As I read the specs...
The Assembly spec defines the interface
element including the remotable attribute in section 6. The attribute
is identified as an alternative to interface type specific mechanisms for
identifying an interface as remotable.
The Java CAA spec defines the interface.java
element and refines the interpretation of the remotable attribute in section
3.1 The attribute is restricted to be a way to override the lack
of an @Remotable annotation on an interface.
The Java CI spec defines the rules for
introspecting the effective componentType of an implementation in sections
8 and 8.1. These rules do not cover the case where there is an explicit
remotable=true attribute on an interface.java element.
I think what all this means is:
1) While a preliminary effective
componentType for a Java implementation can be determined just by introspection,
the final effective componentType is not known until that implementation
is used as the implementation of a component and any interface.java elements
in that component are evaluated.
2) The Java CI spec needs some additional
text in sections 8 and 8.1 covering this case.
Bryan Aupperle, Ph.D.
STSM, WebSphere Enterprise Platform Software Solution Architect
Research Triangle Park, NC
+1 919-254-7508 (T/L 444-7508)
Internet Address:
From:
David Booz/Poughkeepsie/IBM@IBMUS
To:
,
Date:
02/16/2010 02:48 PM
Subject:
[sca-j] Re: [sca-assembly] Re: [sca-j]
ASSEMBLY-218: Review Interface Compatibility logic to accomodate @Remotable
attribute in the SCDL - proposal
Thanks for cleaning up for the issue references.
I don't quite agree with your use cases. IIRC, the reason @remotable
was
added was to allow for the case where a Java interface, which cannot be
modified, is used as a remotable interface. Since there's no opportunity
to add the @Remotable annotation, an alternative mechanism was added.
However, since the SCA java component implementation spec decided not to
support the use of a componentType side file, there is no place other than
a component definition which can contain the remotable assertion. We
did
discuss the subclassing option was way to add remotability to an otherwise
unmodifiable interface class. It was tossed out as too cumbersome.
The cases you cited below are different because they run into the by-value,
by-reference issues that you outlined. I don't think we should be
in the
business of overriding local interfaces and allowing them to be turned
into
remote interfaces (and vice versa which can be even more dangerous). The
problem is that in the absence of an @Local annotation, SCA doesn't know
if
the interface is truly local or remote.
WRT your C1 and C2 example, it certainly doesn't seem like a smart thing
to
do, but I don't see any reason to stop it.
In 1) you suggest removing @remotable from component interfaces. If
we
did, that would disallow the use case for which we added the support in
the
first place.
...still, it's ugly...
Dave Booz
STSM, BPM and SCA Architecture
Co-Chair OASIS SCA-Policy TC and SCA-J TC
"Distributed objects first, then world hunger"
Poughkeepsie, NY (845)-435-6093 or 8-295-6093
e-mail:
|------------>
| From: |
|------------>
>--------------------------------------------------------------------------------------------------------------------------------------------------|
|Raymond Feng/Burlingame/IBM@IBMUS
|
>--------------------------------------------------------------------------------------------------------------------------------------------------|
|------------>
| To: |
|------------>
>--------------------------------------------------------------------------------------------------------------------------------------------------|
|David Booz/Poughkeepsie/IBM@IBMUS
|
>--------------------------------------------------------------------------------------------------------------------------------------------------|
|------------>
| Cc: |
|------------>
>--------------------------------------------------------------------------------------------------------------------------------------------------|
|,
|
>--------------------------------------------------------------------------------------------------------------------------------------------------|
|------------>
| Date: |
|------------>
>--------------------------------------------------------------------------------------------------------------------------------------------------|
|02/16/2010 01:56 PM
|
>--------------------------------------------------------------------------------------------------------------------------------------------------|
|------------>
| Subject: |
|------------>
>--------------------------------------------------------------------------------------------------------------------------------------------------|
|[sca-assembly] Re: [sca-j] ASSEMBLY-218: Review Interface Compatibility
logic to accomodate @Remotable attribute in the SCDL - proposal
|
>--------------------------------------------------------------------------------------------------------------------------------------------------|
Hi,
The links to issues are incorrect. They should be:
http://www.osoa.org/jira/browse/JAVA-125
http://www.osoa.org/jira/browse/JAVA-153
IMO, the @remotable attribute in SCDL is tricky because it is the first
case that component configuration starts to alter the interface
remotability in the componentType. There are basically two use cases:
* A java component implements a "local" java interface and we
want to
expose such service to a remote binding.
* A java component with a reference of a "local" java interface
and we want
to use it to access a remote service.
Did we consider the case that more than one component uses the same Java
implementation class, for example:
Component C1 exposing S1 as web service.
<component name="C1">
<implementation.java class="test.TestServiceImpl"/>
<service name="S1">
<interface.java
interface="test.TestService"
remotable="true"/> <!-- test.TestService interface doesn't
have @Remotable
-->
<binding.ws
.../>
</service>
</component>
Component C2 exposing S1 as a local service for binding.sca.
<component name="C2">
<implementation.java class="test.TestServiceImpl"/>
<service name="S1">
<interface.java
interface="test.TestService"/>
</service>
</component>
Is this allowed? If so, do we expect test.TestServiceImpl to support both
the local pass-by-reference or remote pass-by-value semantics depending
on
the components that use the impl class?
A few crazy thoughts:
1) Based on the original use case, should we only allow the SCA
componentType to set the interface.java to be remotable instead of the
component, for example:
<componentType ...>
<service name="S1">
<interface.java
interface="test.TestService"
remotable="true"/>
</service>
</componentType>
(Please note that componentType can be just an in-memory model. The XML
is
for illustration purpose.)
In fact, when we (Tuscany) implement OSGi remote services using SCA, we
generate the componentType from the OSGi properties
(service.exported.interfaces) so that the java interface is remotable.
2) For the direct use of "local" java interfaces as remotable
in Java
components, can we just have a subinterface so that:
@Remotable
public interface RemotableJavaInterface extends LocalJavaInterface {
}
Thanks,
Raymond
Raymond Feng
Senior Software Engineer, Apache Tuscany PMC Member & Committer
IBM Bay Area Lab, 1001 E Hillsdale Blvd, Suite 400, Foster City, CA 94404,
USA
E-mail: , Notes: Raymond Feng/Burlingame/IBM, Tel:
650-645-8117, T/L: 367-8117
Personal Web Site: www.enjoyjava.com
Apache Tuscany: http://tuscany.apache.org
Co-author of Tuscany In Action: http://www.manning.com/laws
From: David Booz/Poughkeepsie/IBM@IBMUS
To:
Cc:
Date: 02/16/2010 09:38 AM
Subje [sca-j] ASSEMBLY-218: Review Interface Compatibility logic to
ct: accomodate @Remotable attribute in the SCDL - proposal
Following discussion of Assembly-218 [1] on the Assembly TC telecon today,
I took a closer look at the Assembly, Java CAA and Java POJO specs. I've
copied the Java TC on this email for awareness.
The assembly spec rule for matching interfaces based on remotability is
fundamentally important, so I don't think we can change that. The
Java CAA
spec is the one that defines @remotable for <interface.java/>. The
Java
POJO spec contains the componentType introspection rules and describes
the
use of Java POJOs for component implementations.
IMHO, if there is any spec text in error or at the very least that we
should consider updating, it is the Java POJO spec [2]. Issue Java-125
[3]
and Java-153 [4] introduced the @remotable attribute but did not address
the trip hazard that is described in Assembly-218.
In the Java POJO spec [2], see section 2.2 (toward the end) and section
2.3. At both points the spec makes some very specific statements
about
interface remotability but leaves out any mention that the determination
of
remotability could be further altered by the introduction of @remotable
on
a component definition. Interestingly, section 2.2 was heavily updated
when Java issues 125 and 153 were resolved. Those were the Java issues
that introduced @remotable on <interface.java/> in the first place.
The
Java POJO spec is (in general) maniacally focused on the componentType
of a
Java component (I'm sure that's what was in all of the Java TC minds when
it resolved 125 and 153). This trip hazard [1] comes along with the
combination of a component definition which seems to have the ability to
assert remotability into an underlying componentType and an introspected
componentType which says local.
Let's see if others buy this analysis before I suggest moving the issue
to
the Java TC.
[1] http://www.osoa.org/jira/browse/ASSEMBLY-218
[2]
http://www.oasis-open.org/committees/download.php/36414/sca-javaci-1.1-spec-cd02.doc
[3] http://www.osoa.org/jira/browse/ASSEMBLY-125
[4] http://www.osoa.org/jira/browse/ASSEMBLY-153
Dave Booz
STSM, BPM and SCA Architecture
Co-Chair OASIS SCA-Policy TC and SCA-J TC
"Distributed objects first, then world hunger"
Poughkeepsie, NY (845)-435-6093 or 8-295-6093
e-mail:
---------------------------------------------------------------------
To unsubscribe from this mail list, you must leave the OASIS TC that
generates this mail. Follow this link to all your TCs in OASIS at:
https://www.oasis-open.org/apps/org/workgroup/portal/my_workgroups.php
---------------------------------------------------------------------
To unsubscribe from this mail list, you must leave the OASIS TC that
generates this mail. Follow this link to all your TCs in OASIS at:
https://www.oasis-open.org/apps/org/workgroup/portal/my_workgroups.php
