Hi,
Jim and I had a discussion on IRC on the databinding to understand
it should be a concern of "Programming Model" or "Assembly".
Attached is the transcript.
Please express your views.
Thanks,
Raymond
(10:13:10 AM) jmarino [EMAIL PROTECTED] entered the
room.
(10:13:21 AM) jmarino: Hi refng, yt?
(10:13:30 AM) jmarino: sorry rfeng, yt?
(10:13:30 AM) rfeng: yes
(10:13:48 AM) jmarino: hey I'm just thinking about the databinding
stuff. Do you have a sec to chat here?
(10:13:58 AM) rfeng: sure
(10:14:53 AM) jmarino: k. A very high-level question: why is the
databinding specified in the programming model at all as opposed to
being an assembly concept (since it is something about the physical
nature of the wire)?
(10:16:01 AM) rfeng: I would thing assembly level is approparate
(10:16:25 AM) rfeng: think
(10:17:27 AM) rfeng: and it get reflected into the PM?
(10:17:29 AM) jmarino: k so a radical thought: databinding should
never really be expressed as annotations and should never show up
in code except for when code explicitly required a dynamic
interface (e.g. DataObject). In that case the runtime wouldn't
require annotations but could introspect the operation types either
in Java code or WSDL extensions.
(10:18:06 AM) jmarino: it would specified by the binding extension,
or maybe in SCDL
(10:18:17 AM) jmarino: am I way off-base here?
(10:19:16 AM) rfeng: thinking ...
(10:19:53 AM) jmarino: sure np I'm just on a con call ;-)
(10:20:02 AM) rfeng: maybe it's mixed depending on the use cases
(10:20:24 AM) rfeng: three cases:
(10:20:36 AM) rfeng: 1) between mappable interfaces
(10:20:43 AM) rfeng: 2) complex properties
(10:21:03 AM) rfeng: 3) protocol/binding data format
(10:21:24 AM) jmarino: k in those three cases, why is the
databinding requirement need to be known by the code or interface?
(10:22:01 AM) rfeng: I guess it's key to well define mappable
interfaces
(10:22:20 AM) jmarino: k coul you explain that?
(10:23:28 AM) rfeng: for a client in favor of JAXB, can it talk to
a service that requires SDO?
(10:24:56 AM) jmarino: sure but the question is the nature of the
client and service implementations...
(10:25:54 AM) jmarino: do they need to know the databinding
technology? If yes, isn't it because they need to use dynamic APIs
and hence those types should be reflected in the contract operations?
(10:26:35 AM) rfeng: Are you thinking of the case that the same
interface can be used for different databindings?
(10:27:08 AM) jmarino: no that's different. If the same interface
can be used then there should be no knowledge of the databinding by
the implementation
(10:27:18 AM) rfeng: ok
(10:27:45 AM) jmarino: I'm trying to say that most implementations
should never know about the databinding used just as they don't
need to know the physical transport
(10:28:12 AM) rfeng: ok, going back to the JAXB --> SDO case
(10:28:13 AM) jmarino: k hold on for a sec...
(10:28:56 AM) rfeng: let's say: CreditReport checkCredit
(Csutomer ...). sure
(10:29:06 AM) rfeng: Customer
(10:37:47 AM) jmarino: hi I'm back
(10:37:56 AM) jmarino: sorry
(10:39:52 AM) jmarino: rfeng yt still?
(10:39:55 AM) rfeng: yes
(10:40:03 AM) jmarino: sorry I had to talk for a sec on the call
(10:40:07 AM) rfeng: np
(10:40:15 AM) jmarino: can you continue?
(10:40:26 AM) rfeng: so for the operation CreditReport checkCredit
(Customer ...)
(10:41:26 AM) rfeng: the client side passes in JAXB objects and the
service side expects SDOs (CreditReport' and Customer' are
generated SDO classes)
(10:41:48 AM) rfeng: how do we capture this at assembly level?
(10:41:54 AM) jmarino: why does the service side expect SDOs given
that contract (strongly typed)?
(10:43:16 AM) rfeng: maybe because my business objects are
generated SDO classes from the XSD and I don't like JAXB at all :-)
(10:43:36 AM) jmarino: then you should take DataObject as a parameter
(10:43:41 AM) rfeng: I'm not aware of any JAXB clients when the
service is developed
(10:44:00 AM) rfeng: I don't use DataObject because I want strong-
typing
(10:44:13 AM) jmarino: k then you shouldn't care about the databinding
(10:44:26 AM) jmarino: at the programming level
(10:44:47 AM) jmarino: you may have a policy that says "I want to
use SDO, etc. in these palces..."
(10:44:57 AM) rfeng: you're right from the service side, I treat
the annotation is just a convient way to help the runtime for this
case
(10:44:58 AM) jmarino: but the business logic should not care
(10:45:15 AM) jmarino: k that's my point. It shouldn't be in the code.
(10:45:30 AM) jmarino: same as the physical transport
(10:45:38 AM) rfeng: do we have better way to express that?
(10:45:44 AM) jmarino: yes policy
(10:45:58 AM) jmarino: or through introspection of service contracts
(10:46:09 AM) jmarino: (e.g. if it takes DataObject)
(10:46:15 AM) jmarino: also through the binding extension...
(10:46:26 AM) jmarino: e.g. "I only know these formats..."
(10:46:48 AM) rfeng: it's not easy to introspect for cases that
Customer doesn't extend from DataObject
(10:47:15 AM) jmarino: right and in that case the code should not
specify anything since it doesn't need to care what databinding is
in use
(10:47:29 AM) jmarino: it's the same principle as physical bindings
(10:47:59 AM) rfeng: can you give me a pesuo policy for this purpose?
(10:48:45 AM) jmarino: k, for binding type X (e.g. JMS) use SDO to
encode the payload
(10:48:57 AM) jmarino: policy intents should be key
(10:49:14 AM) jmarino: another example...
(10:49:41 AM) jmarino: the "foo" binding only knows how to
serialize using XmlBeans. That is specified by the binding extension
(10:50:00 AM) rfeng: yes
(10:50:26 AM) jmarino: so I'm not sure we need annotations at all
(10:51:21 AM) rfeng: but java annotations could be a convenient
way, because they know what databinding they're using even they
don't care
(10:52:18 AM) jmarino: I think the issue is more fundamental.
Annotations burn things into the programming model that should not
be there. In SCA we do not allow people to burn in the physical
binding...Also, I think not having them simplifies things
(10:52:24 AM) rfeng: does the assembler have the knowledge?
(10:53:14 AM) rfeng: I agree with you on the proper use of annotations
(10:53:17 AM) jmarino: they may or may not. If they don't the
runtime will have the knowledge
(10:54:11 AM) rfeng: How does the runtime acquire the knowledge?
(10:54:35 AM) jmarino: it may know from the binding extension (or
component extension maybe)
(10:54:42 AM) rfeng: yes
(10:54:53 AM) rfeng: axis2 binding may claim AXIOM only
(10:55:20 AM) jmarino: yes. Or it could be constructed in such a
way to allow it to be configured to take almost anything
(10:55:30 AM) jmarino: then the binding could be configured
(10:56:35 AM) rfeng: how about the case that no bindings are
involved, component talks to component over remotable interfaces?
(10:57:08 AM) jmarino: then policy could be used or the runtime
could choose through some algorithm
(10:57:24 AM) rfeng: ok
(10:57:46 AM) jmarino: e.g. "the most performant thing is binary
foo so I will use that"
(10:58:04 AM) rfeng: so the required info can be captured from the
policies or introspections
(10:58:15 AM) jmarino: or from the extensions
(10:58:20 AM) rfeng: yes
(10:58:35 AM) rfeng: and part of them are defined using SCDLs?
(10:58:36 AM) jmarino: this way we keep the programming model
"above" these concerns
(10:58:47 AM) jmarino: yes through policy
(10:59:13 AM) jmarino: I think this is different than just moving
things from annotations to xml though
(10:59:39 AM) jmarino: most of the time nothing will need to be
specified in SCDL or any application artifacts
(10:59:42 AM) rfeng: could the policy def become fairly complicated
as we discussed the parameter-level databindings?
(10:59:45 AM) rfeng: I agree
(11:00:07 AM) jmarino: agree on complexity. I think we will need to
work through that use case
(11:01:42 AM) jmarino: how do you want to proceed?
(11:02:03 AM) rfeng: for WSDL cases, it's usually simple
(11:02:15 AM) rfeng: it's a bit messy for java interfaces
(11:02:40 AM) rfeng: should we keep as is for M2 and fix it in M3?
(11:05:42 AM) jmarino: I don't think we should expose something
fundamental like that if possible
(11:08:25 AM) rfeng: sorry, taking care of my kid for a few mintutes
(11:08:56 AM) jmarino: sure let's reconvene a bit later. I've got
to go too
(11:09:02 AM) rfeng: sure
