[MSHADE-325] Should transitive test dependencies really be eliminated?

**[Fred Eckertson](https://issues.apache.org/jira/secure/ViewProfile.jspa?name=fe2417)** opened **[MSHADE-325](https://issues.apache.org/jira/browse/MSHADE-325?redirect=false)** and commented

The site generation for my project started failing because of a compile error after I upgraded to maven-compiler-plugin 3.8.1. The newer version identifies a change requiring a recompile of the test code. Older versions do not. The second test compile fails because the test code relies on a compile dependency of a test dependency and shade excludes them since they are not direct dependencies.

Is this really correct behavior? 

---
No further details from [MSHADE-325](https://issues.apache.org/jira/browse/MSHADE-325?redirect=false)

Start by reproducing the site-generation failure with maven-compiler-plugin 3.8.1 and inspect how the shade plugin handles test dependencies. Focus on the reported case where test code relies on a compile dependency of a test dependency that is excluded as non-direct. Done means the expected test compilation behavior and transitive-dependency policy are established.

Written by the indexing model from the issue text.