Skip to content

Remold v0.2.5

Choose a tag to compare

@I-hope1 I-hope1 released this 07 Nov 16:12
· 506 commits to main since this release

v0.2.5 Release Notes (draft for commander review)

Status: prepared 2026-11-07 (tag pending commander approval).
Previous release: v0.2.4 (ddfb194).

Highlights

  • Forge / NeoForge: newly declared @Mixin members hot-swap. Adding an @Inject handler method
    or a @Unique field to an existing @Mixin used to be rejected at reload
    (Conform error ... please restart client, then a ClassInfo$Method.isUnique() NPE) because
    MixinInfo.reloadMixin reuses the pre-reload ClassInfo. plugin-forge now adds the new
    members to that same ClassInfo before IMixinTransformer.reload and rolls them back if the
    reload is rejected. Every member lookup is resolved against the host ClassLoader's own
    FieldNode/MethodNode types, so a stale agent-side ASM copy cannot be matched.
  • One ClassInfo member-sync engine. core/src/remold/reflect/ClassInfoMemberSync.java is now the
    only implementation; plugin-mixin (Fabric) and plugin-forge (Forge/NeoForge) are thin adapters
    (type erasure + log prefix). The engine builds its rollback handle before mutating the member
    sets, which also fixes a latent Fabric defect (see Fixes).
  • Real-container verification on three sides — Forge 1.21.1, NeoForge 21.1.265 and Fabric
    1.21.1 — all against the same shared engine, each with its own red baseline where the failure is
    attributable (a reload rejection / re-weave failure), not "nothing happened".
  • JPMS diagnostics. If a host ships Mixin as a named module without an opens edge, the sync path
    and the re-weave bridge now print the exact
    --add-opens org.spongepowered.mixin/org.spongepowered.asm.mixin.transformer=ALL-UNNAMED argument
    instead of a bare reflective failure.

Fixes

  • plugin-mixin MixinClassInfoSync: a failure between the field phase and the method phase leaked
    the members it had already added (the rollback handle was only built after both phases succeeded).
    The shared engine rolls back what was added; pinned by
    ClassInfoMemberSyncTest#halfWayFailureRollsBackWhatWasAlreadyAdded.
  • plugin-forge: Mixin templates that gained a new @Inject handler / @Unique field were rejected
    by the conform check, forcing a restart. Now hot-swapped (see Highlights).
  • Forge/NeoForge re-weave failures caused by reflective access denial now carry the actionable JVM
    argument instead of only the raw IllegalAccessException / InaccessibleObjectException.

Added

  • tests/fabric-mixin-probe/forge-classinfo/: real Forge 1.21.1 dedicated-server E2E (stage 1 adds
    an @Inject handler, stage 2 adds a @Unique field) with a witness-based Verdict whose
    discriminating power is pinned by synthetic red / green / not-attempted / self-contradictory
    fixtures.
  • tests/neoforge-mixin-probe/: real NeoForge 21.1 E2E whose Mixin config is registered by the
    loader itself through META-INF/neoforge.mods.toml [[mixins]] (no manual
    MixinEnvironment.addConfiguration); its verdict additionally rejects silent degradation
    (Failed to sync Mixin class metadata, "no new members" when members were expected) and manual
    registration.
  • docs/neoforge/01-loader-and-classinfo-boundary-audit.md: read-only audit of the NeoForge
    boundary (16-row code-surface truth table against NeoForge 21.1 and 26.x, artifact-level loader
    facts, B1-B7 breakpoints with falsification tests).

Verification

  • JUnit: 469 tests, 0 skipped, 0 failures (./gradlew check), including the new engine contract
    (5), Fabric adapter (2) and Forge cross-ClassLoader adapter (4) cases.
  • ./gradlew check: all tasks pass, including verifyDocStatusMarkers and verifyTestCount.
  • E2E, same probe sequence for red and green:
    • Forge 1.21.1: red VERDICT RED (stage 1 reload rejected with Conform error ... ClassInfo$Method.isUnique() ... "method" is null, stage 2 re-weave failed) -> green
      VERDICT GREEN (new handler +1 per call, new @Unique reads its initializer 456, existing
      members preserved). Re-run after the SSOT refactor: VERDICT GREEN again.
    • NeoForge 21.1.265 (MC 1.21.1, ModLauncher 11.0.5, sponge-mixin 0.16.4, ModDevGradle
      2.0.148): native [[mixins]] registration, red baseline with the sync-disabled plugin jar ->
      VERDICT RED (Conform error ... isUnique() ... "method" is null); fixed build ->
      VERDICT GREEN with Mixin class-info sync: ... addedMethods=1 (stage 1) and addedFields=1
      (stage 2). No InaccessibleObjectException, no --add-opens needed in this dev runtime.
    • Fabric 1.21.1 (KnotServer, Fabric Loader 0.16.9, sponge-mixin 0.15.4+mixin.0.8.7): the
      existing @Inject self-evolution probe re-run unchanged after the refactor -> RESULT PASS
      (failures=[]), with class-info sync ... addedMethods=1 for both handler additions and the
      added handler removed again in the final stage.
  • Mutation checks: reverse-applying the engine/adapter diff turns the unit cases red; reverting the
    NeoForge Verdict turns its acceptance red (both recorded in the campaign evidence ledger), and the
    NeoForge red baseline is itself the "sync disabled" revert experiment.
  • Evidence files: tests/fabric-mixin-probe/forge-classinfo/evidence/forge-1.21.1-{baseline,after-fix,after-ssot-refactor}.txt,
    tests/neoforge-mixin-probe/evidence/neoforge-21.1-{baseline-red,after-fix}.txt,
    tests/fabric-mixin-probe/evidence/inject-evolution-after-ssot.txt.

Known limits / manual acceptance

  • NeoForge 26.x / FML 10+ is not supported and not verified. That generation removed ModLauncher
    (cpw.mods:* is absent from its library list; entry point net.neoforged.fml.startup.Client,
    FML loader 11.x). The cross-ClassLoader bridge in this release is written against ModLauncher
    semantics; do not extrapolate the 21.1 evidence.
  • NeoForge production installer launch path is not verified. The E2E runs a dev runtime
    (ModDevGradle directory classpath). Module ownership can differ in a production launch, so the
    JPMS --add-opens requirement must be re-tested there; the agent currently prints a precise
    diagnostic instead of attempting runtime unsealing (runtime redefineModule is recorded as the
    preferred follow-up in docs/status.md).
  • NeoForge evidence is one dev-run sequence. Mixin session ids are per-process; the recorded
    hashes are not cross-run comparable.
  • Real client GUI paths and the manual acceptance checklist (docs/acceptance/01-manual-checklist.md)
    remain user-side; container probes cover dedicated servers.
  • Fabric evidence for this release is the existing probe re-run (KnotServer), not a client GUI run.