jade-recipe-8-11-corba-removal

Deletes JADE's CORBA dependency instead of retrofitting it: the generated FIPA IDL stubs, the IIOP message transport protocol, and their build.xml wiring. Agent-mode recipe: the shard contract names t

SKILL.md

GitHub

jade-recipe-8-11-corba-removal — delete the CORBA dependency

Java SE 11 removed the java.corba module (JEP 320). org.omg.CORBA, org.omg.CosNaming, org.omg.PortableServer and the RMI-IIOP packages are gone from the runtime image, and JEP 320 states there will be no standalone CORBA unless a third party takes over maintenance.

There are two ways to answer that. Retrofit puts a GlassFish ORB back on the class path and changes no source; the existing recipe jade-recipe-8-11-corba-retrofit does exactly that and stays in the registry as the fallback path. This recipe does the opposite: it deletes the dependency, because the sponsor goal is to remove what Java no longer has or no longer needs, not to re-supply it. The two are alternatives; never run both against the same workspace.

Why deletion is safe here (read this before editing)

  • org.omg.* in this codebase is confined to two places: the generated IDL stubs in src/jade/src/FIPA/ and the single transport src/jade/src/jade/mtp/iiop/MessageTransportProtocol.java.
  • The knowledge graph (03.5-knowledge-graph.json, 1015 files) records zero incoming imports / extends / implements / calls / type_refs edges into that set from the other 972 files. It is a leaf.
  • jade/mtp/iiop/MessageTransportProtocol.java is the only file in the whole workspace that imports the generated FIPA package.
  • IIOP is one of two MTP implementations (the other is jade/mtp/http/), and it is not the default: jade/core/ProfileImpl.java selects jade.mtp.http.MessageTransportProtocol when no -mtps is given. MTPs are loaded by Class.forName from a user-supplied class name, so nothing in the core names IIOP.

This is a breaking change for JADE users — narrow, but real. It belongs in release notes, not in a silent drop. Exactly three capabilities disappear: -mtps jade.mtp.iiop.MessageTransportProtocol, interop with a foreign FIPA platform that advertises only IOR: addresses, and compiling user code against FIPA.* classes shipped in jade.jar. The migration for the first is a one-string config change to the HTTP MTP; for the second there is none.

Scope of one task

You receive a shard contract. This rule is blast_class: signature, so it is planned as a single, sequential shard (parallel_safe: false) — it cannot be split, because the 43 files are one connected component. Edit or delete only files listed in editable_files. read_only_context is readable and must never be written.

Check the shard before you start: editable_files must be exactly 43 entries — the 42 files under src/jade/src/FIPA/ plus src/jade/src/jade/mtp/iiop/MessageTransportProtocol.java — and read_only_context must be empty. Empty read-only context is not a defect here; it is the graph-leaf proof rendered as data, and plan_shards.py emits an empty_read_only_context warning for it. If the shard contains fewer than 43 files, stop and report FAILED with reason SHARD_PLAN_INCOMPLETE, naming the missing files — do not delete a partial component and do not edit outside the shard.

That check exists because it has already failed once. The rule's third pattern, ^\s*package\s+FIPA\s*;, is in the manifest for exactly this reason: with only the org.omg patterns, five of the 42 generated files carried no org.omg reference of their own, and plan_shards.py (which derives editable_files from flagged files plus their one-hop dependents) left FIPA/DateTime.java and FIPA/ReceivedObject.java outside the shard entirely and put FIPA/FipaMessage.java in read_only_context. Deleting the 40 in-shard files would then have left FipaMessage.java declaring a FIPA.Envelope[] field whose type no longer existed — a broken build reached by following the shard contract correctly.

Classification: delete whole file vs edit in place

Use this test on every file, in this order. Do not improvise a fourth case.

1. Delete the whole file when the file exists only to serve CORBA:

  • Anything under src/jade/src/FIPA/ — all 42 .java files. Every one of them carries the generator banner From: FIPA.IDL / By: idltojava; they are idlj output for fipa.idl, not hand-written code. This includes the 5 files that carry no org.omg. reference themselves (AgentID.java, DateTime.java, Envelope.java, FipaMessage.java, ReceivedObject.java): they are IDL value structs whose only consumers are the Helper/Holder classes being deleted, and two of them (FipaMessage -> FIPA.Envelope[], ReceivedObject -> FIPA.DateTime) would not even compile once their siblings are gone. Leaving any of them behind produces dead code or a broken build, never a working API. Do not confuse FIPA.Envelope (an IDL struct, deleted) with jade.domain.FIPAAgentManagement.Envelope (core JADE, kept).
  • src/jade/src/jade/mtp/iiop/MessageTransportProtocol.java, and the now-empty jade/mtp/iiop/ directory with it.
  • src/jade/src/fipa.idl — the IDL source the deleted stubs were generated from. Keeping it invites someone to regenerate them.

2. Edit in place when the file survives but names a deleted artifact. In this codebase that is exactly one file, src/jade/build.xml:

  • delete the idlj target (the <exec executable="idlj.exe"> block) and drop idlj from the depends list of the jade target;
  • delete the check_execIdlj target and the EXECidlj.is.set <condition> in init;
  • remove jade.mtp.iiop from the packagenames attribute of the javadoc task (note the surrounding entries are space-separated, not comma-separated — do not break jade.mtp.http);
  • remove <include name="src/FIPA/**"/> from dist_src, and src/*.idl if no other .idl file remains;
  • fix the stale header comment that promises iiop.jar (line 6); the lib target never produced a separate iiop.jar anyway.

build.xml is not a .java file, so it is not in editable_files and not in the flag index. Edit it anyway — it is part of the same shard's work — and list it explicitly in the result envelope's files[] so the change is auditable. Do not touch any other build file.

3. Leave alone. Not every string with "iiop" or "FIPA" in it is CORBA:

  • jade/domain/FIPANames.java:186 — public static final String IIOP = "fipa.mts.mtp.iiop.std". Keep it. It is a protocol name, not an implementation, and the same file already keeps WAP (line 191), a name with no implementation in the tree. Deleting a public constant would break client code that merely reads it; deleting the implementation does not.
  • jade.domain.FIPAAgentManagement.*, jade.lang.acl.*, and every other FIPA-flavoured package in jade/ — these are JADE's own ontology classes and have nothing to do with the IDL stubs.
  • jade/mtp/http/** and jade/mtp/MTP.java — the surviving MTP and its interface.
  • jade/imtp/** — IMTP (intra-platform) is a different mechanism from MTP (inter-platform). CORBA never touched it.
  • Anything under consumer-playground/ — outside the workspace, and no consumer references IIOP, org.omg, -mtps or -nomtp.

References to deleted classes in files that stay

After the deletions, run this and read every hit before you accept the shard:

grep -rn -E "\borg\.omg\.|^\s*package\s+FIPA\s*;|\bFIPA\.[A-Z]|jade\.mtp\.iiop|jade/mtp/iiop|fipa\.idl|idlj" <workspace>

Resolve each surviving hit into exactly one of these outcomes:

what you find what to do
a compile-time reference (import, type, extends, new) to a deleted class from a surviving file Stop. The premise of the shard is broken — the component was not a leaf after all. Report FAILED with the file and line; do not invent a replacement type.
a build-file reference to a deleted path or target edit it out, per case 2 above
a javadoc/comment reference (@see jade.mtp.iiop..., prose about IIOP) rewrite the sentence to state that the IIOP MTP was removed in this version and HTTP is the transport; never leave a dangling @see/@link to a deleted type, it breaks the javadoc target
a string constant naming the protocol, not the class ("fipa.mts.mtp.iiop.std") keep it — case 3 above
a // JADE-FLAG:CORBA_REMOVAL comment in a file you deleted nothing; it went with the file
a // JADE-FLAG:CORBA_REMOVAL comment in a file you kept it must not exist. If it does, the file was misclassified — re-run the classification test on it.

Verification

The per-shard verifier does not apply to this shard, and you must know why before you reach that step. verify_shard.py compiles every path in editable_files with javac and exits 2 (SHARD_FILE_MISSING) the moment one of them is not on disk. A deletion shard removes all of them, so the exit code reports the intended end state, not a defect. Treat that exit as inapplicable only when it names a file this shard deliberately deleted and the deletion is listed in the result envelope; any other SHARD_FILE_MISSING is a real failure and the shard is rolled back.

Substitute evidence — stronger than per-shard javac, and required before accepting:

  1. Full build, in Docker, on JDK 11: ant clean jade in the jade-ant:11 image must print BUILD SUCCESSFUL and exit 0. Before the removal it prints exactly 100 errors, all of them package org.omg.CORBA does not exist (53) or package org.omg.CORBA.portable does not exist (47); after the removal it must print none. Nothing else is accepted as proof of a build. Capture it to 07-build.log.
  2. jade.jar must contain no FIPA/*.class and no jade/mtp/iiop/*.class.
  3. runtime_verify.py must report all four consumers PASS, unchanged, with no edit on the consumer side.
  4. The grep above returns only hits resolved as "keep" by the table.

gate_signatures.py still applies: build the after-graph and check that no signature change leaked outside the shard's editable set.

At accept time, shard_checkpoint.py --accept exits 1 with "<file> no longer exists on disk at accept time" for each deleted file. That warning is expected here and the shard is still marked ACCEPTED; the exit code 1 is not a failure. --rollback restores every deleted file from its checkpoint blob, so the deletion stays reversible until accept.

Invariants

  • No edit or deletion outside editable_files, with the single documented exception of src/jade/build.xml, which must be named in the envelope.
  • read_only_context is never written.
  • No file is excluded from compilation while being left in the tree. Removing a source file is legal; adding an excludes pattern to build.xml to hide a file that still exists is a pipeline integrity violation (AGENTS.md #11) and is forbidden even if it makes the build pass.
  • No new dependency is added. If the fix needs a jar, it is the wrong fix — that is the retrofit recipe, not this one.
  • No public API of the JADE core changes. The only removed public surface is the FIPA.* package and jade.mtp.iiop.MessageTransportProtocol.

Status to report

status when
FIXED every file in the delete list is gone, build.xml is cleaned, ant clean jade on JDK 11 gives BUILD SUCCESSFUL and exit 0, and 4/4 consumers PASS
NEEDS_REVIEW the deletions are done and the build is green, but a surviving prose/javadoc reference could not be rewritten with certainty
SKIPPED the shard is already CORBA-free (nothing matched)
FAILED a surviving file has a compile-time reference to a deleted class, the build does not reach BUILD SUCCESSFUL, or any consumer stops passing

Never report FIXED with an empty files[]. Never report a status you did not observe: BUILD SUCCESSFUL and exit 0 from Docker are the only accepted proof of a build, and consumer PASS is the only accepted proof of behavior.