There's this one too
https://issues.apache.org/jira/browse/FELIX-1184
It's not clear if the problem generating osgi metadata for shaded jars
is in the shade or bundle plugin so I filed it with both.... hopefully
this won't mean that each project will assume the other will fix it.
thanks
david jencks
On May 27, 2009, at 7:44 PM, David Blevins wrote:
On May 19, 2009, at 4:54 PM, David Blevins wrote:
There are two pom.xmls involved; the one that creates the shaded
module and the one that results from the shaded module. When the
"shading" is being done as part of the build the modules that
depend on the "shaded" module will get the first version of the
pom.xml, i.e. pre-shade, which contains all the dependencies we
don't want and are trying to "shade away" (i.e. get rid of). So we
end up compiling and testing with a classpath that is not the same
classpath we ship with which is risky; i.e. if we had a class that
used org.objectweb.asm.Foo it would still compile and run because
the shaded module is still pulling it in as maven is still using
the pre-shaded pom for the duration of the build. But... maven is
smart enough to publish the updated pom.xml, the post-shade
version, where the shaded dependencies are removed. So when it's
not in the build you get the published version which does not drag
in the unwanted dependencies.
FYI, bug reports have been filed for some of these oddities in the
maven-shade-plugin.
http://jira.codehaus.org/browse/MSHADE-50
http://jira.codehaus.org/browse/MSHADE-51
If you can click the "Vote for it" link it will greatly increase the
chances of them getting fixed. A while back I helped them create a
swizzle-jira report that shows the plugin issues with the highest
number of votes, so it will have a direct impact how the issues are
prioritized.
-David
