app/core/chunk_payload.py aktualisiert
All checks were successful
Deploy mindnet to llm-node / deploy (push) Successful in 3s

This commit is contained in:
Lars 2025-11-08 21:22:17 +01:00
parent 2ddf034983
commit 2de786fc64

View File

@ -1,158 +1,220 @@
# app/core/chunk_payload.py
# Line count: 214
from __future__ import annotations
from typing import Any, Dict, Iterable, List, Optional, Union
# ---- Helpers ----
def _coerce_float(val: Any) -> Optional[float]:
from typing import Any, Dict, Iterable, List, Optional, Tuple, Union
def _get(obj: Any, key: str, default: Any = None) -> Any:
if obj is None:
return default
if hasattr(obj, key):
try:
val = getattr(obj, key)
return val if val is not None else default
except Exception:
pass
if isinstance(obj, dict):
if key in obj:
val = obj.get(key, default)
return val if val is not None else default
return default
def _get_frontmatter(note: Any) -> Dict[str, Any]:
fm = _get(note, "frontmatter", None)
if isinstance(fm, dict):
return fm
meta = _get(note, "meta", None)
if isinstance(meta, dict) and isinstance(meta.get("frontmatter"), dict):
return meta["frontmatter"]
return {}
def _get_from_frontmatter(fm: Dict[str, Any], key: str, default: Any = None) -> Any:
if not isinstance(fm, dict):
return default
if key in fm:
val = fm.get(key, default)
return val if val is not None else default
return default
def _coerce_tags(val: Any) -> List[str]:
if val is None:
return None
try:
if isinstance(val, (int, float)):
return float(val)
if isinstance(val, str):
v = val.strip()
if not v:
return None
return float(v.replace(",", "."))
except Exception:
return None
return None
def _extract_weight(frontmatter: Dict[str, Any], explicit: Optional[float]) -> Optional[float]:
if explicit is not None:
return _coerce_float(explicit)
if frontmatter is None:
return None
if "retriever_weight" in frontmatter:
return _coerce_float(frontmatter.get("retriever_weight"))
# also accept nested style: retriever: { weight: 0.8 }
retriever = frontmatter.get("retriever")
if isinstance(retriever, dict) and "weight" in retriever:
return _coerce_float(retriever.get("weight"))
return None
def _ensure_list(x: Any) -> List[Any]:
if x is None:
return []
if isinstance(x, list):
return x
return [x]
if isinstance(val, list):
return [str(x) for x in val]
if isinstance(val, str):
parts = [t.strip() for t in val.split(",")]
return [p for p in parts if p]
return []
def _resolve_note_id(frontmatter: Dict[str, Any], kw_note_id: Optional[str]) -> Optional[str]:
if kw_note_id:
return kw_note_id
if not isinstance(frontmatter, dict):
return None
return frontmatter.get("id") or frontmatter.get("note_id")
def _base_fields(frontmatter: Dict[str, Any], note_id: Optional[str], path: str) -> Dict[str, Any]:
title = None
typ = None
tags = None
if isinstance(frontmatter, dict):
title = frontmatter.get("title")
typ = frontmatter.get("type") or frontmatter.get("note_type")
# tags can be list[str] or comma separated string
tags = frontmatter.get("tags")
if isinstance(tags, str):
tags = [t.strip() for t in tags.split(",") if t.strip()]
def _resolve_retriever_weight(
fm: Dict[str, Any],
explicit: Optional[float],
) -> Optional[float]:
if explicit is not None:
return explicit
val = _get_from_frontmatter(fm, "retriever_weight", None)
if isinstance(val, (int, float)):
return float(val)
retr = fm.get("retriever")
if isinstance(retr, dict):
v = retr.get("weight")
if isinstance(v, (int, float)):
return float(v)
return None
def _resolve_note_fields(note: Any) -> Dict[str, Any]:
fm = _get_frontmatter(note)
note_id = _get_from_frontmatter(fm, "id", None)
if note_id is None:
note_id = _get(note, "note_id", None)
if note_id is None:
note_id = _get(note, "id", None)
title = _get_from_frontmatter(fm, "title", None)
if title is None:
title = _get(note, "title", None)
ntype = _get_from_frontmatter(fm, "type", None)
if ntype is None:
ntype = _get(note, "type", None)
tags = _get_from_frontmatter(fm, "tags", None)
if tags is None:
tags = _get(note, "tags", None)
tags = _coerce_tags(tags)
path = _get_from_frontmatter(fm, "path", None)
if path is None:
path = _get(note, "path", None)
if path is None:
path = _get(note, "source", None)
if path is None:
path = _get(note, "filepath", None)
return {
"note_id": note_id,
"title": title,
"type": typ,
"type": ntype,
"tags": tags,
"path": path or None,
"path": path,
"frontmatter": fm,
}
# ---- Public API ----
def _extract_chunk_text_and_index(
chunk: Any,
fallback_index: int,
) -> Tuple[str, int]:
"""
Akzeptiert verschiedene Chunk-Formate:
- str (reiner Text)
- dict mit keys: text | window | body | content
- Objekt mit Attributen: text | window | body | content
- (text, idx) Tuple
"""
# Tuple (text, idx)
if isinstance(chunk, tuple) and len(chunk) == 2 and isinstance(chunk[0], str):
txt, idx = chunk
try:
idx_int = int(idx)
except Exception:
idx_int = fallback_index
return txt, idx_int
# String
if isinstance(chunk, str):
return chunk, fallback_index
# Dict
if isinstance(chunk, dict):
txt = (
chunk.get("text")
or chunk.get("window")
or chunk.get("body")
or chunk.get("content")
)
if isinstance(txt, str):
idx = chunk.get("index")
try:
idx_int = int(idx) if idx is not None else fallback_index
except Exception:
idx_int = fallback_index
return txt, idx_int
# Objekt mit Attributen
for attr in ("text", "window", "body", "content"):
if hasattr(chunk, attr):
try:
txt = getattr(chunk, attr)
except Exception:
txt = None
if isinstance(txt, str):
# Optionale "index"-Quelle
idx = None
if hasattr(chunk, "index"):
try:
idx = getattr(chunk, "index")
except Exception:
idx = None
try:
idx_int = int(idx) if idx is not None else fallback_index
except Exception:
idx_int = fallback_index
return txt, idx_int
# Wenn nichts passt -> klarer Fehler
raise ValueError("Unsupported chunk format: cannot extract text/index")
def make_chunk_payloads(
frontmatter: Dict[str, Any],
*args,
note_id: Optional[str] = None,
chunks: Optional[Iterable[Any]] = None,
path: str = "",
chunk_profile: Optional[str] = None,
note: Any,
chunks: Iterable[Any],
*,
retriever_weight: Optional[float] = None,
**kwargs,
base_payload: Optional[Dict[str, Any]] = None,
) -> List[Dict[str, Any]]:
"""
Build chunk payload dictionaries for Qdrant.
This function is intentionally permissive to stay compatible with older callers:
- If `chunks` is a list of dictionaries that already contain payload-like fields,
those are augmented.
- If `chunks` is a list of strings, minimal payloads are created.
- If `chunks` is a list of dicts with keys like `text`, `window`, or `index`, they are normalized.
Always injects `retriever_weight` into each payload when available (from explicit arg or frontmatter).
Erzeugt Qdrant-Payloads für Chunk-Punkte.
- Kopiert Note-Metadaten (note_id/title/type/tags/path)
- Schreibt text + chunk_index je Chunk
- Setzt retriever_weight, wenn vorhanden/angegeben
"""
# Backward-compat for callers that might pass via kwargs
if chunks is None:
chunks = kwargs.get("payloads") or kwargs.get("pls") or kwargs.get("items") or kwargs.get("chunk_items")
note_id_resolved = _resolve_note_id(frontmatter, note_id)
weight = _extract_weight(frontmatter, retriever_weight)
base = _base_fields(frontmatter, note_id_resolved, path)
out: List[Dict[str, Any]] = []
for idx, item in enumerate(_ensure_list(chunks)):
# Case A: already a full payload dict (heuristic: has 'text' or 'window' or 'note_id' keys)
if isinstance(item, dict) and ("text" in item or "window" in item or "note_id" in item):
pl = dict(item) # shallow copy
# ensure base fields exist if missing
for k, v in base.items():
pl.setdefault(k, v)
# ensure chunk_index if not present
pl.setdefault("chunk_index", item.get("index", idx))
# inject retriever_weight
if weight is not None:
pl["retriever_weight"] = weight
out.append(pl)
continue
note_fields = _resolve_note_fields(note)
fm = note_fields["frontmatter"]
rw = _resolve_retriever_weight(fm, retriever_weight)
# Case B: item is a dict with nested 'payload'
if isinstance(item, dict) and "payload" in item and isinstance(item["payload"], dict):
pl = dict(item["payload"])
for k, v in base.items():
pl.setdefault(k, v)
pl.setdefault("chunk_index", pl.get("index", idx))
if weight is not None:
pl["retriever_weight"] = weight
out.append(pl)
continue
# Basisfelder, die jeder Chunk tragen soll
common: Dict[str, Any] = {}
if base_payload:
common.update({k: v for k, v in base_payload.items() if v is not None})
# Case C: item is a plain string -> treat as text (no window context)
if isinstance(item, str):
text_val = item
pl = {
**base,
"chunk_index": idx,
"text": text_val,
"window": text_val,
}
if weight is not None:
pl["retriever_weight"] = weight
out.append(pl)
continue
if note_fields.get("note_id") is not None:
common["note_id"] = note_fields["note_id"]
if note_fields.get("title") is not None:
common["title"] = note_fields["title"]
if note_fields.get("type") is not None:
common["type"] = note_fields["type"]
if note_fields.get("tags"):
common["tags"] = note_fields["tags"]
if note_fields.get("path") is not None:
common["path"] = note_fields["path"]
if rw is not None:
common["retriever_weight"] = rw
# Case D: item has 'text'/'window' under different names
if isinstance(item, dict):
text_val = item.get("text") or item.get("body") or item.get("content") or ""
window_val = item.get("window") or text_val
pl = {
**base,
"chunk_index": item.get("chunk_index", item.get("index", idx)),
"text": text_val,
"window": window_val,
}
if weight is not None:
pl["retriever_weight"] = weight
out.append(pl)
continue
# Fallback: minimal payload
pl = {**base, "chunk_index": idx}
if weight is not None:
pl["retriever_weight"] = weight
out.append(pl)
for i, ch in enumerate(chunks):
text, idx = _extract_chunk_text_and_index(ch, i)
payload = dict(common) # copy
payload["chunk_index"] = idx
payload["text"] = text
out.append(payload)
return out