jade-recipe-11-17-rmic-task-removal

Removes the two Ant <rmic> pregeneration steps from build.xml. The rmic tool they drive was removed in JDK 15, and Ant refuses the task on JDK 15+, which is the only thing stopping the JADE build on J

SKILL.md

GitHub

jade-recipe-11-17-rmic-task-removal — the static stub compiler is gone

rmic was deprecated for removal in JDK 13 and removed in JDK 15. Ant does not fall back; it fails the task outright:

build.xml:154: rmic does not exist under Java 15 and higher, use rmic of an
older JDK and explicitly set the executable attribute

This is the only compile-or-build breakage in the 11-to-17 window. javac itself is clean: a probe of the untouched baseline on jade-ant:17 compiled all 972 sources with zero errors at source/target 11 and again at 17.

Scope of one task

You receive a shard contract. This rule is blast_class: body-local (parallel_safe: true). Edit only that shard's editable_files; read_only_context is read-only.

src/jade/build.xml is absent from the knowledge graph — the graph is built from .java sources — so the shard planner emitted an isolated shard and said so with flagged_file_not_in_graph. That warning is expected here and is not a signal that something is missing.

The two sites

file line construct
src/jade/build.xml 154 <rmic classname="jade.imtp.rmi.ServiceManagerRMIImpl" base="${build}"/>
src/jade/build.xml 155 <rmic classname="jade.imtp.rmi.NodeRMIImpl" base="${build}"/>

Both sit at the end of the jade target, after <javac> and the resource <copy>. Delete the two elements, and the <!-- JADE-FLAG: --> comment the scanner injected after the enclosing tag. Change nothing else: no excludes, no classpath, no target dependency list.

Why deleting them is safe

Static stubs have been obsolete since Java 5. Both named classes extend java.rmi.server.UnicastRemoteObject:

  • src/jade/src/jade/imtp/rmi/ServiceManagerRMIImpl.java:46
  • src/jade/src/jade/imtp/rmi/NodeRMIImpl.java:38

UnicastRemoteObject.exportObject generates a dynamic proxy stub at export time, and the JDK 13 release note that deprecated rmic says so directly: dynamically generated stubs "are functionally equivalent to statically generated stubs". Removing the pregeneration step removes a build dependency, not a runtime capability. Note also that jade.core.ProfileImpl:540 defaults the IMTP to LEAP, so the RMI IMTP is an opt-in transport in the first place.

What the agent may not do

  • Do not set executable= on the <rmic> task to point at an older JDK's rmic. That pins the build to a second JDK installation to keep a step that produces nothing the runtime needs.
  • Do not add an excludes pattern to hide jade/imtp/rmi/** from javac. The sources compile fine; only the pregeneration step is dead. Hiding live source from the compiler to make a build pass is forbidden by AGENTS.md #11.
  • Do not delete the jade.imtp.rmi package. The rule is about the build step, not the transport.

Stop condition

If anything in the workspace or in consumer-playground/ loads a *_Stub class by name, or otherwise depends on a pregenerated stub file rather than on UnicastRemoteObject export, dynamic stubs do not cover it. Report NEEDS_REVIEW naming the site instead of deleting.

Verification

ant clean lib in jade-ant:17 prints BUILD SUCCESSFUL, exits 0, and produces lib/jade.jar. No build file in the workspace still contains <rmic.