Configuration

中文

Lumberjack accepts concrete components, so configure built-in splitting by constructing a splitter with its tokenizer:

from lumberjack import Lumberjack
from lumberjack.block import BlockConfig, BlockKind, MarkdownTableConfig
from lumberjack.splitter import SiblingSplitter
from lumberjack.tokenizer import ApproxByteTokenizer

tokenizer = ApproxByteTokenizer()
splitter = SiblingSplitter(
    tokenizer,
    max_tokens=800,
    ideal_max_tokens_ratio=0.8,
    merge_below_ratio=0.125,
    heading_sensitive=True,
    block_options=(
        MarkdownTableConfig(max_tokens=500, repeat_header=True),
        BlockConfig(BlockKind.CODE_FENCE, split=False),
    ),
)
jack = Lumberjack(tokenizer=tokenizer, splitter=splitter)

block_options is a sequence of typed configuration objects. Duplicate block kinds are rejected. Use MarkdownTableConfig for Markdown tables and HTMLTableConfig for HTML tables; both can repeat headers. BlockConfig covers the other built-in kinds, and CustomBlockConfig covers a parser plugin's custom kind.

Field	Effect
isolated	Keeps the block in its own draft rather than packing it with neighbours.
split	Allows or prevents fallback splitting for the block. A non-splittable oversized block may exceed the budget.
max_tokens	Overrides the splitter budget for that block kind.
repeat_header	Repeats the table header for every generated table piece.

CLI and Web API inputs use mapping forms at their boundary. The Python API deliberately does not accept a dict for block_options; pass typed objects so invalid names and duplicate policies are found early.