mindnet/scripts/import_markdown.py

#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
import_markdown.py — Markdown → Qdrant (Notes, Chunks, Edges) mit Change-Detection & optionalem Purge

Version: 3.0 (2025-09-08)

Änderungen ggü. bisherigen Varianten:
- Stabiler Qdrant-Delete über Filter-API (kompatibel mit aktuellen Clients).
- Option --purge-before-upsert: löscht NUR Chunks/Edges der geänderten Note(n) vor Upsert.
- Change-Detection ausschließlich über Note-Body-Hash (Frontmatter/Dateidatum ignoriert).
- Robuste Edge-Erzeugung:
  - Chunk-Level: belongs_to, prev/next, references (+backlink)
  - Fallback: Note-Level references, falls Chunk-Payloads keine references liefern
- Korrekte Qdrant-Scroll/Query-Nutzung (Rückgabesignaturen neuer Clients).
- Saubere Dry-Run/Apply-Ausgaben je Note als JSON.

Aufruf:
  python3 -m scripts.import_markdown --vault ./vault
  python3 -m scripts.import_markdown --vault ./vault --apply
  python3 -m scripts.import_markdown --vault ./vault --apply --purge-before-upsert
  python3 -m scripts.import_markdown --vault ./vault --note-id 20250821-architektur-ki-trainerassistent-761cfe --apply

Parameter:
  --vault PATH               Pflicht. Root des (Obsidian-)Vaults.
  --apply                    Wirkt schreibend in Qdrant; ohne Flag nur Dry-Run.
  --purge-before-upsert      Vor dem Upsert (nur) für geänderte Notes: zugehörige Chunks & Edges löschen.
  --note-id ID               Optional: Nur eine Note mit genau dieser ID verarbeiten.
  --embed-note               Optional: zusätzlich Volltext-Vektor für die Note erzeugen.
  --force-replace            Ignoriert Change-Detection (erzwingt Upsert + optionalen Purge).

Umgebungsvariablen (optional):
  QDRANT_URL                 z.B. http://127.0.0.1:6333  (falls QDRANT_HOST/PORT nicht gesetzt)
  QDRANT_API_KEY             API Key (falls erforderlich)
  QDRANT_HOST/QDRANT_PORT    Alternative zu QDRANT_URL
  COLLECTION_PREFIX          Default "mindnet"
  VECTOR_DIM                 Default 384

Voraussetzungen:
  - app/core:
      parser.read_markdown, parser.normalize_frontmatter, parser.validate_required_frontmatter
      note_payload.make_note_payload  (liefert u.a. hash_fulltext)
      validate_note.validate_note_payload
      chunker.assemble_chunks
      chunk_payload.make_chunk_payloads   (idealerweise inkl. 'references' je Chunk; wenn nicht → Fallback)
      embed.embed_texts, embed.embed_one
      qdrant.QdrantConfig, qdrant.get_client, qdrant.ensure_collections
      qdrant_points.points_for_note / points_for_chunks / points_for_edges / upsert_batch
      edges.deriv_edges_for_note          (nutzt Chunk-Payloads; wir ergänzen Fallback)
"""

from __future__ import annotations
import argparse
import os
import sys
import json
import glob
from typing import List, Dict, Tuple, Optional
from dotenv import load_dotenv
from qdrant_client.http import models as rest

from app.core.parser import read_markdown, normalize_frontmatter, validate_required_frontmatter
from app.core.note_payload import make_note_payload
from app.core.validate_note import validate_note_payload
from app.core.chunker import assemble_chunks
from app.core.chunk_payload import make_chunk_payloads
from app.core.embed import embed_texts, embed_one
from app.core.qdrant import QdrantConfig, get_client, ensure_collections
from app.core.qdrant_points import (
    points_for_chunks,
    points_for_note,
    points_for_edges,
    upsert_batch,
)
from app.core.edges import deriv_edges_for_note


# -----------------------------
# Utils
# -----------------------------

def iter_md(root: str) -> List[str]:
    files = [p for p in glob.glob(os.path.join(root, "**", "*.md"), recursive=True)]
    out = []
    for p in files:
        pn = p.replace("\\", "/")
        if any(ex in pn for ex in ["/.obsidian/", "/_backup_frontmatter/", "/_imported/"]):
            continue
        out.append(p)
    return sorted(out)


def collections(prefix: str) -> Tuple[str, str, str]:
    return f"{prefix}_notes", f"{prefix}_chunks", f"{prefix}_edges"


def fetch_existing_note_hash(client, prefix: str, note_id: str) -> Optional[str]:
    notes_col, _, _ = collections(prefix)
    f = rest.Filter(
        must=[rest.FieldCondition(key="note_id", match=rest.MatchValue(value=note_id))]
    )
    # neuere Clients: scroll gibt (points, next_page_offset) zurück
    points, _ = client.scroll(
        collection_name=notes_col,
        scroll_filter=f,
        with_payload=True,
        with_vectors=False,
        limit=1,
    )
    if not points:
        return None
    pl = points[0].payload or {}
    return pl.get("hash_fulltext")


def purge_note_artifacts(client, prefix: str, note_id: str) -> None:
    """
    Löscht NUR Chunks & Edges der angegebenen Note (vor Upsert).
    - Chunks: payload.note_id == note_id
    - Edges : payload.source_id startswith note_id#c  ODER payload.target_id startswith note_id#c
              plus note-scope edges, falls vorhanden (source_id == note_id oder target_id == note_id)
    """
    _, chunks_col, edges_col = collections(prefix)

    # Chunks der Note löschen
    f_chunks = rest.Filter(
        must=[rest.FieldCondition(key="note_id", match=rest.MatchValue(value=note_id))]
    )
    client.delete(collection_name=chunks_col, points_selector=f_chunks, wait=True)

    # Edges der Note löschen (Chunk- und Note-Scope)
    should = [
        rest.FieldCondition(key="source_id", match=rest.MatchText(text=f"{note_id}#")),
        rest.FieldCondition(key="target_id", match=rest.MatchText(text=f"{note_id}#")),
        rest.FieldCondition(key="source_id", match=rest.MatchValue(value=note_id)),
        rest.FieldCondition(key="target_id", match=rest.MatchValue(value=note_id)),
    ]
    f_edges = rest.Filter(should=should)
    client.delete(collection_name=edges_col, points_selector=f_edges, wait=True)


def has_chunk_level_refs(chunk_payloads: List[Dict]) -> bool:
    return any(isinstance(ch.get("references"), list) and ch["references"] for ch in chunk_payloads)


def fallback_note_level_edges(note_meta: Dict, chunk_payloads: List[Dict]) -> List[Dict]:
    """
    Falls chunk_payloads keine 'references' enthalten, erzeugen wir
    - belongs_to + prev/next aus chunk_payloads
    - references/backlink auf Note-Scope aus note_meta['references'] (falls vorhanden)
    """
    edges: List[Dict] = []

    # belongs_to + prev/next (Chunk-Scope)
    for ch in chunk_payloads:
        src = ch["id"]
        edges.append({"src_id": src, "dst_id": note_meta["id"], "edge_type": "belongs_to", "scope": "chunk"})
        nb = ch.get("neighbors") or {}
        prev_id = nb.get("prev")
        next_id = nb.get("next")
        if prev_id:
            edges.append({"src_id": src, "dst_id": prev_id, "edge_type": "next", "scope": "chunk"})
            edges.append({"src_id": prev_id, "dst_id": src, "edge_type": "prev", "scope": "chunk"})
        if next_id:
            edges.append({"src_id": src, "dst_id": next_id, "edge_type": "next", "scope": "chunk"})
            edges.append({"src_id": next_id, "dst_id": src, "edge_type": "prev", "scope": "chunk"})

    # Note-Scope references/backlink
    for tid in (note_meta.get("references") or []):
        edges.append({"src_id": note_meta["id"], "dst_id": tid, "edge_type": "references", "scope": "note"})
        edges.append({"src_id": tid, "dst_id": note_meta["id"], "edge_type": "backlink", "scope": "note"})

    # Dedupe
    uniq = {}
    for e in edges:
        k = (e["src_id"], e["edge_type"], e["dst_id"], e.get("scope", "note"))
        uniq[k] = e
    return list(uniq.values())


# -----------------------------
# Main
# -----------------------------

def main():
    load_dotenv()
    ap = argparse.ArgumentParser()
    ap.add_argument("--vault", required=True, help="Pfad zum Obsidian-Vault (Root-Ordner)")
    ap.add_argument("--apply", action="store_true", help="Schreibt in Qdrant; ohne Flag nur Dry-Run")
    ap.add_argument("--purge-before-upsert", action="store_true",
                    help="Vor Upsert Chunks & Edges der GEÄNDERTEN Note löschen")
    ap.add_argument("--note-id", help="Nur eine bestimmte Note-ID verarbeiten")
    ap.add_argument("--embed-note", action="store_true", help="Optional: Note-Volltext einbetten")
    ap.add_argument("--force-replace", action="store_true",
                    help="Änderungserkennung ignorieren und immer upserten (+ optional Purge)")
    args = ap.parse_args()

    # Qdrant-Config
    cfg = QdrantConfig.from_env()
    client = get_client(cfg)
    ensure_collections(client, cfg.prefix, cfg.dim)
    notes_col, chunks_col, edges_col = collections(cfg.prefix)

    root = os.path.abspath(args.vault)
    files = iter_md(root)
    if not files:
        print("Keine Markdown-Dateien gefunden.", file=sys.stderr)
        sys.exit(2)

    processed = 0
    for path in files:
        parsed = read_markdown(path)
        fm = normalize_frontmatter(parsed.frontmatter)

        # Pflichtfelder?
        try:
            validate_required_frontmatter(fm)
        except Exception:
            # Überspringen, aber nicht abbrechen
            continue

        if args.note_id and fm.get("id") != args.note_id:
            continue

        processed += 1

        # Note-Payload & Validierung
        note_pl = make_note_payload(parsed, vault_root=root)
        validate_note_payload(note_pl)
        note_id = note_pl["note_id"]

        # Change-Detection (Body-Hash)
        new_hash = note_pl.get("hash_fulltext")
        old_hash = None
        if not args.force_replace:
            old_hash = fetch_existing_note_hash(client, cfg.prefix, note_id)

        changed = args.force_replace or (old_hash != new_hash)

        # Chunks + Embeddings
        chunks = assemble_chunks(fm["id"], parsed.body, fm.get("type", "concept"))
        chunk_pls = make_chunk_payloads(fm, note_pl["path"], chunks)
        vecs = embed_texts([c.text for c in chunks])

        # Optional: Note-Vektor
        note_vec = embed_one(parsed.body) if args.embed_note else None

        # Edges
        # 1) Versuch: volle Kanten aus Chunk-Payloads
        edges = deriv_edges_for_note(fm, chunk_pls)
        # 2) Fallback, falls keine Chunk-References vorhanden sind
        if not edges or (not has_chunk_level_refs(chunk_pls) and (fm.get("references") or note_pl.get("references"))):
            # Ergänze fm['references'] falls noch nicht im fm vorhanden
            if "references" not in fm and "references" in note_pl:
                fm["references"] = note_pl["references"]
            edges = fallback_note_level_edges(fm, chunk_pls)

        # Zusammenfassung für Log
        summary = {
            "note_id": note_id,
            "title": fm.get("title"),
            "chunks": len(chunk_pls),
            "edges": len(edges),
            "changed": changed,
            "decision": ("apply" if args.apply and changed else
                         "apply-skip-unchanged" if args.apply and not changed else
                         "dry-run"),
            "path": note_pl["path"],
        }
        print(json.dumps(summary, ensure_ascii=False))

        # Apply?
        if not args.apply:
            continue

        # Optionaler Purge NUR für geänderte Notes
        if changed and args.purge_before_upsert:
            purge_note_artifacts(client, cfg.prefix, note_id)

        # Upserts
        # Notes
        notes_name, note_pts = points_for_note(cfg.prefix, note_pl, note_vec, cfg.dim)
        upsert_batch(client, notes_name, note_pts)

        # Chunks
        chunks_name, chunk_pts = points_for_chunks(cfg.prefix, chunk_pls, vecs)
        upsert_batch(client, chunks_name, chunk_pts)

        # Edges
        edges_name, edge_pts = points_for_edges(cfg.prefix, edges)
        upsert_batch(client, edges_name, edge_pts)

    print(f"Done. Processed notes: {processed}")


if __name__ == "__main__":
    main()