generateBaseTerrain now factors the noise+surface stage out of chunk
generation so structure placement can read pre-carve heights exactly
like vanilla's STRUCTURE_STARTS stage. RuinedPortalGenerationPoint
replays the ruined_portals set: the weighted pick across all seven
variants on its own stream, each attempt's setup/template/rotation/
mirror draws on a fresh GenerationContext stream (makeRandom reseeds
per attempt, confirmed from bytecode), findSuitableY with corner-column
settle scan against those heights, and the post-draw 3D-biome filter at
quart-snapped coordinates.
Known gap, documented in STRUCTURE_NOTES.md: no variant is accepted in
the window around the fixture's portal yet while vanilla placed one;
the biome/surface inputs need a vanilla-side probe to bisect. Not wired
into the production replay until that closes.
LoadResolvedTemplate parses the extracted .nbt files with every palette
entry resolved to concrete state IDs through an injected resolver, and
TransformBlockPos reproduces StructureTemplate.transform's mirror-then-
rotate order around the pivot, including all three quarter-turn cases
read off the bytecode. MthGetSeed gives processors their per-block
positional Legacy stream. Verified against the real ruined_portal
templates: every block lands inside the declared size for all eight
mirror/rotation combinations.
Captures the decoded placement grid, the four legacy frequency reducers,
the full ruined-portal findGenerationPoint and findSuitableY algorithms
with their exact draw order, the template name arrays, and the fixture
marker coordinates that pin the grid to vanilla. This is the working map
for the piece-placement ports that follow.
Loads the 26.1.2 structure datapack: 20 structure sets, 34 structure
definitions, and the worldgen biome tags they reference. Random-spread
placement follows RandomSpreadStructurePlacement byte for byte — region
floorDiv by spacing, one Legacy stream salted per region, two
spread-type draws for the offset — and the four legacy frequency
reducers keep their exact draws, including the hardcoded buried-treasure
salt and the outpost's discarded full-range int.
Verified against the fixture: on seed 12345 the ruined_portals set
claims exactly chunk (1,0), the chunk whose captured blocks carry the
ruined portal's obsidian, gold, and crying obsidian. TestStructure-
PlacementGrid pins that plus village separation invariants.
Ports MonsterRoomFeature from bytecode with vanilla draw order and loop
structure: shell validation (solid floor and ceiling, one to five open
side columns), interior cave_air carving, cobblestone walls with the
mossy three-in-four floor roll drawn only where vanilla draws it, chest
placement faced away from its single opaque neighbour with the loot-seed
long consumed even though block entities are not modelled yet, and the
spawner's one-draw mob pick. Writes go through Feature.safeSetBlock
semantics against #minecraft:features_cannot_replace, captured verbatim
from the jar into the embedded datapack data.
The pass slots between geodes and the ores so the rooms' cave_air
pockets exist before ore ellipsoids roll their air-exposure discards.
The committed fixture's four chunks happen to contain no dungeons, so
measured parity holds at 98.028%, but the generator's output can differ
elsewhere: generatorVersion bumps to 22.
TestVanillaBaseTerrainParity compares our noise+surface+carver+aquifer+
vein pipeline against a vanilla capture whose biomes carry no features
and no structure sets: all 393,216 cells match. The plain-state fixture
mismatches the base-terrain diagnostic used to attribute to carver or
aquifer defects are therefore feature outputs wearing plain states —
monster-room cave_air, geode voids, lava-lake lava — and its doc comment
now says so.
Splits fixture mismatches into classes that name the responsible
subsystem: plain-vs-plain cells are direct base-terrain defects and come
with sample coordinates, single-side feature states count as flips per
family, and disk-shaped outputs land in an ambiguous bucket so they
cannot pollute the plain signal.
First run on seed 12345: 966 base cells (0.246%), dominated by vanilla-
carved cave segments missing from our carver output and deepslate-to-
water aquifer edges.
The combined stage-6 pass in placeScheduledUndergroundOresStage already
replays ores, underwater magma, and disks in datapack order; the separate
placeScheduledDisks and placeScheduledUnderwaterMagma passes have no
callers left. Keep the per-position helpers the combined pass and tests
use.
The router's noise caves are one kind of cave. The other kind -- the long
winding tunnels with rooms and side branches, and the ravines that cut down
through the terrain -- is walked, step by step, by a random source, and none of
it existed.
The shape of the work is unusual enough to state plainly. To carve one chunk,
vanilla replays every carver seeded in the 17x17 chunks around it and keeps only
what lands inside, so the same tunnel is walked up to 289 times across a world.
That redundancy is the point: it is what lets a chunk be carved without
generating its neighbours, which is the only way carving fits a generator that
produces one chunk at a time. A carve-once-write-into-neighbours design would be
cheaper and would not reproduce vanilla's mask and ordering.
Two primitives had to be right before any of it could be, and both are pinned
against values captured from the jar:
* setLargeFeatureSeed, which decides which chunks start a cave. It combines
its two products with XOR; setDecorationSeed, which it otherwise resembles,
uses addition and forces the low bit. Getting them the wrong way round moves
every tunnel in the world and nothing complains.
* Mth.sin and Mth.cos, which are a 65536-entry lookup table and not libm.
Mth.sin(-1.0) is -0.8414514 against Math.sin's -0.8414709848078965, and a
tunnel that walks by adding cos(yaw) a hundred times ends up somewhere else
entirely if that difference is smoothed away.
Carving lands between the surface pass and decoration, where vanilla puts it,
and both neighbours matter: the surface rules must already have placed grass for
a cave mouth to be retextured, and decoration must come after so nothing is
planted over a hole. The heights decoration plants against are recomputed
afterwards, which is why vanilla re-primes its heightmaps at the start of the
feature step.
The configs are extracted from the jar rather than transcribed, along with the
flattened #minecraft:overworld_carver_replaceables tag, so the probabilities and
Y ranges are data. Open volume below y=60 rises 28% over sixteen sampled chunks,
tunnels cut at or below y=-56 fill with lava rather than air (869 blocks, no
air), and the cost is inside the noise floor of the density pass.
