Hi Rich,
I think the intent has always been to allow empty policies. It's
useful especially for automatically generated policies to be able to
generate empty policies as a special case of a general scheme for
generating a full policy set.
So I would not want to restrict the schema like your first bullet
point. Rather the change to do would be to add a minOccurs="0" to
the choice, but since it is already possible to generate a policy
with no rules in the current schema (by putting a a dummy combiner
parameter), I don't think it's a big deal.
Best regards,
Erik
On 2012-01-29 19:15, rich levinson wrote:
Hi Erik,
At this point we are simply trying to determine what the schema
should be,
in the sense that if it was reflecting the intended behavior, what
changes
would be appropriate.
For example it would seem the appropriate changes might be:
- A Policy must have one or more Rules.
- A Policy may have zero or more VariableDefinitions
- A Policy may have zero or one CombinerParameters
- A Policy may have zero or one RuleCombinerParameters
If these are the actual requirements that are intended, or if some
other specific requirements are more accurate or appropriate, then
the purpose of the issue being raised is to determine what are the
correct requirements.
Once the requirements are established, then we can evaluate what,
if any, impact that might have on the schema, and whether the
TC thinks the schema should be changed.
At present, it appears that the current schema does not reflect
any kind of plausible real requirements, and this represents
a problem for developers using the specification to build
products,
because they may be uncertain how to handle the use cases that
are made possible w the current schema.
Thanks,
Rich
On 1/27/2012 6:10 PM, Erik Rissanen wrote:
Hi Rich,
I don't think there is anything which needs to be changed here.
It's true that the schema is a bit weird in this respect. It's a
carry over from 2.0, and does not represent any practical
concern.
Empty policies are fine I think, though it does not really make
sense to have an empty policy with variables since there would
be no rule to use them.
But there is not really any issue in how to interpret any
combination which the schema allows and there is no reason for a
product to produce meaningless policies.
I would say that we keep it as it is.
Best regards,
Erik
On 2012-01-24 07:23, rich levinson wrote:
To TC:
We have been looking at the xsd for Policy and there is a
central
"choice" element that does not appear correct, although for
mainstream Policies it probably does not show up.
The choice element is the following:
2028
<xs:choice maxOccurs="unbounded">
<xs:element ref="xacml:CombinerParameters"
minOccurs="0"/>
<xs:element ref="xacml:RuleCombinerParameters"
minOccurs="0"/>
<xs:element ref="xacml:VariableDefinition"/>
<xs:element ref="xacml:Rule"/>
</xs:choice>
This is the construct that allows multiple Rules in a Policy,
which, looking at
the Rule element alone, seems ok, as default is minOccurs="1",
and it
inherits maxOccurs="unbounded" from the choice element itself.
However, very little else about this element appears to make
sense:
- Since this is a choice element, w minOccurs="1", one
could choose any
of the other 3 elements and nothing else, and the result
would be a
Policy with zero Rules. Does this make sense?
- With zero Rules, even if you used the Policy to define
VariableDefinitions,
they cannot be referenced outside the Policy, and since
there are no
Rules in the Policy, there is nothing that would ever use
the
VariableDefinitions.
- Does it make sense to have multiple instances of either
CombinerParameters
or RuleCombinerParameters? i.e. can't all the parameters
be put in one
element in both cases? If not then why are these elements
in the choice
block that allows unbounded instances?
Please advise as to whether the above interpretation is
accurate. If so, we would
like to consider raising this as an issue for action.
Thanks,
Rich
|