I just committed a fix in the mojo sandbox for the shade plugin:
instead of removing the dependencies, it marks them provided.
That should make sure all versions line up if a transitive
dep brings in a 'removed' artifact.
I'm not sure where the removal of provided scoped deps takes place;
probably as soon as they're encountered, so this might not work.
Again, as I've argued numerous times in the past, if provided scope
was transitive this would not be a problem (and loads of other problems
would go away aswell). The filter on provided scope should be in
get(Runtime|Test)(ClasspathElements|Dependencies), not in the artifact
resolver.
-- Kenney
Kenney Westerhof wrote:
Hi,
I added a sleep to the test and quickly copied /tmp/surefire*tmp
to examine them.
The following 2 entries in the surefire tmp file are the cause:
surefire45886tmp:classPathUrl.30=/home/forge/.m2/repository/org/codehaus/plexus/plexus-component-api/1.0-alpha-20/plexus-component-api-1.0-alpha-20.jar
surefire45886tmp:classPathUrl.24=/home/forge/.m2/repository/org/codehaus/plexus/plexus-container-default/1.0-alpha-30/plexus-container-default-1.0-alpha-30.jar
The dependency on plexus-velocity 1.1.5 brings in the dep on p-c-a
1.0-alpha-20
which isn't filtered out by maven 2.0.7, since that uses p-c-d
1.0-alpha-9-stable-1
which is from the era where plexus-component-api didn't exist.
If maven were to use a newer version of plexus with the 2 separate jars,
this problem would go away.
If maven were to use the latest version of plexus with the shaded jar,
then the proper solution in the shading/merging is NOT to just remove
dependencies that
are merged, but still list them but with scope provided so they are
filtered out.
In any case, maven 2.0.7 will not work when projects have dependencies
on the 2-jar
version of the container, unless the p-c-a is filtered out everywhere,
or all
p-c-d poms with shading have p-c-a listed as provided (so that 2 jars
will/might be used
in compilation since provided is only filtered out transitively, but at
least the p-c-a
and p-c-d versions will match).
So whether the test succeed depends on forkmode, even with 2.3. I really
can't get it
to work with 2.3 or 2.3.1 at all unless I set forkMode to never - to me,
they behave
exactly the same. It may depend on what snapshots you have in your local
repo though.
If someone has a working test with surefire 2.3 and forkmode=once
I'd like to get a hold of those surefire*tmp files to see exactly which
jars
are used.
-- Kenney
Brett Porter wrote:
On 12/08/2007, at 6:56 PM, Jason van Zyl wrote:
Looks like a bad plexus snapshot on my end, since it's fine on the
zone too.
testRender(org.apache.maven.doxia.siterenderer.DefaultSiteRendererTest)
Time elapsed: 0.275 sec <<< ERROR!
java.lang.NoSuchMethodError:
org.codehaus.plexus.component.repository.ComponentDescriptor.setSource(Ljava/lang/String;)V
at
org.codehaus.plexus.component.discovery.DefaultComponentDiscoverer.createComponentDescriptors(DefaultComponentDiscoverer.java:68)
I'll poke around some more.
Vincent already brought this up on the list a week or so ago. It's a
problem with surefire not pulling in the right version of plexus. It
should be using the version of plexus specific by the POM, not the
one Maven is running with. So simply forking the tests may do the
trick. But the isolated classloader no longer appears to do what it
was intended to do.
Odd. I've pinned the whole thing to 2.3, using the default forkMode of
once, since that works. It seems it is a regression in 2.3.1-SNAPSHOT,
so I'll add it to the list.
- Brett
---------------------------------------------------------------------
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]
