Coverage for src/qdrant_loader/core/chunking/docling/outcome.py: 100%

1"""The chunker -> mapper handoff type. 

2 

3A :class:`Chunk` is the engine-neutral unit the chunker emits: the body text, the 

4heading-contextualised text that will actually be embedded, and the 

5projected :class:`~.structure.ChunkStructure`. It is the contract boundary between 

6the docling-touching :mod:`.docling_chunker` and the docling-free :mod:`.mapper` 

7that turns chunks into qdrant-loader ``Document`` chunks — so the mapper, and its 

8tests, never need docling. 

9""" 

10 

11from __future__ import annotations 

12 

13from dataclasses import dataclass 

14 

15from .structure import ChunkStructure 

16 

17 

18@dataclass(frozen=True, slots=True) 

19class Chunk: 

20 """One chunk, before projection into the qdrant-loader ``Document`` contract.""" 

21 

22 text: str # the chunk body — docling DocChunk.text 

23 # contextualize() output. Contextual embedding is NOT YET IMPLEMENTED, so this 

24 # mirrors text while the toggle is off; it is the carrier for when it lands. 

25 embed_text: str 

26 structure: ChunkStructure 

27 token_count: int | None = None # tokens in embed_text — drives the budget check