Re: [wsbpel] Issue - 301 - Uninitialized Partner Links

["Monica J. Martin" <Monica.Martin@Sun.COM>]

Dieter,
Have we (Ron and I) missed your response so we can prepare for 
tomorrow's discussion on #301?  Can we proceed to define equivalent 
language discussing how uninitialized partnerLinks are to be handled? 
Otherwise, we leave it to implementation where different approaches may 
be taken. Thanks.

> koenig: I have not yet fully understood the issue 301.
> My picture is as follows:
>
> A. Variables and partner links are allowed to be declared but not used
>   - Note this is also true for correlation sets and message exchanges 

ten-hove: Agreed.

> B. Variables and partner roles of partner links are *used* when their 
> value
> is *read*
>   - Variables are used when they are referenced in <from> / <invoke> /
> <reply>
>   - Partner roles of partner links are used when they are referenced in
> <from> / <invoke>
>
> C. Variables and partner roles of partner links can be uninitialized
>   - If they are *used* anyway then
> uninitializedVariable/uninitializedPartnerRole is thrown
>
> Questions:
>
> 1. Do we have agreement on A. & B. & C.?
> 2. If yes (1.), which part is left unclear by the spec? 

ten-hove: The expected behaviour is unspecified. Consider the following:

   <assign>
     <copy>
       <from partnerLink="foo" endpointReference="partnerRole"/>
       <to partnerLink="bar"/>
     </copy>
   </assign>

when the source partnerLink "foo" is uninitialized.  According to 
section 8.1 (Variables):

   An attempt during process execution to read a variable or, in the
   case of a message type variable, a part of a variable before it is
   initialized MUST result in the standard bpel:uninitializedVariable
   fault.

There is no equivalent language discussing how uninitialized 
partnerLinks are to be handled. The <copy> could cause a fault (but what 
kind?), or the copy could uninitialize the partnerRole of the 
partnerLink named "bar", and wait until the process attempts to 
reference the partnerRole of "bar" (and throws a 
uninitializedPartnerRole fault).

The specification does not answer these questions; implementors must 
choose, presumably by suiting their own tastes. Do we want to have such 
variability in implementations? As it stands, given the scenario 
outlined above, we could have three different behaviours, all for good 
reasons:

   * The <copy> throws an uninitializedVariable fault. (The entire
     partnerLink is uninitialized)
   * The <copy> throws an uninitializedPartnerRole fault (the <from>
     role isn't initialized)
   * The <copy> doesn't throw a fault at all.

Are we happy with this variability? This seems worthy of opening an 
issue to discuss.

-Ron