Shading = repackaging a dependency class inside your own jar (instead of depending on the jar of that class).
I've just spend quite some time trying to find such a classpath issue, which manifested it self as "NoSuchMethodError on slf4j LocationAwareLogger.log(...)"
It turns out the weld-servlet jar was shading some of the slf4j classes.
Shading is evil. Here's why:
- It breaks maven/gradle/buildr/ivy dependency conflict resolution. If guvnor uses slf4j 1.6 and weld-servlet; and weld-servlet shades slf4J 1.5; maven will not detect a slf4j version conflict and will therefor have no chance to put only 1 version of slf4J in the classpath. As a result, there's no telling which slf4J version will be used (the first in the classpath, but that is very unstable)
- It distorts ANT's classpath: putting the weld-servlet jar (with slf4J 1.5) in your classpath can effectively turn the explicit slf4j-1.6.0.jar in your classpath into dead code or combine it with binary incompatibly slf4j extensions such as xstream-for-slf4j-1.6.0.jar (fictive example).
- Because the slf4j classes are actually in the weld-servlet jar, they cause bloat: the user might have already downloaded the real slf4j jar.
- Most importantly, because the (Red Hat and other) guys who create RPM's out of our jars say so. See slide 13 of this Fosdem presentation: http://sochotni.fedorapeople.org/fosdem2011-sochotnicky.pdf (video : http://ia600608.us.archive.org/16/items/fosdem_2011_free_java_guide_to_packaging_for_developers/20110205-fosdem11-guide_to_packaging.ogg)
Summary: shading will cause grief for your users. Why subject your users to grief? In this case, don't give them a good and a bad choice, only give them a good choice.