Hi Anne:
I am fairly certain that JBI had nothing whatsoever to do with the "one
interface per service" decision in WSDL. Really. Though I definitely do
support that decision, I agree we shouldn't get into it here.
Regarding terminology, though - got any other suggestions?
--Glen
----- Original Message -----
From: "Anne Thomas Manes" <[EMAIL PROTECTED]>
To: <[email protected]>
Cc: "Srinath Perera" <[EMAIL PROTECTED]>
Sent: Monday, July 18, 2005 1:19 PM
Subject: Re: [Axis2] CHANGE : Components vs. Services?
My concern is that EJBs, JBs, JSPs, servlets, MDBs, etc. are
"components". "Services" are larger-grained applications that comprise
multiple "components".
Having recently spent the time to read the JBI spec, I kinda get the
impression that it's responsible for the bone-headed decision by the
WSDL 2.0 WG to establish what I view as an arbitrary and inappropriate
constraint that a service can implement only one interface. (A service
should implement at least three interfaces: its functional interface,
its metadata interface, and its management interface, and it's quite
reasonable for a service to implement multiple functional interfaces.
Somehow the "inheritance" argument [an interface can inherit multiple
interfaces] doesn't sit well with me. But I shouldn't rant about it
here...)
In any case, IBM and BEA have no plans to support JBI. Oracle is
clearly on the fence about JBI. IMNSHO, JBI as unnecessary overhead.
Therefore I don't think that JBI terminology should influence Axis
terminology.
Anne
On 7/17/05, Glen Daniels <[EMAIL PROTECTED]> wrote:
Hi Anne:
The idea here is that a "foo" is a deployment unit which a) shares
classloaders, b) is packaged as a single thing, and c) implements 1..N
services (where "service" == WSDL 2.0 service).
JBI calls something like this thing a "component" (the larger class of
which
SE's and BC's are subtypes). I don't see why it's particularly
non-intuitive to call it that.
I'm not stuck on the name at all, and would be fine with something else
(got
any suggestions?), but I do think the concept that one .aar might
implement
multiple WSDL services is something we should integrate (pre-1.0).
--Glen
----- Original Message -----
From: "Anne Thomas Manes" <[EMAIL PROTECTED]>
To: <[email protected]>
Cc: "Srinath Perera" <[EMAIL PROTECTED]>
Sent: Sunday, July 17, 2005 9:59 PM
Subject: Re: [Axis2] CHANGE : Components vs. Services?
Who came up with the concept that a "component" is of larger
granularity than a "service"? that terminology is just remarkably
non-intuitive!
Anne
On 7/16/05, Sanjiva Weerawarana <[EMAIL PROTECTED]> wrote:
> I have a longer reply coming but a definite -1 to doing anything this
> drastic before 1.0.
>
> I'll explain further when I write my detailed response .. apologies for
> the two-stage note :(.
>
> Sanjiva.
>
> On Sat, 2005-07-16 at 21:17 +0600, Srinath Perera wrote:
> > Hi Glen;
> >
> > I think the basic idea is to widen the scope of a componenet! and
> > replace service archive with a componenet archive. This allowed number
> > of services to share a same class loader.
> >
> > mm .. seems ok to me (may be need to think bit more before commit my
> > self :) ).
> > Thanks
> > Srinath
> >
> > On 7/16/05, Glen Daniels <[EMAIL PROTECTED]> wrote:
> > >
> > > Axis2'ers:
> > >
> > > I've been thinking recently about a couple of things with respect to
> > > Axis2. First of all, the idea that we might want to support some
> > > concept of "service groups" - a bunch of individual services which
> > > are
> > > related somehow (via state, implemented with the same code, etc).
> > > Second of all, I'm thinking of building a JBI implementation on top
> > > of
> > > Axis2, and JBI's notion of "components" are deployable units which
> > > can
> > > each provide multiple services.
> > >
> > > What about changing our model slightly to enable "components" to
> > > implement more than one Web Service? This would entail, I believe:
> > >
> > > * Change axis/services to axis/components (just for clarity)
> > >
> > > * Add a "ComponentContext" level to the context stack between
> > > ServiceContext and ConfigurationContext
> > >
> > > * Components would be "engage()"d just like services (although
> > > looking
> > > at the code I don't see this for services yet... need to dig around
> > > more)
> > >
> > > * component.xml (replacement for service.xml) would contain 1..N
> > > <service> elements each of which looks like the current service.xml,
> > > so
> > > the minimal one-service file would be
> > > <component><service>...</service></component>. We could allow
> > > optimizing this to just <service> at the top level too!
> > >
> > > Thoughts?
> > >
> > > Thanks,
> > > --Glen
> > >
> >
>
>
