you propose below.
</mje>
<vdR>
Sounds like we're converging on your point #2, with Dieter's
examples, etc. and we still need to either work through point #1.
</vdR>
OFC7EB8A82.1243F002-ON80257666.00378E3E-80257666.0038B23A@uk.ibm.com"
type="cite">So I'll again propose new syntax. I was
thinking
on the call that it would be an extension activity that would look
something
like:
<bpel:extensionActivity>
<sca-bpel:ifPartnerRoleSet partnerLink="pLinkName">
<bpel:activity/>
<sca-bpel:else>
<bpel:activity/>
</sca-bpel:else>
</sca-bpel:ifPartnerRoleSet partnerLink="pLinkName">
</bpel:extensionActivity>
And we could decompose the functionality into a virtual test and
catch.
However, as I'm typing that, I think that providing an sca-bpel xpath
extension
function would be far simpler. This would follow the same lexical
rules as bpel:getVariableProperty, namely:
[SA00030] The arguments to bpel:getVariableProperty
MUST
be given as quoted strings. The previous requirement MUST be statically
enforced. It is therefore illegal to pass into a WS-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 bpel:getVariableProperty("varA","b:propB") meets
the previous requirement while bpel:getVariableProperty( $varA,
string(bpel:getVariableProperty("varB",
"b:propB") ) does not. Note that the previous requirement institutes
a restriction which does not exist in the XPath standard.
sca-bpel:isPartnerRoleSet('partnerLinkName') would return true if the
role
is set, and false otherwise. Still explain in spec as virtual test
and catch. Or leave that as an exercise to the reader.
<mje>
Could you show an example of this
second
case please?
I think I prefer this second case, but
I'd
like to see it "in action" before I pass judgement...
The first case works, as you show
above,
but it is a tad cumbersome - and if we can use the standard BPEL
if...else construct instead, I think
that
would be simpler and less cumbersome. Placing requirements on the
SCA BPEL runtime for a special XPath
function
should not be too much of a burden I would have thought.
</mje>
<vdR>
I agree that it's not much of a burden, considering especially that
BPEL itself already imposes a few required XPath functions. I
recommend, however, that everyone checks their implementations to see
how much of a burden it would be to support this particular XPath
function.
</vdR>
OFC7EB8A82.1243F002-ON80257666.00378E3E-80257666.0038B23A@uk.ibm.com"
type="cite">Danny
On 8/13/2009 8:24 AM, Mike Edwards wrote:
Folks,
This is all good stuff, but I'd still like us to turn our attention to
the original question of this issue.
How are 0..1 multiplicity references supposed to work?
The text below clarifies the circumstances under which the process will
be introspected to find a 0..1 reference.
But that still does not mean that there is a good description of how
they
work.
Issue 1:
Given that the BPEL spec does not require initialization of a
partnerLink
configured in the way stated to be interpreted as
a 0..1 multiplicity reference, how is any value of a target specified
in
SCA configuration for that reference supposed to
get into the partnerLink variable?
- are we going to make a statement that the variable MUST be
initialized
(removing the MAY wording of the BPEL spec)?
- are we going to a supply the value of the reference in some other way
so that he process can initialize the partnerLink variable
(I am thinking of a procedure similar to that of the handling of 0..n
and
1..n references)
Issue 2:
Given that a 0..1 multiplicity reference can be configured at the SCA
level
with 0 targets, how is the BPEL process supposed to
find out that there are no targets? It is not possible to inspect
the contents of a partnerLink directly from the BPEL process.
I heard one approach mentioned in previous discussions - let the
invocation
of the partnerLink take place and fail with a
fault and deal with the fault. I must say that I don't like this
procedure at all - faults should be used for genuine problem
situations. A 0..1 reference that is unwired is not a problem - it
is expected.
One possibility is to have the partnerLink initialized but without an
EPR
present - I believe that this CAN be tested for and
also makes no presumption about the form of the epr - the partnerLink
would
contain the <sref/> with no child elements,
which can be tested for by the BPEL process fairly simply.
Yours, Mike.
Strategist - Emerging Technologies, SCA & SDO.
Co Chair OASIS SCA Assembly TC.
IBM Hursley Park, Mail Point 146, Winchester, SO21 2JN, Great Britain.
Phone & FAX: +44-1962-818014 Mobile: +44-7802-467431
Email: mike_edwards@uk.ibm.com
I found more problems with that section:
The 1st bullet in (Multi-Reference) is wrong. It says -
Multi-Reference. [SBPEL2010]
If the partner link is declared with an sca-bpel:multiRefFrom="aVariableName"
extension, the multiplicity of the SCA reference MUST be determined by
the multiplicity attribute of sca-bpel:multiReference
extension used in the corresponding variable. Details of these
extensions
are described in section ŭ3.2.
It is not the the partnerLink that determines this, it is the variable.
It should instead say:
Multi-Reference. [SBPEL2010]
If the variable is declared with an sca-bpel:multiReference
extension, the multiplicity of the SCA reference MUST be determined by
the multiplicity attribute of sca-bpel:multiReference
extension. Details of these extensions are described in section 3.2
This is one of the reasons I thought it was the partnerLink with
multiRefFrom
that was exposed as a reference and not the variable, hence my initial
proposal for issue 52 was crafted the way it was.
-Anish
--
Michael Rowley wrote:
Proposal:
Change the sentences in section 2.1.1 which currently reads:
3. Stub
Reference. [SBPEL2012]
If neither [SBPEL2010]
nor [SBPEL2011]
apply and the analysis of the process determines that the first use of
the partner link by any activity is in an <assign>
activity that sets the partner role, then the multiplicity MUST be "0..1"
and the attribute wiredByImpl
MUST be set to "true".
to instead read:
3. Stub
Reference. [SBPEL2012]
If [SBPEL2010]
does not apply and the partner link has initializePartnerRole="no",
then
the multiplicity MUST be "0..1"
and the attribute wiredByImpl
MUST be set to "true".
Note that (4), which follows this, will turn any other partnerLink into
a reference with multiplicity of 0..1 (not wiredByImpl).
Michael
-----Original Message-----
From: Anish Karmarkar [mailto:Anish.Karmarkar@oracle.com]
Sent: Thursday, August 06, 2009 3:39 PM
To: OASIS BPEL
Subject: [sca-bpel] [Issue 53] BPEL Process: References with
Multiplicity
0..1 - how are they supposed to work?
This is now issue 53
http://osoa.org/jira/browse/BPEL-53
Mike Edwards wrote:
>
> Raiser: Mike
Edwards
>
> Target:
>
> Description:
>
>
> Multi-valued references (ie 0..n and 1..n) are discussed in the
> specification at some length, particularly in section
> 3.2. Unfortunately, the case of 0..1 references is mentioned
only in
> passing, although it has some significant
> matters that need discussion and clarification.
>
> Section 2.1.1 makes it clear that 0..1 multiplicity references do
exist
> for BPEL processes.
>
> Basically, it states that where a partnerLink is determined to be
a
> reference, it will be 0..1 if @initializePartnerRole="no"
> OR @initializePartnerRole is omitted. It must also be the case
that the
> partnerLink must not have its "first use" as the
> target of an <assign/> operation.
>
> Now, reading the BPEL 2.0 specification:
>
> @initializePartnerRole="no" means that the
BPEL Processor MUST NOT
> initialize the partnerLink variable.
>
> @initializePartnerRole omitted means that the BPEL Processor
MAY
> initialize the partnerLink variable.
>
>
> Now, the problems are these:
>
> A) If the partnerLink is NOT initialized, then it is essential
that it
> is initialized by the BPEL process itself, since any
> attempt to use that partnerLink variable will otherwise result in
a fault.
>
> So, if the partnerLink is marked @initializePartnerRole="no",
the
> partnerLink must be supplied with a value - but
> how is such a value supplied to the BPEL process by SCA? The
SCA BPEL
> spec talks about the way in which
> serviceReferences are supplied in a variable typed as a
> serviceReferenceList for multi-valued references.
> There is no such equivalent for 0..1 references.
>
> Even if the partnerLink is not marked with @initializePartnerRole
at
> all, it is STILL possible for the partnerLink to
> be uninitialized and so unusable without being assigned a value by
the
> BPEL process - this is the result of that
> "MAY" in the BPEL 2.0 spec, listed above. Again, there
is no means by
> which the BPEL process can get a
> serviceReference to assign into the partnerLink.
>
>
> B) If the partnerLink IS initialized, how can the BPEL process
determine
> the difference between the case where the
> reference is wired and the case where the reference is unwired?
>
> There is no standard means for doing this in BPEL.
>
> So, I think we may have a problem with 0..1 references in the SCA
BPEL
> spec.
>
> Proposal:
>
> None
>
>
> Yours, Mike.
>
> Strategist - Emerging Technologies, SCA & SDO.
> Co Chair OASIS SCA Assembly TC.
> IBM Hursley Park, Mail Point 146, Winchester, SO21 2JN, Great
Britain.
> Phone & FAX: +44-1962-818014 Mobile: +44-7802-467431
> Email: mike_edwards@uk.ibm.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/
>
>
>
>
>
>
---------------------------------------------------------------------
To unsubscribe from this mail list, you must leave the OASIS TC that
generates this mail. Follow this link to all your TCs in OASIS at:
https://www.oasis-open.org/apps/org/workgroup/portal/my_workgroups.php
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
|