Re: [wsbpel] Issue - 166 - Let's try it again

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

I have reviewed the mails on this thread and below I map out the 
positions I have seen various people take:

*--Assign has ACID semantics (names of supporters)
  |  |
  |  ---- Yaron's Goal Based Proposal [1] - (Yaron & Ron)
  |  |
  |  ---- Ron's Detailed Definition [2] - (Frank)
  |  |
  |  ---- Current Language Is Fine [3] - (Dieter, Andrew)
  |
  --Assign is only atomic regarding the output, not ACID
     |
     ---- Current Language Is Fine [4] - (Danny)
     |
     ---- We need to add nested serializable scopes [5] -
              (Suggested by Alex but not necessarily supported by him)

I believe that Assaf and Bernd fall into the first branch of the tree 
(Assign is ACID) but I haven't seen them support any particular proposal.

Alex hasn't come down for any particular proposal so I don't know where 
in the tree he falls.

That means 7 people believe that assign is ACID but can't agree on how 
that should be expressed in the spec and 1 person who believes that 
assign is atomic only with respect to the output part of the assign 
operation.

Should we have a straw poll on the question of ACID versus atomic or can 
we agree that the majority believes that assign has ACID semantics and 
move forward on that basis?

If we can move forward on that basis then the issue with the current 
language becomes editorial and we can close issue 166 and ask the 
editors team to make the spec language more explicit.

		Yaron

[1] http://lists.oasis-open.org/archives/wsbpel/200410/msg00011.html and 
http://lists.oasis-open.org/archives/wsbpel/200410/msg00046.html
[2] http://lists.oasis-open.org/archives/wsbpel/200410/msg00015.html and 
http://lists.oasis-open.org/archives/wsbpel/200410/msg00042.html
[3] http://lists.oasis-open.org/archives/wsbpel/200410/msg00016.html
[4] http://lists.oasis-open.org/archives/wsbpel/200409/msg00207.html
[5] http://lists.oasis-open.org/archives/wsbpel/200410/msg00043.html

Ron Ten-Hove wrote:
> 
> 
> Yaron Y. Goland wrote:
> 
>  > From the perspective of the assign it should look to it "as if" it is
>  > the only activity running. So long as the behavior of the process
>  > provides that illusion then the process is compliant.
>  >
>  > As you point out the actual requirements to make this happen do not
>  > require any grand serialization. A little strategic locking can quite
>  > nicely handle matters. But that is an implementation specific detail.
> 
> Upon further thought, and consultations with my developers, I concur.
> The language you propose is a) simple, and b) leaves room for "clever"
> implementations to add those extra "ilities" that are necessary in
> commercial products.
> 
>  >
>  > The point is that the standards language provides a goal and then
>  > leaves it to the implementer to figure out the best way to meet that
>  > goal. That's why I didn't dive into the details.
> 
> Thanks for indulging my appetite for details, in this brief digression.
> I don't think the spec needs to go any further than what you have proposed.
> 
> -Ron
>