RE: [wsbpel] Issue - 11 - Partial Schema Compliance

["Yaron Goland" <ygoland@bea.com>]

I could also fairly happily live with 'validation only for external
services' with the introduction of the validateVariable function but only if
we do enforce partial validation. What I mean by partial validation is that
if what you put into the variable can never result in a schema compliant
value then the assign fails.

If the schema is:
<foo><bar/><yack/></foo>

and you do an assign of <foo><blah/></foo> then the assign should fail
because the result is 100% invalid. That is different than an assign of
<foo><bar/></foo>. Technically the result is still invalid (there is no
<yack/>) but it is at least 'partially' valid. Programmers need help
building valid structures which is why partial validation is so important.

I also have to admit that I don't fully understand Ugo's proposal for using
null. Null (through the nillable declaration) is an explicit state in XML
Schema so it would seem that Ugo's proposal would require changing a
schema's definition if the schema doesn't use nillable. My suspicion is that
I just don't understand his proposal.

		Yaron

-----Original Message-----
From: Chris Keller [mailto:chris.keller@active-endpoints.com]
Sent: Wednesday, August 06, 2003 2:59 PM
To: 'Danny van der Rijn'; wsbpel@lists.oasis-open.org
Subject: RE: [wsbpel] Issue - 11 - Partial Schema Compliance


I'd vote for the validation "only when dealing with external services".
But in addition we should add a new function like
bpws:validateVariable('x').  This could return a Boolean, which would be
true if valid and false if not (or we could have it return a new bpel
type, which gave more info on what was invalid).  In this way users can
build valid messages via assigns and then decide if and when they may
need to test the data (e.g. they can put it in a switch with a throw if
the data is invalid).  But they should never receive or send out bad
data so that validation should be mandatory.

-----Original Message-----
From: Danny van der Rijn [mailto:dannyv@tibco.com]
Sent: Wednesday, August 06, 2003 5:26 PM
To: wsbpel@lists.oasis-open.org
Subject: Re: [wsbpel] Issue - 11 - Partial Schema Compliance

I was going to raise this as a separate issue, and was just beginning to
write it up.  Not sure if it should be a separate issue at this point or
not, since its existence as an issue depends on a solution to the
insertion
issue.

I think that yaron's example can be easily fixed by saying that schema
validation occurs after all <copy> elements of an <assign> activity have
been completed, and not after each <copy> clause.  So the example turns
into
something like:

<assign>
  <copy>
    <from something or another that contains the integer value>
    <to variable="Foo" part="p1" query="/foo/bar"/>
  </copy>
  <copy>
    <from something or another that contains the Icky value>
    <to variable="Foo" part="p1" query="/foo/icky"/>
  </copy>
</assign>

however, if the validation boundary (more on this in a sec) is arrived
at,
and there is a validation problem, a new standard fault needs to be
thrown:

invalidVariableContent - Thrown when the content of a variable fails
validation

it would be nice if the fault could include information about which
variable
it was.

but that only fixes your example.  here's a more difficult example:

a variable has 2 parts, both required, one of which is repeating and has
cardinality > 1.  Think of a purchase order request/response, as a
common
example

Variable: PO
Part: PO
Schema:
<PurchaseOrder>
   <PurchaseOrderID/>
   <LineItem/>+
</foo>

Variable: POResponse
Part: POResponse
Value: Uninitalized
Schema:
<PurchaseOrderResponse>
   <PurchaseOrderID/>
   <LineItemResponse/>+
</foo>

Variable: LineIn
Part: LineIn
Schema: <LineItem/>

Variable: LineOut
Part: LineOut
Schema: <LineItemResponse/>

Now imagine that the process <receive>s a PO, and loops over each line
item
to construct a response.  imagine the following code fragments (an
analogous
example inside a <sequence> is trivial to imagine):

<flow>
...
<assign name="AssignID">
  <copy>
    <from variable="PO" part="PO"
query="/PurchaseOrder/PurchaseOrderID"/>
    <to variable="POResponse" part="POResponse"
query="/PurchaseOrderResponse/PurchaseOrderID"/>
  </copy>
<assign>
...
<while "there are line items, loop over them">

  <assign>
    <copy>
      <from variable="PO" part="PO"
query="/PurchaseOrder/LineItem[loopIndex]" />
      <to variable="LineIn"/>
    </copy>
  </assign>

  <invoke operation="ProcessOrderLine" inputVariable="LineIn"
outputVariable="LineOut"/>

  <assign name="AssignResponseLine">
    <copy>
      <from variable="LineOut"/>
      <to variable="POResponse" part="POResponse"
query="/PurchaseOrderResponse/LineItemResponse[loopIndex]"/>
    </copy>
  </assign>

</while>
...
</flow>

if the AssignID <assign> is reached first, it would cause an
invalidVariableContent fault to be raised, since there are no line
items.
if, however, the AssignResponseLine <assign> is reached first, it would
cause the same, since there is no PurchaseOrderID

i don't really have a nice solution to this problem.  i do have some
thoughts, though, not that i like any of them.

- create an attribute on some (proper?) subset of activities stating
that
some list of variables (/subparts?) is to be exempt from validation
during
the execution of the activity.  then, for instance, you can wrap the
above
code fragment with an <empty> or <scope> (or whatever) that has the
attribute novalidation="POResponse/POResponse" (syntax is
variable/part?).
validation would occur upon completion of the activity.

- don't be so granular:  validation="no"  simpler to deal with for the
language, doesn't provide as much benefit to the user of a strongly
typed
system.

- implicitly allow what yaron calls "automatic incremental validation."
question then is when is full validation enforced.  i'm not sure that
yaron's proposal of "only when dealing with external services" is good
enough.

- never validate(!)

- programming convention (as per yaron) with type <any> shamelessly
overused.

BPEL makes use of a strongly typed system, but current limitations makes
it
unusable.  to paraphrase yaron:

> The issue here is one of programming model friendliness. For BPEL to
be
> successful we need a programming model that programmers will be
comfortable
> with and I'm fairly confident that [one where data is messy to deal
with]
isn't
> one programmers are going to successfully use.

danny

----- Original Message -----
From: "Yaron Goland" <ygoland@bea.com>
To: <wsbpel@lists.oasis-open.org>
Sent: Wednesday, August 06, 2003 11:19 AM
Subject: [wsbpel] Issue - 11 - Partial Schema Compliance


> A related issue is what happens if the copy results in a non-schema
> compliant value?
>
> For example:
>
> Variable: Foo
> Part: P1
> Value: Uninitialized
> Schema:
> <foo>
>    <bar>INT</bar>
>    (<blah/>|<icky/>)
> </foo>
>
> Now lets say that the programmer is building up the Foo value. Some
event
> has occurred which has told the programmer the value that bar should
have
so
> naturally the programmer wants to execute:
>
> <copy>
> <from something or another that contains the integer value>
> <to variable="Foo" part="p1" query="/foo/bar"/>
> </copy>
>
> Assuming we use the idea that queries on to elements that resolve to 0
nodes
> are inserts and assuming we accept that in cases where there are no
siblings
> of the insertion point then there is no need to use
beforeinsertionPoint
or
> afterinsertionPoint since the insertion point is unambiguous then the
result
> would be:
>
> <foo>
>    <bar>Whatever the INT value was that got copied in</bar>
> </foo>
>
> But this result violates the schema for the Foo variable which
mandates
the
> presence of either a blah or icky element in addition to the bar
element.
> Therefore the COPY would fail schema validation.
>
> It looks to me like the way to get around this problem is to define a
> variable Temp whose schema is <ANY> and then build the legal foo value
there
> and when one has successfully put everything together then one can
copy
that
> value into Foo. But doesn't that seem ham fisted to anyone? In essence
it
> means that all values are untyped until the last possible moment when
one
> has a big bang attempt to set the value and see if it is schema
compliant.
>
> Another approach would be to define a temp variable whose schema was:
>
> Variable: Temp
> Part: P1
> Value: Uninitalized
> Schema:
> <foo>
>    <bar>INT</bar>?
>    (<blah/>|<icky/>)?
> </foo>
>
> This would allow one to build up the temp value incrementally and
still
get
> schema validation. Then when one was done one could copy the result
into
> Foo.
>
> But the level of sophistication this requires on the part of the
> programmers, e.g. the ability to take all of their typed values,
analyze
> them and figure out how to re-write their schemas so as to allow for
partial
> validation, seems way beyond the norm.
>
> What I suspect will really happen is that programmers will learn to
define
> their temp variables as schema type <ANY>. They will build up their
values
> inside of the <ANY> variables and then when they are done they will
try to
> copy from the temp variable into the final location (Foo in this case)
and
> pray it works. If it doesn't they are probably out of luck since it is
> unlikely that they will be able to do much with the error message.
>
> A workable programming model demands a way to build up values
incrementally
> with feedback on validity (e.g. error: the element you just inserted
does
> not exist in the schema).
>
> What I think we need is a model where when manipulating variables one
can
> get some sort of automatic incremental validation and only when trying
to
> communicate the value externally through a Web Service message will
you
get
> full validation where the value must be perfect or you get an error.
>
> For example, in an incremental validation model the programmer could
execute
> the following assuming Foo is uninitialized:
>
> <copy>
> <from something or another that contains the integer value>
> <to variable="Foo" part="p1" query="/foo/bar"/>
> </copy>
>
> The result is that Foo would contain <foo><bar>Some INT</bar></foo>
which
is
> 'partially' valid. E.g. it doesn't fully specify the schema but it
doesn't
> contradict it either.
>
> However, in a partial schema validation model the following would be
> illegal:
>
> <copy>
> <from something or another that contains the integer value>
> <to variable="Foo" part="p1" query="/foo/bark"/>
> </copy>
>
> The reason being that /foo/bark/ does not exist in the schema and so
could
> never be valid.
>
> The issue here is one of programming model friendliness. For BPEL to
be
> successful we need a programming model that programmers will be
comfortable
> with and I'm fairly confident that a model in which all variables stay
> untyped while they are being built up until the last possible instant
isn't
> one programmers are going to successfully use.
>
> Just a thought,
>
> Yaron
>
>
> -----Original Message-----
> From: Danny van der Rijn [mailto:dannyv@tibco.com]
> Sent: Wednesday, August 06, 2003 10:08 AM
> To: wsbpel@lists.oasis-open.org
> Subject: [wsbpel] Issue - 11 Query in <to> close should allow
assigning
> to new locations
>
>
> I would like to open discussion on this issue (and volunteer to
champion
> it).
>
> The issue is that in section 14.3 dealing with assignment, there is
the
> following verbiage:
>
> "For XPath 1.0, the value of the query attribute MUST be an absolute
> locationPath (with '/' meaning the root of the document fragment
> representing the entire part). It is used to identify the root of a
subtree
> within the document fragment representing the part. <b>The location
path
> MUST select exactly one node. If the location path selects zero nodes
or
> more than one node during execution, then the standard fault
> bpws:selectionFailure MUST be thrown by a compliant
implementation.</b>"
> (emphasis added)
>
> This means that the only way to get an initialized value into a
variable
is
> to receive a message from elsewhere.  This is, I believe, far too
> restrictive.  I propose that the wording be changed to
>
>  "For XPath 1.0, the value of the query attribute MUST be an absolute
> locationPath (with '/' meaning the root of the document fragment
> representing the entire part). It is used to identify the root of a
subtree
> within the document fragment representing the part. <b>When the
expression
> is used inside a <from> element, the location path MUST select exactly
one
> node. If the location path selects zero nodes or more than one node
during
> execution, then the standard fault bpws:selectionFailure MUST be
thrown by
a
> compliant implementation.  When the expression is used inside a <to>
> element, and it selects 0 nodes, then the expression should be treated
as
> the location that the value will have after it is inserted.</b>"
>
> This probably isn't the best wording, and I'd love someone to clean it
up
a
> bit, but I think you get the general idea.
>
> Danny van der Rijn
>
>
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: wsbpel-unsubscribe@lists.oasis-open.org
> For additional commands, e-mail: wsbpel-help@lists.oasis-open.org
>
>
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: wsbpel-unsubscribe@lists.oasis-open.org
> For additional commands, e-mail: wsbpel-help@lists.oasis-open.org
>
>



---------------------------------------------------------------------
To unsubscribe, e-mail: wsbpel-unsubscribe@lists.oasis-open.org
For additional commands, e-mail: wsbpel-help@lists.oasis-open.org





---------------------------------------------------------------------
To unsubscribe, e-mail: wsbpel-unsubscribe@lists.oasis-open.org
For additional commands, e-mail: wsbpel-help@lists.oasis-open.org