Fix JDK 25 IllegalAccessError from module split in JavaUnrestrictedClassLoader

[mcebanupgrade]

We've encountered this issue with the latest version of the rewrite lib and plugin. I asked Claude Code to analyze, fix and confirm the fix locally and here's the result.
I fully admit that the proposed changes here are beyond my understanding. The details bellow, generated by Claude

Summary

• On JDK 25, running OpenRewrite via Maven plugin causes IllegalAccessError: class com.sun.tools.javac.parser.JavaTokenizer (in unnamed module) cannot access class com.sun.tools.javac.parser.Tokens$Comment$CommentStyle (in module jdk.compiler) because Maven's ClassRealm establishes loader constraints that force some jdk.compiler classes to fall back to parent delegation, creating a module split
• The existing LinkageError catch from Fix JDK 25 LinkageError in JavaUnrestrictedClassLoader #6736 correctly falls back to the parent classloader, but the returned class ends up in jdk.compiler module while classes already loaded via defineClass remain in the unnamed module — and jdk.compiler doesn't export its internal packages to unnamed modules
• Fix adds a dynamic module export from the named module to the classloader's unnamed module when falling back, using sun.misc.Unsafe → trusted MethodHandles.Lookup → Module.implAddExports (all via reflection for Java 8 source compatibility)

Test plan

• ./gradlew :rewrite-java:test --tests "org.openrewrite.java.JavaUnrestrictedClassLoaderTest" — 3 tests pass (export mechanism, unnamed module no-op, idempotency)
• ./gradlew :rewrite-java-25:compatibilityTest --tests "org.openrewrite.java.tree.AnnotationTest" — no regressions
• Manual: mvn clean -Popenrewrite rewrite:run -Drewrite.activeRecipes=org.openrewrite.java.migrate.UpgradeToJava25 on a real project with JDK 25 — BUILD SUCCESS (previously failed with IllegalAccessError)

🤖 Generated with Claude Code

[timtebeek]

Thanks for the help here @mcebanupgrade ! Sam has wrestled with classloaders quite a bit more than I have, so I've tagged him for review.

[sambsnyd]

I would like to accept this but the author's lack of confidence gives me pause. Have you tested that this change resolves the issue beyond the provided test? Knowing that this has been validated by more than Claude's assumptions would give me more confidence.

[sambsnyd]

Reading this again I think it looks OK. Seems reasonable to me that this would solve the problem.

[mcebanupgrade]

I would like to accept this but the author's lack of confidence gives me pause. Have you tested that this change resolves the issue beyond the provided test? Knowing that this has been validated by more than Claude's assumptions would give me more confidence.

I've tested that this fix indeed fixes my issue - without the fix, the recipe was failing and how it's passing

My lack of confidence is mostly about the nature of the changes since I'm not familiar with this API and I also can't say with confidence that this change won't cause some other unwanted side-effects.