Trainer_LLM/llm-api/plan_router.py

# -*- coding: utf-8 -*-
"""
plan_router.py – v0.12.1 (WP-15)

Minimal-CRUD für Plan-Templates & Pläne (POST/GET) + Idempotenz via Fingerprint.
NEU in v0.12.1:
- Fix: Filter `section` bei `/plans` nutzt jetzt **materialisierte** `plan_section_names` (robust gg. Qdrant-Versionen).
- Neues Payload-Feld beim POST `/plan`: `plan_section_names` (Liste) wird automatisch gesetzt und indexiert.
- Swagger-Doku aktualisiert.
"""
from fastapi import APIRouter, HTTPException, Query
from pydantic import BaseModel, Field
from typing import List, Optional, Dict, Any
from uuid import uuid4
from datetime import datetime, timezone
import hashlib
import json
import os

from clients import model, qdrant
from qdrant_client.models import PointStruct, Filter, FieldCondition, MatchValue, VectorParams, Distance

router = APIRouter(tags=["plans"])

# -----------------
# Konfiguration
# -----------------
PLAN_COLLECTION = os.getenv("PLAN_COLLECTION") or os.getenv("QDRANT_COLLECTION_PLANS", "plans")
PLAN_TEMPLATE_COLLECTION = os.getenv("PLAN_TEMPLATE_COLLECTION", "plan_templates")
PLAN_SESSION_COLLECTION = os.getenv("PLAN_SESSION_COLLECTION", "plan_sessions")
EXERCISE_COLLECTION = os.getenv("EXERCISE_COLLECTION", "exercises")

# -----------------
# Modelle
# -----------------
class TemplateSection(BaseModel):
    name: str
    target_minutes: int
    must_keywords: List[str] = []
    ideal_keywords: List[str] = []
    supplement_keywords: List[str] = []
    forbid_keywords: List[str] = []
    capability_targets: Dict[str, int] = {}

class PlanTemplate(BaseModel):
    id: str = Field(default_factory=lambda: str(uuid4()))
    name: str
    discipline: str
    age_group: str
    target_group: str
    total_minutes: int
    sections: List[TemplateSection] = []
    goals: List[str] = []
    equipment_allowed: List[str] = []
    created_by: str
    version: str

class PlanItem(BaseModel):
    exercise_external_id: str
    duration: int
    why: str

class PlanSection(BaseModel):
    name: str
    items: List[PlanItem] = []
    minutes: int

class Plan(BaseModel):
    id: str = Field(default_factory=lambda: str(uuid4()))
    template_id: Optional[str] = None
    title: str
    discipline: str
    age_group: str
    target_group: str
    total_minutes: int
    sections: List[PlanSection] = []
    goals: List[str] = []
    capability_summary: Dict[str, int] = {}
    novelty_against_last_n: Optional[float] = None
    fingerprint: Optional[str] = None
    created_by: str
    created_at: datetime = Field(default_factory=lambda: datetime.now(timezone.utc))
    source: str = "API"

class PlanTemplateList(BaseModel):
    items: List[PlanTemplate]
    limit: int
    offset: int
    count: int

class PlanList(BaseModel):
    items: List[Plan]
    limit: int
    offset: int
    count: int

# -----------------
# Helpers
# -----------------

def _ensure_collection(name: str):
    if not qdrant.collection_exists(name):
        qdrant.recreate_collection(
            collection_name=name,
            vectors_config=VectorParams(size=model.get_sentence_embedding_dimension(), distance=Distance.COSINE),
        )


def _norm_list(xs: List[str]) -> List[str]:
    seen, out = set(), []
    for x in xs or []:
        s = str(x).strip()
        k = s.casefold()
        if s and k not in seen:
            seen.add(k)
            out.append(s)
    return sorted(out, key=str.casefold)


def _template_embed_text(tpl: PlanTemplate) -> str:
    parts = [tpl.name, tpl.discipline, tpl.age_group, tpl.target_group]
    parts += tpl.goals
    parts += [s.name for s in tpl.sections]
    return ". ".join([p for p in parts if p])


def _plan_embed_text(p: Plan) -> str:
    parts = [p.title, p.discipline, p.age_group, p.target_group]
    parts += p.goals
    parts += [s.name for s in p.sections]
    return ". ".join([p for p in parts if p])


def _embed(text: str):
    return model.encode(text or "").tolist()


def _fingerprint_for_plan(p: Plan) -> str:
    core = {
        "title": p.title,
        "total_minutes": int(p.total_minutes),
        "items": [
            {"exercise_external_id": it.exercise_external_id, "duration": int(it.duration)}
            for sec in p.sections
            for it in (sec.items or [])
        ],
    }
    raw = json.dumps(core, sort_keys=True, ensure_ascii=False)
    return hashlib.sha256(raw.encode("utf-8")).hexdigest()


def _get_by_field(collection: str, key: str, value: Any) -> Optional[Dict[str, Any]]:
    flt = Filter(must=[FieldCondition(key=key, match=MatchValue(value=value))])
    pts, _ = qdrant.scroll(collection_name=collection, scroll_filter=flt, limit=1, with_payload=True)
    if not pts:
        return None
    payload = dict(pts[0].payload or {})
    payload.setdefault("id", str(pts[0].id))
    return payload


def _as_model(model_cls, payload: Dict[str, Any]):
    fields = getattr(model_cls, "model_fields", None) or getattr(model_cls, "__fields__", {})
    allowed = set(fields.keys())
    data = {k: payload[k] for k in payload.keys() if k in allowed}
    return model_cls(**data)


def _truthy(val: Optional[str]) -> bool:
    return str(val or "").strip().lower() in {"1", "true", "yes", "on"}


def _exists_in_collection(collection: str, key: str, value: Any) -> bool:
    flt = Filter(must=[FieldCondition(key=key, match=MatchValue(value=value))])
    pts, _ = qdrant.scroll(collection_name=collection, scroll_filter=flt, limit=1, with_payload=False)
    return bool(pts)

# -----------------
# CRUD (bestehend)
# -----------------
@router.post(
    "/plan_templates",
    response_model=PlanTemplate,
    summary="Create a plan template",
    description=(
        "Erstellt ein Plan-Template (Strukturplanung).\n\n"
        "• Mehrere Sections erlaubt.\n"
        "• Section-Felder: must/ideal/supplement/forbid keywords + capability_targets.\n"
        "• Materialisierte Facettenfelder (section_*) werden intern geschrieben, um Qdrant-Filter zu beschleunigen."
    ),
)
def create_plan_template(t: PlanTemplate):
    _ensure_collection(PLAN_TEMPLATE_COLLECTION)
    payload = t.model_dump()
    payload["goals"] = _norm_list(payload.get("goals"))
    sections = payload.get("sections", []) or []
    for s in sections:
        s["must_keywords"] = _norm_list(s.get("must_keywords") or [])
        s["ideal_keywords"] = _norm_list(s.get("ideal_keywords") or [])
        s["supplement_keywords"] = _norm_list(s.get("supplement_keywords") or [])
        s["forbid_keywords"] = _norm_list(s.get("forbid_keywords") or [])

    payload["section_names"] = _norm_list([s.get("name", "") for s in sections])
    payload["section_must_keywords"] = _norm_list([kw for s in sections for kw in (s.get("must_keywords") or [])])
    payload["section_ideal_keywords"] = _norm_list([kw for s in sections for kw in (s.get("ideal_keywords") or [])])
    payload["section_supplement_keywords"] = _norm_list([kw for s in sections for kw in (s.get("supplement_keywords") or [])])
    payload["section_forbid_keywords"] = _norm_list([kw for s in sections for kw in (s.get("forbid_keywords") or [])])

    vec = _embed(_template_embed_text(t))
    qdrant.upsert(collection_name=PLAN_TEMPLATE_COLLECTION, points=[PointStruct(id=str(t.id), vector=vec, payload=payload)])
    return t


@router.get(
    "/plan_templates/{tpl_id}",
    response_model=PlanTemplate,
    summary="Read a plan template by id",
    description="Liest ein Template anhand seiner ID und gibt nur die Schemafelder zurück (zusätzliche Payload wird herausgefiltert).",
)
def get_plan_template(tpl_id: str):
    _ensure_collection(PLAN_TEMPLATE_COLLECTION)
    found = _get_by_field(PLAN_TEMPLATE_COLLECTION, "id", tpl_id)
    if not found:
        raise HTTPException(status_code=404, detail="not found")
    return _as_model(PlanTemplate, found)


@router.post(
    "/plan",
    response_model=Plan,
    summary="Create a concrete training plan",
    description=(
        "Erstellt einen konkreten Trainingsplan.\n\n"
        "Idempotenz: gleicher Fingerprint (title + items) → gleicher Plan (kein Duplikat).\n"
        "Optional: Validierung von template_id und Exercises (Strict-Mode)."
    ),
)
def create_plan(p: Plan):
    _ensure_collection(PLAN_COLLECTION)
    if p.template_id:
        if not _exists_in_collection(PLAN_TEMPLATE_COLLECTION, "id", p.template_id):
            raise HTTPException(status_code=422, detail=f"Unknown template_id: {p.template_id}")

    if _truthy(os.getenv("PLAN_STRICT_EXERCISES")):
        missing: List[str] = []
        for sec in p.sections or []:
            for it in sec.items or []:
                ex