I recently raised a JDK issue [1] which, when triaged, was closed as not an issue. I think is incorrect and was advised by Dalibor Topic [2] to email this list so here I am.
In a nutshell, it appears that when using the module path and --patch-module, a call to ClassLoader.getResources() will only return resources locations that are within the patch and ignores those that are in the original module. There's a Git repository [3] that should reproduce the problem when you run with ./mvnw test using Java 9 or later. The problem appears when using Maven as Surefire launches the test VM with the main classes on the module path and the test classes patched into the module. As a result, tests are able to load classes from both the module and the patch but getResources() only returns locations within the patch. As far as I can tell, Surefire is configuring the JVM correctly when it launches it so this appears to be a bug in the JDK. Thanks, Andy [1] https://bugs.java.com/bugdatabase/view_bug.do?bug_id=JDK-8206162 [2] https://twitter.com/robilad/status/1013850133589610496 [3] https://github.com/wilkinsona/module-path-scanning <https://twitter.com/robilad/status/1013850133589610496>
