Re: Simplifying the meaning of assertions and wsp:optional


Looks like I'm already -3 due to big -1 from Umit :-)
Seriously though, I feel we're getting sidetracked again. Yakov, This is a very interesting and useful message, thanks, finally I actually started understanding better what you were talking about in the other email...

The thing is I've never ever proposed to drop optionality. I proposed to *drop wsp:optional*. We treat wsp:optional as being equal to Optionality, as if without it optionality doesn't exist or indeed we can't do policies without it. It's completely not true.

How would you interpret the following normal expression :

Isn't it obvious we have <foo/> as an optional assertion ? wsp:optional is just a shortcut, it's just a little helper for a compact form editor. But yet we look at it as some special attribute which brings some additional semantics we can't live without, etc, etc...What is this complexity for ? One can easily express optionality without it...

I have a very simple, basic and valid case IMHO. I just want my service be searcheable. No sophisticated scenarious, just want my service be selectable. I'd like to say that my service is the most responsive service and I'd like those who understand to use it as a base for the selection. 
I don't want people saying that it's completely wrong to say <oasis:fastestToReply wsp:optional="true"/>, because according to the current wording it's indeed looks strange... 

So please don't regard my proposal as a proposal to drop Optionality. The proposal was about improving the wording for wsp:optional or dropping wsp:optional.

But as I said in my reply to Umit, I'd happy if a wording is changed appropriately, so that a policy author can freely advertize assertions like <oasis:fastestToReply/> without feeling wsp:optional is misused...

Enjoy, Sergey

  +1 to Umit. I think the optionally should stay.


  Just summarizing the problem and also responding on the different email threads related to the issue 3564: "Optional Assertions may not be usable." I think that trying to interpret the Framework policies in the context of the requester/provider and requirements/capabilities, and to separate policy semantics with regard to requester and provider, seems to be counterproductive because of the three fundamental problems:

  1. Presence of intermediaries between requester and provider (always)

  2. Multiple entities, requester and provider consist of (almost always)

  3. Bidirectional nature of messages/transactions (always)


  #1 and #2 were already illustrated by the HTTPS/HTTP example, which clearly shows that a "requirement" may be imposed on the requester without taking into account the provider's "capabilities"


  Let's look at another typical example (authentication domain) to show that the provider capabilities may not necessarily impose any requirements on the requester: 

  A provider's policy may specify WS-Security token with the SAML profile as the only acceptable authentication scheme. Let's assume that the client application "knows" about the policy. Does this mean that the requester must generate the token and attach it to the request? I don't think so. Some intermediary may be involved which would generate the token from credentials provided by the requester. One can argue that the intermediary becomes the actual "requester" but not the client app, and after this we can get into a discussion about whether the intermediary is actually a "provider" for the client app. Just to make it a little bit more complicated - in the context of the same scenario we can specify a second policy for the intermediary to generate the WS-Security token. So the intermediary definitely becomes the requester and the provider at the same time. 


  As for #3, I think the REST scenario proves the same point - that trying to resolve these definition issues above just unnecessarily complicates things.


  In my previous email I stated that the requester/provider and requirements/capabilities paradigm may be applicable to some policy domains. At this point, I don't see which policy domain requires this. The problem with wsp:optional only exists in the "requester/provider/requirements/capabilities" context. Using the 'behavior' and 'subject' concepts in conjunction with creating multiple assertions, alternatives and policies, and using the specification's merge and intersection mechanisms, should be sufficient to support the wide variety of use cases. The issue of a policy covering one or more entities (and possible relationship between the entities) should be (and can be) handled at the policy domain level and/or by the enterprise architecture.


  Email threads - I apologize in advance if I missed somebody:




  Yakov Sverdlov





  From: [] On Behalf Of Yalcinalp, Umit
  Sent: Sunday, October 01, 2006 8:37 PM
  To: Sergey Beryozkin;
  Subject: RE: Simplifying the meaning of assertions and wsp:optional 


  A Big -1 to dropping Optionality. It is completely backwards incompatible with the current practice and existing assertions. Our charter indicates that we should retain backwards compatibility as a goal. 


  Explaination of what it is should not require us to drop the functionality. 


  I will write more about some wording suggestings in a different email.  






    From: [] On Behalf Of Sergey Beryozkin
    Sent: Friday, Sep 29, 2006 9:05 AM
    Subject: Simplifying the meaning of assertions and wsp:optional 

    Hi there


    After reading and reflecting on a lot of interesting messages on what wsp:optional means, on what is the differences between requirements and capabilities are and what provideres and requestors should do about various types of assertions, I'd like to offer to your attention a modest proposal on simplifying the way assertions and wsp:optional are covered in the WS-Policy Framework and primer/policy guidelines. This is all really based on what I've learnt form the others while reading those emails and from the spec... 


    The following is how (in a simplistic way) we might want to talk about assertions and wsp:optional.


    1. Any policy assertion, either marked as optional or not, is first and foremost is a requirement. It's a requirement to a requester to understand it and do something about it. What is it that a requester should do is part of a policy assertion documentation/spec. 

    In many cases a requirement would require a requester to make a commitment to engage in some kind of activity (direct or out-of-band) during the interaction.

    Alternatively, a requirement can simply be "to understand it and use it for choosing a given provider among other providers". In a sense, a requester should do nothing here but use it during the selection process, and this in itself encapsulates a requirement : an ability to use it during the selction process.


    Requirement is a capability and should be used interchangeably. It's a capability because it is something a provider knows about and can do something about. It's a requirement because a client needs to do something about it (engaing in a behavior, using for selction, etc )


    2. Assertion may or may not be optional. This *only* means that a requester is given an option to ignore a given requirement. No assertion can be ignored by a provider. Provider is guaranteed to support all assertions. Optionality is something which is only for a requester to worry about.


    3. Assertions must be understood by both parties. Spec says about it already, but it's worth highlighting it.


    Given above 3 points, we can state that an assertion like sp:HttpsToken and oasis:replyInTenSecs are equal WS-Policy citizens because in both cases there's something a client can do about them. In the former case

    a client will understand that it needs to use HTTPs in order to be able to talk a provider. In the latter case a client will understand that a service is very responsive and might use this fact as a basis for choosing this provider among others.


    Now about wsp:optional (based on above 3 points).


    Two options are proposed :

    Option1. Retain wsp:optional but explain that wsp:optional is just a syntactic shortcut to nominate a requirement as being optional for a requester to understand/do anything about. As well captured elsewhere, at the moment an optional "capability" in a compact form suddenly becomes a requirement in a normal form. This is confusing. wsp:optional is a way to nominate an optional requirement. 


    Option2. (Preferred) Drop wsp:optional altogether. Why ? Because IMHO it doesn't bring anything useful but the complexity. It makes it more complex for a policy engine to convert a policy alternative from a compact form to a normal one, and for policy authors to understand how to work with it and when it's appropriate to use it. 

    Lets explain clearly that for a given assertion/requirement be optional it should be avalable in one alternative and not in the other one and this is all... It will add a bt more work for a policy author, but IMHO not a lot.


    Finally :


    Point 2 above refers to oasis:replyIn10Secs assertion. A client can not really do  something about it as far as the interaction is concerned, but it still can do something about it : use it to select a provider, for ex. For such assertions not to interfere with those policy-aware clients which are not aware of what oasis:replyIn10Secs means and which may or may not have some priory polic requirements, we should recommend that when possible, policy authors should attempt to give an option to ignore such assertions by using policy alternatives as appropriate. Note no 'optional' qualifier is used here :-)


    So that's it... Does it make it a bit simplier ? Criticize away please :-)




    Sergey Beryozkin

    Iona Technologies



Received on Monday, 2 October 2006 14:31:53 UTC