|
Ron,
Assigning a UUID to a process opens the door to the
problems related to change management, versioning and life cycle
management. Unless we decide to get down to the level of what a BPEL deployment
unit looks like, I am not sure that adding UUID brings any
value.
Edwin
Satish Thatte wrote:
See
below
[Ten-Hove] A process category may fall
mostly into the "deployment" bucket, and thus be ultra vires. Alternatively, perhaps it
would be useful to give each process type a globally unique
identifier.
[Satish
Thatte] It is easy to manufacture a unique QName from the targetNamespace
and NCName of the process -- one may claim that this is in fact implied by
the usual semantics of
targetNamespaces. [Ten-Hove] This
technique would work, but it is a convention, not a guaranteed property of
process definitions. Unless one has complete control of tools, deployment and
run-time, a convention cannot be used to guarantee this uniqueness
property.
As an alternative to an explicit uuid attribute, we may try
adding wording requiring the convention you suggest. Something like:
The name and
targetNamespace of the process must be globally
unique. If an existing process is modified, the name and/or tns must be
modified to ensure that the new version of the process has a unique
identity, such that it cannot be confused for the old.
I don't
particularly like this -- it sounds a lot like a hack, and makes it sound like
we never heard of UUIDs! But personal tastes aside, does this sound
reasonable?
Cheers,
-Ron
|