[EverNife]

Problem

Fixes #189. Crucible's ChunkProviderServer patch changed chunksToUnload (field_73248_b) from java.util.Set to CraftBukkit's LongHashSet. Any mod that expects Forge's type breaks. Lumi, for example, crashes at boot with InvalidMixinException: @Shadow field field_73248_b was not located, because a mixin @Shadow matches the field by name and descriptor (Ljava/util/Set;).

LongHashSet can't just implement Set: its toArray() returns long[], which clashes with Set.toArray(). Its keys are also different. Forge keys the set by ChunkCoordIntPair.chunkXZ2Int(x, z), while LongHashSet uses LongHash.toLong(x, z). So a mod calling contains(chunkXZ2Int(x, z)) would get the wrong answer even if the type matched.

Change

• New io.github.crucible.util.ChunkKeySet, an AbstractSet<Long> backed by the same LongHashSet as before. The Set<Long> methods (add, contains, remove, iterator, including Iterator.remove) take and return vanilla chunkXZ2Int keys and convert them to LongHash internally. Objects that aren't a Long are ignored, like any Set<Long> does.
• The (x, z) overloads and popFirst() stay unboxed and in LongHash encoding. That's the path Crucible's own code uses, so the unload loop performs the same as before.
• ChunkProviderServer declares public Set chunksToUnload again. A new public final ChunkKeySet unloadQueue field points at the same object, and Crucible's callers (ChunkProviderServer, WorldServer, CauldronHooks, CraftWorld) use it.

Code compiled against the old LongHashSet chunksToUnload field needs to switch to unloadQueue. Nothing in this repo still references the old type.

Testing

A minimal coremod with a mixin in Lumi's shape (@Shadow private Set chunksToUnload), loaded through UniMixins 0.3.1. On the first unloadQueuedChunks it queues a far chunk through unloadChunksIfNotNearSpawn and then exercises the field only through java.util.Set with vanilla keys:

Build	Result
before this PR	crash: InvalidMixinException: @Shadow field field_73248_b was not located in the target class, the same as the issue's report
this PR	boots; all checks pass: a chunk queued by the server is visible via contains, the iterator yields the vanilla key, remove/add by key, Iterator.remove, size, and a non-Long argument is ignored

Lumi itself was not run.