fix: Search for packages in all available Java modules

[slarse]

Fix #3769

The issue describes the symptoms and the root cause of them. The tl;dr is that PackageFactory.get(String) only searches for packages in the unnamed module, causing type resolution in named modules to not work correctly. This PR fixes that by searching for packages in all modules.

As noted in #3769, there's probably a whole lot of code that assumes that there's only the unnamed module, and therefore use CtModel.getRootPackage() (which returns the root package of the unnamed module) as the definitive root package, which as of Java 9 is not appropriate.

[slarse]

This is very odd. If I now run all tests locally with IntelliJ IDEAs test runner, I get a test failure on LogTest.NonParameterizedTest.testMavenLauncherLogs().

Expected :Running in FULLCLASSPATH mode. Source folders and dependencies are inferred from the pom.xml file (doc: http://spoon.gforge.inria.fr/launcher.html).
Actual   :Running in FULLCLASSPATH mode. Classpath is manually set (doc: http://spoon.gforge.inria.fr/launcher.html).

If I run that test individually, it does not fail. If I run all tests with Maven, there are no failures either. Similarly, there's no failure in CI. Something odd is going on with that test, and I can't see how it's related to the changes I just performed.

EDIT: This appears to be a problem with IntelliJ IDEA's test runner, in combination with use of the nested test classes in LogTest. IntelliJ executes the tests twice, seemingly with some lingering state, causing failures. Most likely adding a new test case caused some unfortunate reshuffling of state. It's not an issue related to this PR so I'll ignore the local failure.

[slarse]

Ready for review.

[monperrus]

This is a serious bug! I'm not that surprised given the relatively low usage of modules in the field.

The fix LGTM, thanks @slarse !

[slarse]

This is a serious bug! I'm not that surprised given the relatively low usage of modules in the field.

Indeed! I'm hoping this solves the most severe issues we're facing in Sorald. But I have a nagging feeling this isn't the last module-related bug we'll see. It's just such a significant metamodel shift from the pre-Java 8 times.

[slarse]

@monperrus Think this could be merged today? It's a blocker for handling Java 9+ projects with Sorald.

[monperrus]

Very important fix wrt modules, thanks a lot @slarse