Rationale for encouraging POP handles be UIDs

[Michael Freedman <michael.freedman@oracle.com>]

I have asked us to strongly suggest/require POP handles be unique.  By 
unique I mean that a consumer can use the value of a POP handle to match 
POPs across registrations.  I.e. if two registrations that a consumer 
thinks are related [i.e. the same type of producer] offers the same POP 
handle, then the consumer can equate the POP in one registration to the 
other.

Here is the use case/reason why consumers need this support/clarification:
Portlets are beginning to move beyond typical usage where they are 
integrated into a Portal application to being viewed more generally as 
UI components that can be integrated into any ole web application.  This 
transformation changes the deployment model considerably.

Today many portal style deployments follow a model where the portal 
product is by and large deployed as a tool.  Developers then use the 
Portal [tool] to register their content producers, construct their 
portal application and then publish their portal application to 
appropriate sites.

When portlets are used as UI components in web applications, however, 
the web application is packaged and deployed not as a tool but a 
running/working application.  During the development process [of the web 
application] content producers are registered, portlets included and 
likely customized.  This application is then packaged into a 
distributable entity, sometimes with and sometimes without packages 
representing the producers.  At deployment, the intent is to recreate 
the application including its connections with its producers/portlets.  
Often this later requires some deployment intervention, specifically in 
the case that the referenced producers are to be migrated to new 
installations [of that producer].  For example, a packaged application 
may have a reference to the many portlets its has created from using 
Producer XXX at location http://yyy [this reference was established in 
the development environment].  The application is being deployed by a 
customer who independently installs Producer XXX at location 
http://zzz.  During the deployment process the deployer is asked if they 
want to modify the connection information for Producer XXX --  and as a 
local installation is being used the deployer says yes and changes the 
URL reference from http://yyy to http://zzz.  Following this, the 
deployment process auto-egisters the producer [at location http://zzz] 
using registration information captured when the application was 
developed, communicates to the new producer to import its portlets and 
finishes up. 

So why the need for unique POP handles?  The answer lies in that its 
reasonable to expect such consumer applications to manage [its cache of] 
portlet meta data in an efficient manner by relying on a single record 
for all the portlets derived from a common POP.  In such an 
implementation, the consumer needs a mechanism for rewiring these 
references when a producer connection is retargeted.  I.e. at deployment 
a new registration happens resulting in a new set of POPs and associated 
portlet meta data being read into the application cache.  The 
application now wants to ensure that all its portlets from that producer 
are properly rewired to point to this new meta data. Somewhere along the 
line the consumer is going to need to figure out how POPs in the 
packaged application relate to the POPs in the actual deployment. [i.e. 
either from the POP point of view to figure out which meta data record 
this portlet description shoudl overwrite or from the cloned portlet 
point of view if the cloned portlet indirectly references its 
description via a POP reference/key].  Being able to rely on the strong 
suggestion/requirement that POeP handles are unique minimizes th guess 
work involved.  I.e. trying to implement a heuristic that matches meta 
data values to identify the relationship.
   -Mike-