Added chunk types (left separate: broken and slow)
This commit is contained in:
parent
4ba6b67d75
commit
d2f9820b86
1 changed files with 190 additions and 0 deletions
190
src/aiocraft/mc/chunk.py
Normal file
190
src/aiocraft/mc/chunk.py
Normal file
|
@ -0,0 +1,190 @@
|
||||||
|
import io
|
||||||
|
import math
|
||||||
|
import logging
|
||||||
|
|
||||||
|
from typing import Dict, Tuple, Any
|
||||||
|
|
||||||
|
import numpy as np
|
||||||
|
|
||||||
|
from aiocraft.mc.types import VarInt, Short, UnsignedByte, Type, Context
|
||||||
|
|
||||||
|
class BitStream:
|
||||||
|
data : bytes
|
||||||
|
cursor : int
|
||||||
|
size : int
|
||||||
|
def __init__(self, data:bytes, size:int):
|
||||||
|
self.data = data
|
||||||
|
self.cursor = 0
|
||||||
|
self.size = size if size > 0 else len(self.data) * 8
|
||||||
|
def __len__(self) -> int:
|
||||||
|
return self.size - self.cursor
|
||||||
|
def read(self, size:int) -> int:
|
||||||
|
if len(self) < size:
|
||||||
|
raise ValueError(f"Not enough bits ({len(self)} left, {size} requested)")
|
||||||
|
# Calculate splice indexes
|
||||||
|
start_byte = (self.cursor//8)
|
||||||
|
end_byte = math.ceil((self.cursor + size) / 8)
|
||||||
|
# Construct int from bytes
|
||||||
|
buf = int.from_bytes(
|
||||||
|
self.data[start_byte:end_byte],
|
||||||
|
byteorder='little', signed=False
|
||||||
|
)
|
||||||
|
# Trim extra bytes
|
||||||
|
end_offset = (self.cursor + size) % 8
|
||||||
|
if end_offset > 0:
|
||||||
|
buf = buf >> (8 - end_offset) # There's an extra 1 to the left in air, maybe shift 1 bit less?
|
||||||
|
start_offset = self.cursor % 8
|
||||||
|
buf = buf & (( 1 << size ) - 1)
|
||||||
|
# Increment and return
|
||||||
|
self.cursor += size
|
||||||
|
return buf
|
||||||
|
|
||||||
|
class PalettedContainer(Type):
|
||||||
|
pytype : type
|
||||||
|
threshold : int
|
||||||
|
size : int
|
||||||
|
|
||||||
|
def __init__(self, threshold:int, size:int):
|
||||||
|
self.threshold = threshold
|
||||||
|
self.size = size
|
||||||
|
|
||||||
|
def write(self, data, buffer:io.BytesIO, ctx:Context):
|
||||||
|
raise NotImplementedError
|
||||||
|
|
||||||
|
def read(self, buffer:io.BytesIO, ctx:Context):
|
||||||
|
bits = UnsignedByte.read(buffer, ctx=ctx) # FIXME if bits > 4 it reads trash
|
||||||
|
if bits < 4:
|
||||||
|
bits = 4
|
||||||
|
if bits >= self.threshold:
|
||||||
|
bits = 13 # this should not be hardcoded but we have no way to calculate all possible block states
|
||||||
|
palette_len = VarInt.read(buffer, ctx=ctx)
|
||||||
|
palette = np.zeros((palette_len,), dtype='int32')
|
||||||
|
for i in range(palette_len):
|
||||||
|
palette[i] = VarInt.read(buffer, ctx=ctx)
|
||||||
|
container_size = VarInt.read(buffer, ctx=ctx)
|
||||||
|
stream = BitStream(buffer.read(container_size * 8), container_size*8*8) # a Long is 64 bits long
|
||||||
|
section = np.zeros((self.size, self.size, self.size), dtype='int32')
|
||||||
|
for y in range(self.size):
|
||||||
|
for z in range(self.size):
|
||||||
|
for x in range(self.size):
|
||||||
|
val = stream.read(bits)
|
||||||
|
if val >= len(palette):
|
||||||
|
logging.warning("out of bounds : %d (%d)", val, len(palette)) # FIXME happens a lot!
|
||||||
|
section[x, y, z] = val
|
||||||
|
continue
|
||||||
|
section[x, y, z] = palette[val] if bits < self.threshold else val
|
||||||
|
return section
|
||||||
|
|
||||||
|
BiomeContainer = PalettedContainer(4, 4)
|
||||||
|
BlockStateContainer = PalettedContainer(9, 16)
|
||||||
|
|
||||||
|
class HalfByteArrayType(Type):
|
||||||
|
size : int
|
||||||
|
|
||||||
|
def __init__(self, size:int):
|
||||||
|
self.size = size
|
||||||
|
|
||||||
|
def write(self, data, buffer:io.BytesIO, ctx:Context):
|
||||||
|
raise NotImplementedError
|
||||||
|
|
||||||
|
def read(self, buffer:io.BytesIO, ctx:Context):
|
||||||
|
section = np.empty((self.size, self.size, self.size), dtype='int32')
|
||||||
|
bit_buffer = BitStream(buffer.read((self.size**3)//2), (self.size**3)*4)
|
||||||
|
for y in range(self.size):
|
||||||
|
for z in range(self.size):
|
||||||
|
for x in range(self.size):
|
||||||
|
section[x, y, z] = bit_buffer.read(4)
|
||||||
|
return section
|
||||||
|
|
||||||
|
BlockLightSection = HalfByteArrayType(16)
|
||||||
|
SkyLightSection = HalfByteArrayType(16)
|
||||||
|
|
||||||
|
class NewChunkSectionType(Type):
|
||||||
|
pytype : type
|
||||||
|
|
||||||
|
def write(self, data, buffer:io.BytesIO, ctx:Context):
|
||||||
|
raise NotImplementedError
|
||||||
|
|
||||||
|
def read(self, buffer:io.BytesIO, ctx:Context):
|
||||||
|
block_count = Short.read(buffer, ctx=ctx)
|
||||||
|
block_states = BlockStateContainer.read(buffer, ctx=ctx)
|
||||||
|
biomes = BiomeContainer.read(buffer, ctx=ctx)
|
||||||
|
return (
|
||||||
|
block_count,
|
||||||
|
block_states,
|
||||||
|
biomes
|
||||||
|
)
|
||||||
|
|
||||||
|
class OldChunkSectionType(Type):
|
||||||
|
pytype : type
|
||||||
|
|
||||||
|
def write(self, data, buffer:io.BytesIO, ctx:Context):
|
||||||
|
raise NotImplementedError
|
||||||
|
|
||||||
|
def read(self, buffer:io.BytesIO, ctx:Context):
|
||||||
|
block_states = BlockStateContainer.read(buffer, ctx=ctx)
|
||||||
|
block_light = BlockLightSection.read(buffer, ctx=ctx)
|
||||||
|
if ctx.overworld:
|
||||||
|
sky_light = SkyLightSection.read(buffer, ctx=ctx)
|
||||||
|
else:
|
||||||
|
sky_light = np.empty((16, 16, 16), dtype='int32')
|
||||||
|
return (
|
||||||
|
block_states,
|
||||||
|
block_light,
|
||||||
|
sky_light
|
||||||
|
)
|
||||||
|
|
||||||
|
ChunkSection = OldChunkSectionType()
|
||||||
|
|
||||||
|
class Chunk(Type):
|
||||||
|
x : int
|
||||||
|
z : int
|
||||||
|
bitmask : int
|
||||||
|
ground_up_continuous : bool
|
||||||
|
blocks : np.ndarray
|
||||||
|
block_light : np.ndarray
|
||||||
|
sky_light : np.ndarray
|
||||||
|
biomes: bytes
|
||||||
|
|
||||||
|
def __init__(self, x:int, z:int, bitmask:int, ground_up_continuous:bool):
|
||||||
|
self.x = x
|
||||||
|
self.z = z
|
||||||
|
self.bitmask = bitmask
|
||||||
|
self.blocks = np.zeros((16, 256, 16), dtype='int32')
|
||||||
|
self.block_light = np.zeros((16, 256, 16), dtype='int32')
|
||||||
|
self.sky_light = np.zeros((16, 256, 16), dtype='int32')
|
||||||
|
self.ground_up_continuous = ground_up_continuous
|
||||||
|
|
||||||
|
def __getitem__(self, item:Any):
|
||||||
|
return self.blocks[item]
|
||||||
|
|
||||||
|
def read(self, buffer:io.BytesIO, ctx:Context):
|
||||||
|
for i in range(16):
|
||||||
|
if (self.bitmask >> i) & 1:
|
||||||
|
block_states, block_light, sky_light = ChunkSection.read(buffer, ctx=ctx)
|
||||||
|
self.blocks[:, i*16 : (i+1)*16, :] = block_states
|
||||||
|
self.block_light[:, i*16 : (i+1)*16, :] = block_light
|
||||||
|
self.sky_light[:, i*16 : (i+1)*16, :] = sky_light
|
||||||
|
if self.ground_up_continuous:
|
||||||
|
self.biomes = buffer.read(256) # 16x16
|
||||||
|
if buffer.read():
|
||||||
|
logging.warning("Leftover data in chunk buffer")
|
||||||
|
return self
|
||||||
|
|
||||||
|
class World:
|
||||||
|
chunks : Dict[Tuple[int, int], Chunk]
|
||||||
|
|
||||||
|
def __init__(self):
|
||||||
|
self.chunks = {}
|
||||||
|
|
||||||
|
def __getitem__(self, item:Tuple[int, int, int]):
|
||||||
|
return self.get(*item)
|
||||||
|
|
||||||
|
def get(self, x:int, y:int, z:int) -> int:
|
||||||
|
coord = (x//16, z//16)
|
||||||
|
if coord not in self.chunks:
|
||||||
|
raise KeyError(f"Chunk {coord} not loaded")
|
||||||
|
return self.chunks[coord][int(x%16), int(y), int(z%16)]
|
||||||
|
|
||||||
|
def put(self, chunk:Chunk, x:int, z:int):
|
||||||
|
self.chunks[(x,z)] = chunk
|
Loading…
Reference in a new issue