On 4/15/10 9:17, Yair Ogen wrote:
Thank you very much and yes this is exactly what I wanted to achieve, I know
regular Java does not allow this, however I thought that OSGi special
ClassLoader model resolves this issue.

I think this is a real life scenario. for example, assume based on
my previously described Component Graph that the logic behind the scenes is
that CompB was tested with CompA a long time ago. Comp C uses Comp B but
also uses new version of CompA since it is a newer component.

It is a real requirement for me (I think this is common to all?) that the
system engineers do not allow me to have B use the new CompA since this
combination was never tested. They do not want to have full regressions
tests either.

So, the route CompC -->  CompA should use newer version, but the route CompC
-->  CompB -->  CompA should use the older version.

So, I need to tell them this is not possible?

To be clear, if CompC somehow gets objects from CompB that are instances from classes from CompA, then it will not work. This is because CompC is expecting the CompA instances to be 1.0.1, but the instances that CompB returns are 1.0.0. A given class can only see one version of a type at a time.

However, if CompB only uses class from CompA internally and never exposes them to CompC, then it is possible for both CompB and CompC to use different versions of that type.

There is a slight variation where CompB might expose CompA types, but CompC doesn't actually ever use those types from CompB. In this case, you could technically cheat and modify your CompB bundle metadata so it doesn't report the "uses" constraints. But I wouldn't recommend this since it would be really fragile and would likely blow up in other situations where you reuse CompB.

-> richard

Regards,

Yair

On Thu, Apr 15, 2010 at 4:03 PM, Richard S. Hall<[email protected]>wrote:

On 4/15/10 2:17, Yair Ogen wrote:

Right. this was another problem.

now A both verrsion and B work ok.

Problem in C:

org.osgi.framework.BundleException: The bundle could not be resolved.
Reason: Package uses conflict: Import-Package: com.something.b.main;
version="0.0.0"

Is this due to conflict between the imported dependency from to Be and the
one imported directly from C?

Is this situation supposed to be supported? I mean C using one version of
and using B that uses another version of A? How can I accomplish this
goal?


As I mentioned in my original response, it is supported assuming B doesn't
expose A to C...it sounds like com.something.b.main exposes A v1 to C who is
only allowed to see A v1.0.1.

For example, assume B has a class like:

public interface com.something.b.main.Foo {
    com.something.classes.Bar getBar();
}

Assuming com.something.classes.Bar comes from bundle A, then bundle C will
not be allowed to use com.something.b.main, because it would cause it to see
a different version of the package than what it is allowed to see. If this
is specifically what you are trying to achieve, then you can't do this with
Java at all. A given class can only see one version of another class at a
time...otherwise you get class cast exceptions.

->  richard


  Any help appreciated.
Thanks,

Yair

On Wed, Apr 14, 2010 at 10:38 PM, Richard S. Hall<[email protected]
wrote:


On 4/14/10 14:59, Yair Ogen wrote:



Added the suggested Export-Package as well as Private-Package for the
activator.

install fails now on:

org.codehaus.classworlds.NoSuchRealmException: plexus.core
         at
org.codehaus.classworlds.ClassWorld.getRealm(ClassWorld.java:128)
         at
org.codehaus.classworlds.Launcher.mainWithExitCode(Launcher.java:434)
         at org.codehaus.classworlds.Launcher.main(Launcher.java:375)




This sounds specific to the scenario you are trying to implement. Not
sure,
I have no idea what Class Worlds is...

->   richard


  Yair


On Wed, Apr 14, 2010 at 8:42 PM, Richard S. Hall<[email protected]


wrote:





To include packages without exporting them, you should use
Private-Package.
For example, you shouldn't export your activator package.

Is sounds like you want something like this:

CompA v1
Export-Package: com.something.classes; version=1.0.0

CompA v1.0.1
Export-Package: com.something.classes; version=1.0.1

CompB
Import-Package: com.something.classes; version="[1.0.0, 1.0.0]"

Export-Package: com.something.b.main
Private-Package: com.something.activator.b

CompC
Import-Package: com.something.classes; version="[1.0.1, 1.0.1]",
com.something.b.main
Private-Package: com.something.activator.c

->    richard







Regards,

Yair

On Wed, Apr 14, 2010 at 4:40 PM, Richard S. Hall<[email protected]




wrote:








On 4/14/10 9:19, Christopher Brind wrote:







Hi Richard,

Would you agree that a good rule of thumb (subject to specific
circumstance
of course) is that bundles should only import packages OR export
packages,
but not both?








Well, as long as you extend that with, ..."and package API separately
from
the implementation."

Then, yes, that'd be a good rule of thumb for the average case, since
it
would result in the fewest issues overall at the expense of creating
more
bundles to manage.

->     richard


  Cheers,






Chris


On 14 April 2010 14:13, Richard S. Hall<[email protected]>
  wrote:









Technically, what you want to do is possible assuming that CompB
doesn't
expose CompA v1 to CompC. From the sounds of it, you've packaged
your
bundles incorrectly. Don't embed the Person package into your
bundles,
package them as separate bundles and them use Import-Package in the
client
bundles to access them. You might want to start with some
introductory
OSGi
tutorials to understand the basics.

->      richard


On 4/14/10 6:53, Yair Ogen wrote:









Hi,

I am a newbie. I've created dummy projects to test this out.

What I am trying to achieve is as follows:

CompA version 1.0.0 defines class Person that has 2 properties:
name,
lastName
CompA version 1.0.1 defines class Person that has 2 properties:
name,
myLastName
CompB version 1.0.0 calls API on person compiled with CompA
version
1.0.0.
CompC version 1.0.0 calls API on person compiled with CompA
version
1.0.1.
Additionally calls API on CompB.

So, CompA, both versions, are not dependent on anything.
CompB is dependant on CompA version *1.0.0*.
CompC is dependant on CompB version 1.0.0 and CompA version
*1.0.1*.

When I read about OSGi I thought that I should be able to see that
the
API
used directly between C and A will activate Person from A version
1.0.1,
but
the API called from C to B to A will activate the person from A
version
1.0.0.

This is not the case. Pure B API activate the "right" person.

Person used from C (regardless of the origin - API in C or API in
B)
activates the same Person - 1.0.1.

When investigating I saw that the person class was embedded in B
and
in
C,
so of course they can only work with one class at a time.

My question is  - isn't there a non embedded approach, where both
A
versions
jars are in the classpath and the container will pick up the
correct
dependency on the fly?

This must work for me, because this is the heart of what I need
from
OSGi
-
Manage multiple version of the same artifacts in a single jvm...

Regards,

Yair











---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]














---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]











---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]








---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]





---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]



---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to