ForgeMultipart/JAVA_MIGRATION_HANDOFF.md

Java migration — working handoff

Start here. Work in this migration checkout on algent/java, based on master; inspect branch, status and recent commits before editing and preserve existing work. The detached review worktree is not the migration checkout; codex/tile-compatibility-fixes was deleted after its fixes reached algent/java.

Document	Purpose
Plan	Phase gates, API policy and upstream cleanup
Java API index	Consumer entry points, guides, compiling examples and remaining API gaps
ABI inventory	Shipping binary names/descriptors and reflective constraints
Consumer audit	Runtime behavior, data/lifecycle contracts and consumer source
Divergences	Intentional effective compatibility differences
Manual checks	Client/release checks with concrete item examples
Profile	Measured workloads, results and rerun commands
History	Dated completed-port findings and reference evidence; read as needed

Current state and next target

573 plain-JVM tests and 289 Java 8 Forge tests pass, with zero failures/errors/skips. Sources total 225 Java files and 9 Scala files / 747 nonblank Scala lines. The packaged inventory has 444 classes.

Review follow-up: restored packet-scheduler callback mutation behavior with the original Scala hash-map traversal, virtual tile accessor dispatch throughout TMultiPart, and null-safe equality for scheduled-tick deduplication. Six JVM and two Forge regression cases cover the fixes. Callable signatures remain unchanged; the packet traversal callback adds one anonymous class. See the latest history entries for the individual fixes and validation.

Accessor review fixes: TileMultipart, Microblock and BlockMicroMaterial now use their virtual state accessors throughout the restored paths. Eleven new regression cases pass against the original Scala jar and the fixed port; all 519 archived consumers pass with their recorded version. The supplied consumer checkouts confirm extension/API use, but no override triggering these three regressions was found. All 443 class/member APIs, 17 ScalaSignature payloads and 116 generated dumps are retained. The agreed next milestone is the documented consumer-facing Java API, followed by consumer release/adoption and final Scala removal; see the plan's API migration design and Phases 8–10.

Latest API work adds supported multipart button orientation mapping. Et Futurum Requiem 78a5744dfd33 currently reflects ButtonPart.metaSideMap and sideMetaMap, then writes the missing floor and ceiling pairs. Baseline commit cd13ad6 freezes the default maps, those exact four writes and placement metadata on all six faces. ButtonPart.setOrientation(int, ForgeDirection) performs the same update through one validated public API, keeps both maps one-to-one when replacing a pair and rejects invalid input before mutation. Its compiling example uses two direct common-initialization calls without Scala or reflection.

All 573 JVM / 289 Forge tests pass; 573 archived JVM and 287 archived Forge callers also pass. All 444 production classes and 17 ScalaSignature payloads remain. The 3,763 old methods are unchanged and one public static method is added; all 134 generated dumps match by name and hash. The mutable arrays retain their exact public static shape for old Et Futurum releases. Consumer patch/release/pack adoption and physical-client validation remain open. Evidence: run/migration-button-orientation-reference/.

Direct typed calls are the intended end state for supported integrations. Optional dependencies should isolate typed compatibility code behind presence/version checks; reflection snippets in earlier guides are temporary legacy interoperability options, not a reason to leave consumers depending on reflection. Add purposeful missing capabilities where needed, without exposing unrelated internals. Keep existing private fields/bridges until release and adoption.

Naming correction: the planned loadParts(Collection) overload made javac require scala.collection.Iterable even for Java arguments. The distinct loadPartList name allows compilation without Scala on the example's classpath. The old registerParts family also required scala.collection.Seq/scala.Function2 for Java array/factory arguments; registerPartFactory compiles without Scala. Keep this gate for remaining APIs; import/bytecode checks alone are insufficient. Internal storage writes and reconstruction still dispatch through the old setter/loader hooks, not the Java siblings.

The API index orients consumers to entry points, compiling examples and migration status. Material enumeration and tile read/traversal/loading/storage APIs are also complete. No supplied consumer directly calls the two-argument tile occlusion hook; existing canAddPart/canReplacePart calls need no rename. GuideNH and Schematica have Java loading replacements, but consumer source changes, other reflection/extension contracts, releases and pack adoption remain pending. Checkouts remain reference-only. The installed +719 pack scan retains the +678 floor: 27 consumers, 35 inherited types, 255 members, 76 other types and 20 strings.

Next priority: move the completed FMP-side Java replacements into Phase 10 consumer patches and releases. Reuse the documented Java surface and preserve each consumer's lifecycle, state and failure policy. ProjectRed's illuminated microblocks remain the representative external extension case. Consumer mods may remain Scala internally while adopting this API. The next bounded source migration can replace Iguana's ItemSaw.harvestLevel reflection with the supported setter and order that work before ForgeMicroblock post-init; its supplied checkout remains reference-only. Converter registration/lifecycle and stable tile capability access guidance are complete. The illuminated microblock example supplies the representative Java extension, but physical-client construction, connector-dependent halos, lighting and consumer adoption still need their recorded checks.

Progress: all ten Phase 9.1 table rows and Phase 9.2's internal markers are complete, plus Schematica registry lookup, staged Java tile generation, microblock creation, typed material access, the Java illuminated extension example, slot refresh, custom Java tile-trait authoring, button orientation mapping and saw-strength mutation. The identified FMP-side reflection gaps now have typed replacements; consumer patches, releases and pack adoption still precede final Scala removal and client/pack release validation.

Extra Utilities remains an active supported consumer. UtilitiesInExcess is the intended replacement, but the support switch awaits approval and actual target-pack adoption. Retain existing contracts until those gates pass. The overall remaining-work summary separates FMP work from downstream gates.

A separate Phase 4b performance pass is planned once the API and representative extension workloads are stable, alongside consumer migration. Use fresh realistic profiles and repeated paired runs; distinguish FMP implementation gains from migrated-consumer gains, covering hot paths plus startup, transitions, rendering, network and memory costs. Follow the measurement protocol. The historical Phase 4 results do not replace that pass, and no new performance gain has been measured for this update.

Pause mechanical extraction of retained Scala shells unless it enables that API, fixes a demonstrated issue or has a measured benefit. ScalaSignature.ClassSymbolRef.info remains an optional bounded extraction, not the default next task. Keep case-class/product/serialization shapes and simple model accessors supported until their users are retired. The external ProjectRed Scala-trait fixture and ScalaSignature model bridges remain required. Actual client generation, GPU output and full-pack checks remain manual.

Commit 5f0e329b established Java 21 method-body syntax in StackAnalyserLogic through scoped compilation and Java 8 downgrading, while retaining Scala 2.11.5 on Java 8. Prefer modern syntax where it improves readability and the compilation boundary supports it; it need not wait for Scala removal. Defer a specific change if Scala parsing, compile order, ABI or downgrader/runtime support blocks it, recording the blocker and revisit condition. Keep the global modern-syntax setting disabled while it breaks Scala compilation. See the plan's modern Java readability policy and JVM_DOWNGRADER_HANDOFF.md for the original 398-test integration checkpoint and expansion constraints. The initial Java API can operate over retained Scala storage and compatibility shells. Record legacy FMP uses, replacements, consumer releases, target-pack adoption and verification in the consumer audit. Source patches alone do not permit removal. Once the adoption gates pass, replace or retire FMP's remaining internal Scala users before dropping its compiler/runtime dependency. Other mods' own Scala requirements are outside this removal target.

Remaining Scala units:

Files under src/main/scala/codechicken	Why they remain
multipart/asm/ASMMixinCompiler.scala	Retained nested models, construction callbacks and Scala entry-point shell
multipart/asm/ScalaSignature.scala	Named models, primitive/erased bridges and five generic inner-construction branches
multipart/asm/StackAnalyser.scala	Class/companion/models over Java StackAnalyserLogic; state, handler callback, bridges and model queries remain Scala
microblock/MicroblockTraits.scala, FaceMicroblockTraits.scala, CornerMicroblockTraits.scala, EdgeMicroblockTraits.scala	Nine retained state/inheritance/bridge shells over Java implementation, including post lifecycle/super dispatch
microblock/HollowMicroblockTraits.scala	Hollow server/client inheritance, initializer and super shells over Java behavior
microblock/TMicroOcclusion.scala	Occlusion/client inheritance, state/accessor and lifecycle/super shells over Java behavior

Both generators, both registries, core tile/part classes, ordinary microblock helpers/factories, handlers, networking, placement/render helpers and built-in tile traits are Java. The low-risk queue and immediate consumer gate are complete. Per-port evidence lives in the history; do not repeat it here as unfinished work.

Required workflow

The user explicitly requested characterization and regression tests for every migration target. That authorization persists. Six early ASM extractions lacked characterization; their backfill is complete. Do not repeat that gap.

1. Read both consumer audits before changing any compatibility surface.
2. Write meaningful tests on the untouched implementation. Use plain JVM tests for isolated behavior and Forge for initialization-dependent code. Run them, then commit separately as test: characterize X.
3. Save the reference dev jar, source, reports and relevant generated outputs under ignored run/migration-X-reference/ before editing. Freeze compiled consumers where the audit shows a load-bearing bridge; do not recompile those fixtures against the port. No stashing or checkout switching is necessary when the baseline is saved first.
4. Port only the bounded unit. Preserve binary names, descriptors, modifiers, Scala-facing bridges, virtual dispatch, cache/state/serialization behavior and failure ordering. Keep bug fixes and compiler algorithm changes separate.
5. Run formatting/checkstyle/build and Forge, then compare APIs and generated output. Repeat from a clean build; stop the Gradle daemon first on Windows. Do not weaken tests to make the port pass.
6. Update the history and this handoff. Add a divergence only for a new effective difference; shared compiler artifacts already have one ledger entry. Commit separately as refactor: port X to Java on algent/java.
7. Rebuild after the final commit and verify all five @Mod versions in both dev/release jars match the clean version in their filenames. Keep compileScala.scalaCompileOptions.force = true: Zinc otherwise leaves stale Java annotation values when Tags.VERSION changes. Reconsider it only with the planned source-layout cleanup.

.\gradlew.bat spotlessApply checkstyleTest build
.\gradlew.bat runFunctionalTestServer
.\gradlew.bat --stop
.\gradlew.bat clean spotlessApply checkstyleTest build
.\gradlew.bat runFunctionalTestServer

JUnit XML: build/test-results/test/TEST-*.xml and run/server/junit-out/TEST-*.xml. Count tests, failures, errors and skips; do not infer them from Gradle's task summary. The Forge runner validates its own reports, and its dependent CI job must stay required. The ignored local server EULA is already accepted. Archived test mods predating the illuminated fixture lamp must use a separate disposable world; reusing the current test world triggers Forge's missing-mapping prompt. Restore server.properties after such archived runs.

Generated dumps: run/server/asm/multipart/**/*.txt; enumerate recursively and assert a nonzero expected count. Compare names and hashes first. Investigate differences before normalizing only proven debug/private-name changes. The current baseline emits 134 dumps, including the deterministic definition, compiler and feature fixtures.

Binary and fixture checks

Compare member names and descriptors together, with modifiers, generic signatures, private reflective fields and the emitted class inventory; descriptor counts alone miss changes. Inspect javap -p -s or ASM output. Existing ignored reference directories contain comparison scripts; adapt their explicit target exclusions and expected counts. Preserve model serialization and ScalaSignature bridges; routine unreferenced closure removal is covered in the ledger.

Re-scan a pack with java tools/AbiScan.java "<instance>/.minecraft/mods" ForgeMultipart using JDK 17+. Diff against src/test/fixtures/abi/gtnh-daily-678-consumers.txt (27 consumers); the source audit adds the +700 release provenance and UtilitiesInExcess. The clone root is E:\Development\GTNH\Projects\ForgeMultipart_Java_Port\fmp-consumers.

For a new frozen Scala consumer, run Scala 2.11.5 under Java 8, using the reference jar plus the main compile classpath and scala.tools.nsc.Main -target:jvm-1.8. Encode the compiled bytes under src/test/resources/compat/, record source/class SHA-256s in src/test/fixtures/README.md, and follow an existing *BinaryCompatibilityTest. Compilation against the migrated artifact would defeat the binary-compatibility check.

Retained compiler constraints

• Microblock trait inheritance still needs Scala metadata. Abstract Java mixins and Java @SideOnly filtering are supported, but that does not supply Scala multiple-trait linearization. getAndRegisterParentTraits collects Scala-signature traits, while Java registration only incorporates an already registered superclass trait, not implemented Scala interfaces. CommonMicroblockClient combines three trait parents, including the still-Scala TMicroOcclusionClient. Keep its declarations and bridges while extracting behavior; replacing them outright needs a separately characterized compiler/inheritance strategy. Extra Utilities, ForgeRelocationFMP and GuideNH also use these runtime interface types.
• External Scala traits remain supported. ProjectRed registers LightMicroblock through MicroblockGenerator.registerTrait; keep its ScalaSignature ingestion, $class helper and generated dispatch. The Java path now supports its @SideOnly requirement, but retiring this external Scala path still needs a released ProjectRed Java rewrite. Do not flatten generated microblock inheritance to bypass that consumer transition.
• ScalaSignature models cannot be mechanically replaced. Primitive literal value() methods coexist with erased Object value() bridges; Java source cannot declare both. Five generic inner constructors also need their retained Scala construction branches and exact outer-instance bindings.
• ClassInfo bridges remain Scala. Javac expects an extra $ in nested ClassInfo$ model names; its interpretation of IterableLike.view()'s generic Object return emits a missing descriptor instead of IterableView. Preserve construction/access callbacks, the initial Scala .view, lazy collection builders and virtual dispatch. Keep path-dependent symbol types out of Java-callable signatures. Joint helpers resolve ASMMixinCompiler$.MODULE$ internally because Scala cannot resolve a Java method parameter naming its own generated $ companion class.
• StackAnalyser's shell stays coordinated with its companion/models. Scala 2.11.5 rejects a Java class beside its same-name Scala object. A full shell port needs a coordinated binary/model/serialization bridge strategy.
• Characterized algorithm quirks stay separate from this port. Primitive NEWARRAY throws MatchError: 188; wide argument counts can misindex getSuper; Scala String alias parameters decode as Lscala/Predef/String;. Successful external fixtures use java.lang.String; failing alias fixtures retain the original failure.
• Cache and error quirks are intentional preservation. Metadata keys distinguish dotted/slashed strings, while byte-cache publication normalizes and invalidates only that key. Parse failures keep cached bytes; null results are cached, load failures are retried, and dump failures follow publication. Ordinary define duplicates remain InvocationTargetException-wrapped; only direct LinkageErrors reach its case-sensitive duplicate handler, whose null message still throws. Do not unwrap or repair these during mechanical extraction.

Gotchas, all discovered the hard way

Default methods vs superclass. Emit a Java default only when no class in an implementor's superclass chain declares the same member. A superclass method always beats an interface default on the JVM, and the failure is silent. TFacePart, TEdgePart, Saw, JIconHitEffects.getBreakingIcon and IMicroMaterial's three members passed this test. TCuboidPart, TNormalOcclusion, TIconHitEffects, TItemMultiPart.onItemUse and TRandomUpdateTick.onWorldJoin failed it and had to stay abstract with explicit forwarders in implementors.

Audit existing super call sites, not just missing overrides. PostMicroblock already overrode occlusionTest and ended with super.occlusionTest(npart), which used to route through the trait. After the port it reached the superclass directly and silently dropped the box test, while still compiling.

A trait gaining a default changes its concrete implementors. ItemSaw and MissingMicroMaterial stopped declaring members they never overrode, because Scala no longer emits a forwarder. Diff those classes too.

Scala trait extending a class is a bare interface in bytecode. It carries none of the class's members, so Java code must cast. Seen with JIconHitEffects, TRandomUpdateTick, TileMultipartClient. There will be more.

Trait super accessors are ACC_SYNTHETIC. javac cannot see or implement them. Where one is load-bearing, emit it as a non-synthetic default returning the identity for the chain, as codechicken$multipart$TNormalOcclusion$$super$occlusionTest does. For retained Scala traits, keep the accessor and its call in Scala instead. PostMicroblockTraitLogic.occlusionResult returns 0/1 for a completed decision or -1 for the Scala bridge to invoke the original super chain; no callback allocation, eager super evaluation or new public accessor is needed.

Recompiled Scala consumers. These are source-only breaks; binaries are fine. Expect and document them: x.partList(i) becomes x.partList.apply(i); assignment to a Java accessor needs _$eq; a Scala object becoming a Java class loses apply sugar and must be referenced as X$.MODULE$ when passed as a value; Tuple2 members typed Boolean in Scala arrive as Object and need casting; and Scala 2.11 will not adapt a lambda to Function1[_, BoxedUnit], so operate { p => ... } needs an explicit AbstractFunction1. A Java static array getter also loses Scala property/indexing composition: the converted Face, Corner, Edge, and Post bounds facades need an extra empty argument list before the index when recompiling, while existing bytecode remains compatible.

A companion object can be load-bearing without a single bytecode reference. The inventory's 17 MODULE$ list is not the whole test — check the reflective string constants too. MultipartHelper$ is in neither the MODULE$ list nor any consumer's constant pool as a type, but guidenh names it as a string, so it was kept. IconHitEffects$ was in neither and was dropped. Check both lists before deleting a companion.

Do not call a side-only Scala object's static facade from common Java. MultipartProxy.postInit() exists in the raw jar but inherits @SideOnly(CLIENT) from the client override, so Forge strips that forwarder on a dedicated server and a Java call fails with NoSuchMethodError. Call MultipartProxy$.MODULE$.postInit() as the Scala reference did; after side stripping, virtual resolution reaches the server superclass implementation.

Generated version constants require full Zinc recompilation. Tags.VERSION is a compile-time constant embedded in the joint-compiled Java @Mod annotations. Gradle notices the generated classpath change, but Zinc does not invalidate those Java consumers. compileScala.scalaCompileOptions.force = true keeps incremental jars current; do not remove the version assertions or this build guard without replacing both with an equivalent mechanism.

A Scala-to-Java source replacement needs one clean verification. Zinc can retain the deleted Scala source's class files while compiling the new Java file with the same binary name. An incremental test and jar can then exercise and package the old implementation despite a green build. GrassMicroMaterial exposed this: only clean changed the class SourceFile from Scala to Java. Stop the Gradle daemon first if Windows holds build/rfg/recompiled_minecraft open, then run the focused test and ABI comparison from the clean output.

Classes that cannot class-initialize headless make good probes. MultipartSaveLoad reflects into TileEntity's static maps through ObfMapping and always throws under a plain JVM. That turns "did this branch reach the loader?" into an assertion: returning normally proves the guard short-circuited, and assertThrows(LinkageError.class, ...) proves the other branch did reach it. Use LinkageError, not the exact type — the first attempt raises ExceptionInInitializerError and every later one raises NoClassDefFoundError, so test order would otherwise matter.

Where a trait member cannot be a default, use an interface static. TScheduledPacketPart.read is shadowed by TMultiPart.read, so the dispatch that lived in the $class helper became TScheduledPacketPart.readMask(part, packet), with a one-line forwarder documented in the javadoc. Same shape as a $class bridge, but idiomatic and callable from Java.

private[multipart] becomes public, not package-private. It reaches codechicken.multipart.handler too, and Java has no scope spanning a package and its siblings. PacketScheduler.sendScheduled and the generator companions keep those methods public; do not add static forwarders for originally companion-only methods.

A Forge FakePlayer unlocks the server half of player-dependent code. FakePlayerFactory.getMinecraft(worldServer) needs no connection and its world is not remote, which is the branch a server takes. It made ControlKeyModifer.isControlDown testable. The client branch of anything stays manual.

A Scala package object cannot be ported. It compiles to a class named package, and that is a Java keyword, so no Java source can declare or name it. Both were removed by inlining their single member. Any future one has the same three options: leave it in Scala, invent a Java type to hold its members, or inline it.

Scala's uniform access hides field versus no-arg method. renderer.hasOverrideBlockTexture reads the same in Scala either way; it is a method on RenderBlocks and Java needs the parentheses. It fails at compile time rather than silently, but expect it in every remaining renderer conversion.

Preserve inferred Scala return types. Adding : String to FieldMixin.accessName changes its ScalaSignature payload despite preserving the JVM descriptor. Let the Java delegate's return type remain inferred and compare the metadata as well as callable signatures.

Java 8 target. Joint main sources and tests cannot use List.of, var or switch expressions. The scoped StackAnalyserLogic stage supports verified Java 21 method bodies; its packaged output must still target Java 8. These are current compilation constraints, not a preference for old syntax. Move additional source units onto the scoped path only when compatible; otherwise retain the working syntax and record why modernization is deferred.

Two test classes sharing global registry state must guard their registrations, and the registries' error paths call a logger that is null until preInit, so they cannot run headless at all.

Java traits cannot directly touch inherited members with the current transformer. An inherited field read is mistaken for trait-owned state, and an inherited virtual call takes a broken cast-rewrite path. TRedstoneTile keeps coordinate, partList, and virtual partMap access in package-private TRedstoneTileAccess; use that pattern or fix and characterize the generator before converting another trait with the same bytecode shape.

Java traits also cannot directly read fields owned by method arguments. The transformer rewrites every GETFIELD as trait-owned state without checking its owner. TFluidHandlerTile therefore keeps FluidStack.amount reads and writes in package-private TFluidHandlerTileAccess. A static helper outside the transformed class is enough when no trait-to-superclass cast is involved.

Java traits cannot allocate primitive arrays. The stack analyzer has no case for JVM opcode NEWARRAY and throws MatchError: 188 while registering the trait. JInventoryTile therefore converts its collected sided slots to int[] in package-private JInventoryTileAccess. Reference arrays use ANEWARRAY and remain safe inside traits.

Registered Java traits may now extend another registered Java trait. Register the parent first. The transformer adds the parent runtime interface and linearization entry, while parent validation resolves through it to the concrete TileMultipart base. This path is characterized by TRandomDisplayTickTile; do not generalize it to arbitrary class inheritance without another fixture.

Mark runtime-only Java-trait fields transient when Scala did not copy them. autoCompleteJavaTrait copies every ordinary field through generated copyFrom, but now excludes transient fields. TileMultipartClient uses this for render caches and its derived dynamic flag, preserving the Scala trait's absence of a copyFrom override.

Do not invoke a registered mixin type directly from Java source. The untransformed input is a class, so javac emits class opcodes that become invalid when Forge rewrites it to an interface. In-repo client calls use additive no-op hooks on TileMultipart so generated overrides dispatch through the stable superclass. Dedicated-server tests use targeted method handles for trait-only methods; ordinary reflection enumerates the client-only RenderBlocks descriptor and triggers the side transformer.

Type the access shim's parameters Object. TTileChangeTileAccess first typed them as the trait itself and cast to TileMultipart inside. Because the untransformed trait extends TileMultipart, javac elided the cast, and once Forge rewrote the trait to an interface the verifier rejected the getfield. TRedstoneTileAccess only escapes this because IRedstoneTile is unrelated to TileMultipart, forcing a real checkcast. Object always works.

A Java mixin trait may not carry an inner class. registerJavaTrait throws on a non-empty InnerClasses attribute. An anonymous AbstractFunction1 for an operate callback, a lambda, or a string switch all trip it. Put the callback in the access shim as a named class, as TTileChangeTileAccess.NeighborTileChanged does.

Abstract Java mixins declare contracts without helper bodies. Their abstract methods stay on the generated interface so concrete mixin methods can dispatch through them, but do not enter MixinInfo.methods until a later mixin implements them. Their source constructor must still be no-argument; it may directly call a superclass constructor with arguments. Registration removes that call and its argument evaluation because the generated composite has already invoked the real base constructor, then preserves the remaining field initialization.

Release gaps and cleanup

• The full manual checklist is still open. The user confirmed the ProjectRed placement crash is fixed (36e1a58), but that does not establish all rendering, particles, selection/collision, pick-block, activation or lighting behavior. The headless renderer tests cover compiled dispatch, not GPU output.
• Existing binaries retain the runtime tile interfaces. Recompilation against raw Java mixin inputs can emit invalid class/field opcodes after Forge transforms them. Stable capability guidance covers redstone, ordinary base/interface calls and slot refresh; custom authoring guidance covers marker registration, stable capabilities and transformer constraints. The server pass-through fixture does not cover client exclusion.
• Run representative full-pack CPU/allocation and packaged modern-Java validation before performance/release claims. Focused results and reruns are in the profile. The current deobfuscated dedicated-server harness uses Java 8.
• Core NBT, descriptions and one-byte shape updates are frozen; characterize each remaining shape's state immediately before changing it. Keep material/type registration order and handshake formats intact.
• The optional successful MCPC compatibility hook still requires an actual patched World to exercise end to end.
• Keep the plan's pre-merge gates: source relocation where joint compilation permits it, version-guard review, README refresh, durable-document organization and recorded client checks. History is now under docs/migration/; relocating the other durable documents remains open. GTNHLib/UniMixins are added only when a concrete change needs them.