I recently worked on some projects that also need OSGi settings and
found an interesting thing.
It seems the easiest way to get the exports, imports and other OSGi
settings right is not to use central defaults and instead do all
settings per project while relying on defaults as much as possible.
I use these settings in the parent:
https://github.com/apache/aries-rsa/blob/master/parent/pom.xml#L199-L223
So basically what I do is to only activate the bundle plugin in the
parent and set it to not auto export packages. For bnd-maven-plugin this
should not be necessary as it is the default.
I then delegate the OSGi settings to a bnd.bnd file. This is largely
equal to the pom config but a little less verbose.
An example can be found at:
https://github.com/apache/aries-rsa/tree/master/rsa
https://github.com/apache/aries-rsa/blob/master/rsa/bnd.bnd
As you see the pom of each project contains no OSGi informations at all
and the bnd.bnd files are very small.
So I would recommend this approach for camel too.
Christian
On 24.03.2016 16:18, Claus Ibsen wrote:
On Thu, Mar 24, 2016 at 4:10 PM, Quinn Stevenson
<qu...@pronoia-solutions.com> wrote:
I’d be happy to take a shot at the conversion. Is there an appropriate JIRA
created already? Or should I continue what you started on the osgi-trouble
branch?
I suggest to start a new branch. The osgi-trouble branch includes
another upgrade as well that surfaced a bug in the old felix
bundle-plugin. If we get the osgi fixed then we can cherry-pick the
caffiene upgrade that are in this branch also.
The new branch could try to aim at
- try to use the bnd plugin
- if that fails try to upgrade to newer felix plugin
- upgrade to newer osgi (is that version 5.0 ?)
- there is some default osgi imports in the parent pom.xml that likely
need upgrades (i think its in the parent pom)
And a related project is to upgrade the osgi tests in the tests
directory to use karaf 4.x by default.
On Mar 24, 2016, at 8:37 AM, Claus Ibsen <claus.ib...@gmail.com> wrote:
Hi
Thanks for sharing the details about the bnd maven plugin. Sounds
promising if its more active maintained and is better.
Anyone is surely welcome to give it a go on the Camel master branch.
The build system is a bit complicated as there is some default stuff
in parent pom.xml and some ant magic to "massage" maven vs osgi
versions when using SNAPSHOTs and whatnot. Its all part of some old
stuff we needed many years ago when OSGi was new and more buggy.
I am not so sure we need all that anymore, it would be lovely to make
the build system simpler and easier.
Sadly I have not seen any tools that can compare a set of JARs against
other JARs to see if their MANIFEST.MF is "the same". Its a bit scary
if the new plugin generates "wrong" imports/exports and the only way
to be sure it works is to run it all in a real osgi container and try
all the components for real. Not only just see if the component can be
installed.
But then this is what the community is for - to help test - especially
for the people who are using OSGi.
People who are not, you are missing out all the fun ;) ..... or maybe not.
A fallback plan is to keep using the old 2.3.7 plugin and then maybe
"hand craft" the camel-core pom.xml instead of generating it to
workaround its issue with Java 1.8 and the caffeine cache. But then we
are stuck on this old dead horse still.
On Thu, Mar 24, 2016 at 2:47 PM, Quinn Stevenson
<qu...@pronoia-solutions.com> wrote:
Antonin/Claus -
I’ve used the bnd-maven-plugin, and it dramatically reduced the amount of
configuration I had to do for my bundles. I hit a bug in maven-bundle-plugin
(https://issues.apache.org/jira/browse/FELIX-5179) and moving to the
bnd-maven-plugin allowed me to what I needed to do. I even provided a patch
for the maven-bundle-plugin, but it has yet to be applied.
I haven’t explored the intricacies of the Camel build as far as bundle
manifests are concerned, but I think it would be worthwhile to try the
bnd-maven-plugin.
On Mar 24, 2016, at 2:28 AM, Antonin Stefanutti <anto...@stefanutti.fr> wrote:
Hi Claus,
Just in case for info, there is apparently a new BND Maven plugin [1] that is
supposed to alleviate some of the issues encountered with maven-bundle-plugin.
I haven’t tried it (nor am knowledgeable in the area) but that may be good to
know at some point for that piece of work.
[1]: http://njbartlett.name/2015/03/27/announcing-bnd-maven-plugin.html
Antonin
On 24 Mar 2016, at 07:44, Claus Ibsen <claus.ib...@gmail.com> wrote:
Hi
m)
Upgrade OSGi
We are using osgi 4.3.1 version which whatever OSGi version that is.
But there is a OSGi 5.0 that newer Karaf containers uses.
But the big pain is upgrading maven-bundle-plugin. We are currently
using an old 2.3.7. But the newer versions have their new sets of
problems / fixes.
i have struggled with newer versions generating missing details in the
manifest.mf files. For example camel-core did not export all its
packages etc. A bit scary. But we do have a fair bit of maven
properties and other osgi "magic" to make the build process build OSGi
modules across all the 250 or so artifacts.
I pushed to a branch called osgi-trouble where you can see some of this problems
https://github.com/apache/camel/commits/osgi-trouble
Using the latest 3.0.1 bundle plugin fails to build camel-core. It
complains something about the osgi activator.
On Wed, Mar 23, 2016 at 11:07 AM, Claus Ibsen <claus.ib...@gmail.com> wrote:
Hi
So Camel 2.17 was the last release supporting Java 1.7.
The next Camel 2.18 is requiring Java 1.8.
Here is some thoughts of mine about this release up for discussion.
a)
I see the overall goal of Camel 2.18 as a stepping stone towards Java
1.8 and Camel 3.0.
By that I mean the release should be a way of moving our existing
users from Java 1.7 and the current Camel APIs and the likes gradually
towards Java 1.8 and eventually Camel 3.0.
In other words we should not get carried away to change/break APIs and
whatnot just because Java 1.8 lambdas and functions.
There are too many current users that rely on the current Camel API
and we cannot go around change processor / expression / predicate /
aggregation strategy and other interfaces to be java 8 functional if
that means current code cannot compile. And certainly not adding
Optional<X> as return types all over.
The following releases (Camel 2.19 or 3.0) can pick up that torch and
be more Java 1.8 aggressive. For example Camel 3.0 can expect API
changes that are Java 8 lambda / functional based. And as well changes
in the DSL to go with that.
There are some minor code changes needed to make the source compile as
source 1.8 to go in this Camel 2.18 let alone.
b)
Drop components that do not support and run on Java 1.8
And potentially remove some deprecated components
c)
Drop karaf 2.x.
And move to karaf 4.x for all our testing.
d)
Drop Jetty 8.x.
This also requires to upgrade at least two components that currently
rely on Jetty 8 to use Jetty 9.
e)
Upgrade to latest Jetty 9.
Jetty 9.3 (or is it 9.4) requires Java 1.8
f)
Drop support for older versions of Spring. We have a number of
camel-test-spring3 etc modules that can be dropped. And maybe even
spring 4.0. as its also EOL.
g)
Potentially move spring-dm out of camel-spring into a camel-spring-dm
module. So camel-spring can use latest version of Spring safely. This
also makes it easier to deprecated spring-dm and remove it eventually.
The Karaf team is working on a sping -> blueprint layer so you can use
spring xml files but Karaf will "convert" that under the hood to
blueprint and run it as blueprint. When that is ready we no longer
need spring-dm.
h)
Continue adding components docs in the source, eg src/main/doc files.
So we eventually have as many/all of them. This is an ongoing effort,
as we need to do this for the EIPs and the other parts of the docs.
However I see this as a great step for a new documentation and
website, that IMHO is a big goal for Camel 3.0. To make the project
website fresh and modern. And make the documentation easier for end
users to use and view.
i)
Add camel-hysterix component and integrate camel's circuit breaker
into turbine/hysterix so you can see metrics from camel in the
dashboard. Eg to integrate with the popular Netflix OSS stack.
j)
Split camel-cxf into modules so we can separate WS and RS and also
spring vs blueprint. Today its big ball of dependencies that is a bit
hard to slice and dice. Specially for MSA style with REST and you dont
want to add in a bunch of extra not needed JARs.
k)
Continue as usual by adding new components, data formats, fix bugs, and so on.
l)
Timeline. This release do not need to have 6-8 months timeframe. We
could try to get this "stepping stone" release done sooner, so it can
be released during/shortly after summer.
There is plenty of "first work" that we must do with the java 8
upgrade and dropping older techs etc, that we have our hands full for
a while.
Doing a release with these changes allows our end users to migrate
along in a easy way, than a big bang - breaking apis - release would
do. And the latter would be more appropriate to be released as Camel
3.0.
Then towards the end of this year, we can see where we are and plan
for a Camel 3.0 with a new website and documentation that such a
release deserve. For example if we release Camel 3.0 in start of 2017
then its also Camel's 10 year birthday year.
And doing such a release with a rewamped website with fresh looking
documentation and content, is what helps the project a lot.
The current website looks the same as it did when it was created:
https://web.archive.org/web/20070701184530/http://activemq.apache.org/camel/
PS: We surely also need a better "what is Camel" story on the front
page. Its still that very first one with all the tech jumble that was
initially created.
PPS: I would also love to see a new Camel logo. The current one is a
bit dull and boring.
--
Claus Ibsen
-----------------
http://davsclaus.com @davsclaus
Camel in Action 2: https://www.manning.com/ibsen2
--
Claus Ibsen
-----------------
http://davsclaus.com @davsclaus
Camel in Action 2: https://www.manning.com/ibsen2
--
Claus Ibsen
-----------------
http://davsclaus.com @davsclaus
Camel in Action 2: https://www.manning.com/ibsen2
--
Christian Schneider
http://www.liquid-reality.de
Open Source Architect
http://www.talend.com
