RE: [xacml] New issue: XACML 3.0 higher order bag functions may have significant problems

["Tyson, Paul H" <PTyson@bellhelicopter.textron.com>]

Here’s an example showing what the 3.0 definition of any-of can do but the 2.0 couldn’t.

 

To test whether any of a bag of time values is within a certain range, we can say:

 

any-of(‘time-in-range’, {05:00,06:00, 07:00}, 06:30, 14:30)

 

There are 4 arguments to ‘any-of’:

1.       The function name, ‘time-in-range’

2.       A bag of time values

3.       A single time value marking the beginning of the time range.

4.       A single time value marking the end of the time range.

 

This will be evaluated as 3 calls to ‘time-in-range’:

                time-in-range(05:00,06:30,14:30)

                time-in-range(06:00,06:30,14:30)

                time-in-range(07:00,06:30,14:30)

 

The last of these will return ‘true’, so the any-of function will return true.

 

Time-in-range is the only built-in Boolean function of more than 2 arguments, but Jan could probably supply more examples from GeoXACML.

 

The extended functionality is not everything you would want if you were used to  Lisp or some similar language, but I think it is a useful extension.  However, the 3.0 function needs to be re-identified so 1.0 processors can remain compliant.

 

Regards,

--Paul

 

From: xacml@lists.oasis-open.org [mailto:xacml@lists.oasis-open.org] On Behalf Of rich levinson
Sent: Thursday, 12 January, 2012 15:56
To: herrmanj@informatik.tu-muenchen.de
Cc: Erik Rissanen; xacml@lists.oasis-open.org
Subject: Re: [xacml] New issue: XACML 3.0 higher order bag functions may have significant problems

 

Hi Jan,

It sounds to me that your comment is that we need a multi-datatype
functionality. My understanding is that at present "bags" are only
of a single datatype, as described in section 7.3.2 "Attribute Bags":

"3257 There SHALL be no notion of a bag containing bags,
or a bag containing values of differing types;
i.e., a bag in XACML SHALL contain only values that
are of the same data-type."

Given this restriction, I do not see the value of providing a list of
parameters of differing datatypes to test against the bag of a
specific datatype.

For example, if I have a bag of "apples", then it makes sense
to ask:

Is "apple-01" in this bag?
(I think the any-of (fcn-id, apple-id, apple-bag) fcn satisfies that)

It also makes sense to ask:

Is at least one of "apple-01", "apple-02", "apple-03" in this bag?
(I think the any-of-any(fcn-id, apple-bag-1, apple-bag2) satisfies that)
(and therefore there is no need for:

any-of(fcn-id, apple-id1, appleid2, appleid3, apple-bag)

It sounds to me like the multi-data type suggestion would be like this:

any-of(fcn-id, "apple-01", "orange-02", "lemon-03", apple-bag)

This does not seem to make a lot of sense to me. Even though you can't
put apple, orange, lemon in one bag to use the any-of-any fcn, I see no
strong motivation to ask if a specific orange is in the apple-bag.

I do think having a single apple vs an apple-bag is useful, because everyone
is not going to want to take the step of putting their single apple into a bag
to do an apple to apple comparison. This is analogous to testing if a String
is in a String[], vs having to put the single String in a 1 member String array.

Apologies for the somewhat overly simplistic representation of the problem,
but I am trying to get at what the core of the new functionality is, and either
I am still missing something either in the functionality itself or in the
possible value of it. However, in these simplistic terms, it does not appear
to me to warrant changing 2.0 to move in this direction, at least for this
specific "any-of" functionality.

  Thanks,
  Rich

 

On 1/12/2012 4:55 AM, herrmanj@informatik.tu-muenchen.de wrote:

Hi Rich,

see inline.

Best regards

Jan

Hi Erik,

I expected that there was background material that provided the

motivation,

however, looking at the result, it seems to me that the "any-of" fcn has

been expanded in an unnecessary way, since it appears to me that the

change of the 2nd parameter in 2.0 to be n-1 parameters in 3.0, really

is providing individual elements in separate parameters that could just

as well be provided in a bag as they already are in the "any-of-any" fcn.

Aggregating the arguments of n-arry functions in a bag of datatype-xy only

works if they all have the same datatype wich is in general not the case

(e.g. is-within-distance(polygonA, polygonB, 500meters)).

Also, the "any-of" is ref'd in some of the other fcns, presumably as

being an "atomic" one element 2nd parameter fcn. That ref'ing I expect

becomes less meaningful as the 2nd parameter expands to effectively

be a bag implemented as multiple parameters.

It looks to me as if what might have seemed like a good idea at the

time may have emerged as being duplication of existing functionality

by effectively providing a less usable alternative to a bag as an

n-1 list of individual parameters.

I think in the email you ref'd when you said:

    "With this definition the any-of function actually becomes

    a special case of any-of-any, with one primitive type argument

    and one bag argument."

that you and I are possibly in technical agreement, although we may have

different perspectives on it.

Unless there has been some real functionality added w this n-1 construct,

my recommendation would be to fall back to the 2.0 impl and remove

the n-1 parameter construct in 3.0.

due to the limitation mentioned above i disagree and would recommend to

keep the more flexible definition as it is now.

   Thanks,

   Rich

On 1/11/2012 11:49 PM, Erik Rissanen wrote:

Hi Rich,

Please see this post:

http://lists.oasis-open.org/archives/xacml/200912/msg00087.html

(And other discussion on this issue.)

Best regards,

Erik

On 2012-01-12 00:19, rich levinson wrote:

I was perusing thru section A.3.12 today, looking at some of the

details

of the higher order bag functions, and it  appears to me that there

have

been some changes made in 3.0 that do not look right.

For example, let's compare the 2.0  and 3.0 definitions of "any-of".

Some of the

diffs are subtle, so I have bolded, italicized the points I want to

call attention to.

Note: I have tried to line up the paragraphs line by line of the 2

defns. Do not

pay too much attention to the line numbers as they are sometimes off by

1

from the original because of quirks in copy and paste.

Here are the points about the 2.0 vs 3.0 definitions:

   1. Both 2.0 and 3.0 have the same name: /*xacml:1.0:function:any-of,

      */therefore, one would expect the functions to be identical. I

think

      we will find that the are not only not identical, but have some

      very strange differences.

   2. There is a difference in the hi-level defn of the fcns, where 2.0

uses

      the singular (compares 1 value to bag of values), and 3.0 uses

the plural

      (compares multiple values to one bag of values).

      It appears that the change to plural is intentional, as will be

evidenced

      in the points below, however, at this stage, it would appear that

at

      the very minimum that the "name" of the 3.0 fcn should be

different

      than the name of the 2.0 fcn.

   3. The most important change is that the 2nd argument of 2.0, has

become

      "n-1" arguments in 3.0. i.e. the "signature" of "any-of" has

changed from

      3 arguments to "n+1" arguments, because of the expansion of the

2nd

      argument. Aside from this change in quantity of arguments, I

believe

      the semantics have also changed, and maybe not in a good way.

   4. In 2.0, the function was effectively:

          "Does the 2nd argument appear in the bag in the 3rd

argument?"

      In 3.0, this function appears to have become:

          "Does the 2nd argument appear in the bag in the last

argument?

            Does the 3rd argument appear in the bag in the last

argument?

             ...

            Does the nth argument appear in the bag in the last

argument?

            If any of the above answers are "yes" then return true, ow

false."

      We do not know on a true return which of the 2nd thru nth args

      matched something in the bag. Maybe all of them matched

      something, maybe only one, or maybe some number in between.

   5. Based on prev bullet 4, this new variation on the "any-of" fcn

      begins to look suspiciously like the "any-of-any" fcn that

compares

      2 bags looking for anything in bag 1 that matches anything in bag

2.

      Therefore, it appears to me that the 3.0 defn of "any-of"

actually

      "implements" "any-of-any", which seems pointless.

I think there may be other similar issues in this section, but this one

is the only I have had time to look at in detail.

As usual, I am aware I may be missing something, in which case,

in advance, "Never mind" :).

The text of 2.0 and 3.0 follows:

  Thanks,

  Rich

First, here is*2.0*:

    4558 • urn:oasis:names:tc:/*xacml:1.0:function:any-of*/

    4559 This function applies a Boolean function between /*a specific

primitive value*/ and a bag of

    4560 values, and SHALL return "True" if and only if the predicate

is "True" for at least one

    4561 element of the bag.

    4562 This function SHALL take three arguments.

              The first argument SHALL be an <xacml:Function> element

that names a Boolean function

                that takes/*two arguments*/ of primitive types.

    /*          The second argument SHALL be a value of a primitive

data-type.

              The third argument SHALL be a bag of a primitive

data-type. */

              The _expression_ SHALL be evaluated as if the function

named in the <xacml:Function> argument

              were applied to the second argument and each element of

the third argument (the bag)

              and the results are combined with

“urn:oasis:names:tc:xacml:1.0:function:or”.

Now, here is*3.0*:

    4680ï‚· urn:oasis:names:tc:/*xacml:1.0:function:any-of*/

    4681 This function applies a Boolean function between /*specific

primitive values*/ and a bag of

    4682 values, and SHALL return "True" if and only if the predicate

is "True" for at least one

    4683 element of the bag.

              This function SHALL take n+1 arguments, where n is one or

greater.

              The first argument SHALL be an <Function> element that

names a Boolean function

                that takes/*n arguments*/ of primitive types.

    /*          Under the remaining n arguments, n-1 parameters SHALL

be values of primitive data-types and

              one SHALL be a bag of a primitive data-type.*/

              The _expression_ SHALL be evaluated as if the function

named in the <Function> argument

              were applied to the n-1 non-bag arguments and each

element of the bag argument

              and the results are combined with

“urn:oasis:names:tc:xacml:1.0:function:or”.