Because we don't test what we provide (as release or SNAPSHOT), which is
not good IMHO.
Regards
JB
On 10/03/2012 12:14 PM, Christian Schneider wrote:
Not sure I get you. Why should we need to switch to Equinox in the distro?
We can simply do it in the tests till the fixed version of Felix is
available and then switch back. If we switch back before the release of
Karaf 3 no one will notice and we have working tests in the mean time.
Christian
Am 03.10.2012 12:06, schrieb Jean-Baptiste Onofré:
If we switch to Equinox for the test, it means that we have to switch
to Equinox in our standard distribution: that's not acceptable.
So no, I prefer to keep like this and use -DskipTests waiting for a
new Felix framework version. Anyway, maybe we can investigate on the
Felix issue to find some workaround.
Regards
JB
On 10/03/2012 11:08 AM, Christian Schneider wrote:
The ConcurrentModificationException seems to randomly happen on almost
all tests. Not sure if there is a pattern here. In any case I think we
can not really flag some
tests to solve it. So why not switch to equinox till the issue is
solved. We will switch back before we do a release and make sure all
tests work on felix too.
I guess we will not release Karaf 3 until a fixed Felix version is
available. I am not sure if the Exceptions also happen in production but
if they do it will make Karaf on Felix quite unreliable as it is now.
Btw. When I tried to switch to Equinox using
editConfigurationFilePut(CustomProperties.KARAF_FRAMEWORK, "equinox"),
I get java.lang.ClassNotFoundException:
org.apache.felix.service.command.CommandProcessor. So maybe we should
test more on Equinox :-)
Christian
On 10/03/2012 10:53 AM, Jean-Baptiste Onofré wrote:
Hi,
For the framework, I disagree, as Felix is our "core" framework, it
doesn't make sense to switch to Equinox for the itests: we don't test
what we provide by default.
If it's not a random issue, we can flag the itest with @Ignore waiting
for the Felix fix. Else, I propose to live like this and update to a
new framework version including the fix.
For the port number, we already implemented a port "manager" for
itests. I will implement the same in itests.
Regards
JB
On 10/03/2012 08:19 AM, Christian Schneider wrote:
Hi all,
the tests on trunk are almost fully working now. Two causes of
failures
remain:
- Some tests fail with ConcurrentModificationException. This is a
known
bug in felix. The bug is not yet solved:
https://issues.apache.org/jira/browse/FELIX-3242
I propose we switch to Equinox for the tests until the problem is
solved.
- Some tests fail because of already used ports like 8181 or 4444.
We should create a mechanism to find unused ports and use them
dynamically. CXF and Camel already use such a feature. As a fast
fix we
can simply use different ports for now but this will
not help for long.
What do you think ?
Christian
--
Jean-Baptiste Onofré
[email protected]
http://blog.nanthrax.net
Talend - http://www.talend.com
