Proposal on name/identity discussion

[Alan Conway <aconway@redhat.com>]

---- proposed changes:

2.5.1: Remove "name" from required attribures. Keep "type" and "identity"

New text:

Entities can always be identified by the unique "identity" attribute, which is
provided by the implementation.

An implementation MAY also provide other identifying attributes (such as "name")
that can be used to identify entities. Such attributes are referred to as "index
attributes". The existence of more index attributes is part of the definition of
an entity type, like any other attribute.

The value of index attributes may be provided by the implementation, or may be
required from the user in the CREATE request. Index attributes may be read-only
or modifiable. This is all part of the definition of an entity type, like any
other attribute.

Index attributes are only special in that they can be used to identify entities
in READ, UPDATE and DELETE requests instead of the special "identity" attribute.

3.3.1.1: Create request: remove name from app properties.

3.3.* READ, UPDATE, DELETE:

Replace the "name" property with:

index (string) [exactly one of index and identity is required]
The name of an index attribute to be used instead of "identity" to identify the entity.

key   (string) [required with index]
The value of the index attribute to be used to identify the entity.

----

Justification: The idea is to 
- not force "name" on implementations that don't need/want it.
- allow implementations that want "name" to use it with minimal changes.
- provide additional flexibility for implementations that want identifying
  attributes other than a universal user-supplied "name", e.g. only on selected
  entity types or provided by system not user or even multiple index attributes on
  some entities.
- not force extra uniqueness or search requirements on implementations unless
  they opt for them. The whole index attribute concept is optional, things like
  multi-index, different indexes per entity type etc. are optional.

So for an implementation that wants the old "name" attribute behavior:
- All entity types are defined with a required, no-default, string valued attribute "name" (they already do)
- The implementation must ensure that user-provided names are unique (it already does)
- Only change: you must say READ:{index="name", key="foo"} instead of READ:{name="foo"}

Note we COULD make this fully backward compatible by assuming that any property
other than "identity" in a READ/UPDATE/DELETE request is the name of an index
attribute.  I would be OK with that, but I expect many might find it too loose.



A simpler proposal (which I don't like but would make my present implementaiton compliant :) is to simply remove the requirement to pass "name" on create and have it default to the same value as "identity".
That makes "name" a useless appendix for me, but at least it is not a problem - forcing my users to come up with names for things they have no reason to give names to.