[ebxml-iic-framework] RE: Re-ordering precedence rules

[Jacques Durand <JDurand@fsw.fujitsu.com>]

Mike: inline

-----Original Message-----
From: Michael Kass [mailto:michael.kass@nist.gov]
Sent: Friday, January 24, 2003 3:57 PM
To: Jacques Durand
Cc: ebxml-iic-framework@lists.oasis-open.org
Subject: RE: Re-ordering precedence rules

Jacques,

  More comments below

At 02:16 PM 1/24/2003 -0800, you wrote:

inline <Jacques>

-----Original Message-----
From: Michael Kass [mailto:michael.kass@nist.gov]
Sent: Friday, January 24, 2003 12:12 PM
To: jacques Durand
Cc: ebxml-iic-framework@lists.oasis-open.org
Subject: Re-ordering precedence rules


Jacques,

    I had the order reversed on rule precedence for the Config parameters.
Here is the correct selection rules for the Test Driver.  The logic is:

For "non-driver mode" ( stand-alone Test Driver)

If someone explicitly declares any of these values in the actual XML
PutMessage declaration, then that value should be used by the Test Driver
to construct
the message.
If the value is not in the PutMessage declaration, then the Configuration
Parameter, if present is used to assign the value.
For parameters that are "auto-generated" by the Test Driver ( e.g.
Timestamp ), if there is no XML declaration, and no Config parameter value
set, then the Test Driver  "auto-generated" value is used.
For parameters that are also present in the CPPA ( SenderParty and
ReceiverParty) , an explicit XML PutMessage declaration takes
precedence.  If an XML message declaration is not present, then if a Config
parameter is present, it is used.  If neither a message declaration nor a
Config parameter is set, the the CPPA value is used.

<Jacques> that seems to cover it all. Bottom line is that XML PutMessage script
should have precedence. For some of these paraemters however,
the ebXML implementation will even have higher precedence, like MesageId.
as it has the "last word" before sending the message.

[MIKE] - Understood. "service mode" is a different animal.

Only thing I think we don't need to refer to here:
the CPA doc (not CPPA, which is the name of the spec and of the TC...)
probably should not be relied upon (not supposed to be interpreted by
Test Driver - so far in this Framework 1.0).

[MIKE] - CPA ( meaning CPAId ) is provided only to pass the ID to the Test Driver to build the message.
(in connection mode) And in "service mode", won't the "interfaced MSH" have to know the  semantic meaning of  the CPAId
and configure itself appropriately? So the Test Driver has to tell the MSH the id of the cpa, that it will
"implicitly" understand and load.  Since the implementation will likely send out the majority of "first messages", it would
need to know the CPAId to use, and configure itself appropriately, I would think.  Comments?
[Jacques Durand] Right. CPAId is used. the CPA  content itself does not need to be known outside the

ebXML implementation. (But do we agree this CPAId can be passed and specified separately from CPA doc?)

[MIKE] - In connection mode, I know we are NOT expecting a Test Driver to load a CPA, or CPA-like document. However, much of the info in a CPA
( like endPoint, PartyId, SyncReplyMode, Retries, RetryInterval, AckRequested ) would really help in writing tests.
Particularly for timing ( calculating timeouts ).   I do not believe that we could formally "parameterize" all that is needed
to configure a Test Driver ( i.e expand our list of parameters that we've agreed upon ) 
We can, however "informally" do so through the "wildcard" <ConfigurationItem> element,
with its name/value pair.  So if we "change a CPA" during for a new <ConfigurationGroup/>, we would also have to change the
"timeout" <ConfigurationItem/> parameter, if the CPA retries/retry-interval changes. ( i.e. the Test Driver, while basically "dumb"
may have to store data that it can use for XPath evaluation.. such as determining if a message arrived too late.
 Does this seem reasonable?
[Jacques Durand] You are talking about two problems it seems:

(a)- how to conveniently specify header data in "connection" mode, where the Test Driver has to do it all by itself.

(b)-flexibility in changing CPA data.

For (a), before considering something like adding a "CPA interpreter" to Test Driver, I'd consider low-key solutions

of the kind: recommending explicitly the use of parameterized message header docs, like we used for illustartion.

A test suite would use as many as needed, given that the current parameterization ($parameter, substituted by values in COnfigurationGroup) is not enough to render changes in XML header needed to reflect various CPAs, e.g. presence of

new elements related to signature, acks, reliability, etc. I think that is quite acceptable for this version, if

you do not want to build all this header in a cumbersome message expression.

So some predefined parameterized header name could be referred to in the putMessage op, and override or modified

at will in the message expression (or declaration). Actually, the Interop test cases currently use such references

but mostly for sake of documentation. We could make that more processeable by Test Driver...

Opinion?

For (b), not sure if I understand well your point, but it's OK to require in this version of TestFramework, that

CPA contetn is fixed (except for overrides that are allow in the ebMS spec, in message header) and will not vary for

a test case execution. In any case, change in parameterized headers + putMessage dcl,

might again facilitate the script writing for generation of messages, that should accompany a change in CPA. 

 

 [MIKE] - It seems logical that we could do this, since we are following the "spirit/semantics" of a CPA, using our own defined parameters to get the
job done.  However, our Test Suite would not be portable, unless someone else implemented a Test Driver that
understands our "timeout" <ConfigurationItem> parameter... i.e.. it hasn't been "formalized" as a parameter

Comments?

In fact, CPA data will be relied
upon through the implementation where it is installed: we know
this will determine some header attributes (security, reliability, possibly
sender and receiver...) Things that we cannot control anyway by Test Driver in "service" mode.

[MIKE] - Assuming we are in "service mode" yes.  However, even in "service mode", the Test Driver
could no doubt benefit from the data in the CPA.  However I understand that at this point
CPA is at best an abstraction to the Test Driver.

In your table below:
-  i'd replace "CPA" more generally by "ebXML implementation",
and that would be the highest precedence, possibly for senderparty, receiverparty,
messageId, timestamp.
- I dont think we need "Test Driver autogenerates": all parameters are actually
set by test suite or by ebXML implementation itself, including Timestamp.

[MIKE] - I disagree that we will be coding Timestamps by hand in "connection mode". I agree that in
"service mode", the implementation does the auto-generation.
But in "connection mode",  I'd much rather have a Test Driver pound out the clock times. 
The Test Suite is just an XML file. The Test Driver will generate Timestamps and MessageIds by default,
unless we "override" them with a Configuration parameter, or explicitly declare them in the Test Suite document.
[Jacques Durand] You may be right. I still believe that a testbed shou7ld allow to manipulate everything

including timestamps, but we should also be able to generate "clock-based" timestamps as well. So I guess we allow for both...

We must say then that a Test Driver MUST be able to generate a default, clock-based timestamp value in case

no value is specified in test cases (either config, or putMessage..). And our test scripting (including config data)

 should include keywords that refer to such predefined function, like "TestDriver.function.timestamp".

Maybe I'm not interpreting what you said correctly?

The table below illustrates this:

Parameter Name  CPPA Equivalent                         Precedence1                     Precedence2             Precedence3

------------------------
-------------------------                               -------------------                     -------------------
-------------------
$SenderParty            <tp:PartyInfo><tp:PartyId/></tp:PartyInfo>      XML PutMessage
Declaration     Config Param            CPPA
$ReceiverParty  <tp:PartyInfo><PartyId/></tp:PartyInfo>         XML PutMessage
Declaration     Config Param            CPPA
$CPA (Id)                                                               XML PutMessage Declaration      Config Param
$ConversationId                                                 XML PutMessage Declaration      Config Param            Test Driver

auto-generates
$Action                                                         XML PutMessage Declaration      Config Param           

$MessageId                                                              XML PutMessage Declaration      Config Param            Test Driver

auto-generates
$Timestamp                                                              XML PutMessage Declaration      Config Param            Test Driver

auto-generates

For "driver mode" ( Test Driver coupled with MSH ) the same rules as above
would hold, however the Test Driver may not be able to set these values
through the MSH interface.

[MIKE] - Understood

<Jacques> for Test Driver in "connection" mode (generating duirectly the message on wire),
I believe precedences are same (except ebXML implementation e.g. MSH is not involved anymore
on driver side)

Cheers,

Jacques
\

Do you agree with this?

Thanks,
Mike