MZC/KnowMathBot/backend/app/routers/assessment.py

"""
题库与测评路由：习题 CRUD、作答提交、规则判题与大模型兜底、错题与权重更新。
"""
from uuid import UUID
from fastapi import APIRouter, Depends, HTTPException
import httpx
import json
import re
from sqlmodel import Session, select
from app.core.config import settings
from app.db import get_session
from app.deps import get_current_user, require_role
from app.models import Exercise, ExerciseAttempt, User, UserRole, WrongBookItem
from app.integrations.neo4j import Neo4jClient, get_neo4j_client
from app.schemas import (
    AttemptPublic,
    AttemptSubmitRequest,
    ExerciseBatchRequest,
    ExerciseCreateRequest,
    ExercisePublic,
)

from app.utils.exercise_constants import RESOLUTION_EMPTY_PLACEHOLDER, normalize_resolution_for_store
from app.utils.knowledge_codes import split_knowledge_codes
from app.utils.neo4j_exercise import enrich_exercise_from_neo4j

router = APIRouter(prefix="/assessment", tags=["assessment"])


def _normalize_answer_text(s: str) -> str:
    s = (s or "").strip()
    if not s:
        return ""
    s = s.replace("＝", "=").replace("，", ",")
    s = re.sub(r"\s+", "", s)
    return s

def _normalize_boolean_judgment(s: str) -> str | None:
    """
    Map common true/false answer forms to canonical "T" / "F".
    Used when question_type == "判断".
    """
    raw = _normalize_answer_text(s)
    if not raw:
        return None
    low = raw.lower()
    if raw in {"对", "是", "√", "正确"} or low in {"t", "true", "y", "yes", "1", "right"}:
        return "T"
    if raw in {"错", "否", "×", "错误"} or low in {"f", "false", "n", "no", "0", "wrong"}:
        return "F"
    if len(raw) == 1 and low == "x":
        return "F"
    return None

def _extract_standard_answer(body: str) -> str | None:
    if not body:
        return None
    m = re.search(
        r"(标准答案|参考答案|正确答案)\s*[:：]\s*(?P<ans>.+?)(?:\n\s*(评分标准|评分|解析|题解)\s*[:：]|$)",
        body,
        flags=re.I | re.S,
    )
    if not m:
        return None
    ans = (m.group("ans") or "").strip()
    return ans if ans else None

def _extract_expected_answer(ex: Exercise) -> str | None:
    """
    Prefer teacher-provided `correct_answer`.
    Fallback to parsing `body` ("标准答案/正确答案/参考答案...") for legacy data.
    """
    if ex.correct_answer:
        s = str(ex.correct_answer).strip()
        return s if s else None
    return _extract_standard_answer(ex.body)

def _try_rule_based_grade(ex: Exercise, student_answer: str) -> tuple[bool | None, float | None]:
    expected = _extract_expected_answer(ex)
    if not expected:
        return None, None

    expected_candidates = [c.strip() for c in re.split(r"[|/;；\n]+", expected) if c.strip()]
    stu = _normalize_answer_text(student_answer)
    if not stu:
        return False, 0.0

    if (getattr(ex, "question_type", None) or "").strip() == "判断":
        st_j = _normalize_boolean_judgment(student_answer)
        if st_j:
            for cand in expected_candidates:
                exp_j = _normalize_boolean_judgment(cand)
                if exp_j and st_j == exp_j:
                    return True, 1.0
            # 学生已是明确的“对/错”表述，但标准答案无法解析为布尔：退回字符串规则
            if any(_normalize_boolean_judgment(c) for c in expected_candidates):
                return False, 0.0

    for cand in expected_candidates:
        cand_norm = _normalize_answer_text(cand)
        if cand_norm and cand_norm == stu:
            return True, 1.0
        if cand_norm and (cand_norm in stu or stu in cand_norm):
            return True, 0.8

    return False, 0.0

def _extract_json_from_text(text: str) -> dict | None:
    if not text:
        return None
    i = text.find("{")
    j = text.rfind("}")
    if i < 0 or j <= i:
        return None
    try:
        return json.loads(text[i : j + 1])
    except Exception:
        return None

def _judge_with_deepseek(ex: Exercise, student_answer: str) -> tuple[bool, float, str]:
    api_key = settings.deepseek_api_key.strip()
    if not api_key:
        return False, 0.0, "DeepSeek 未配置 API Key"

    messages = [
        {
            "role": "system",
            "content": "你是中学数学自动判题助理。请严格按要求输出JSON，不要输出多余文本。",
        },
        {
            "role": "user",
            "content": (
                "请判断学生的作答是否正确，并给出置信度分数score。\n"
                f"题目标题：{ex.title}\n"
                f"题目内容：{ex.body}\n"
                f"学生答案：{student_answer}\n\n"
                "输出JSON格式（必须严格可解析）："
                "{\"is_correct\": true/false, \"score\": 0.0-1.0, \"reason\": \"一句话简要\"}"
            ),
        },
    ]

    payload = {"model": settings.deepseek_model, "messages": messages, "temperature": 0.0}
    url = f"{settings.deepseek_base_url.rstrip('/')}/chat/completions"
    headers = {"Authorization": f"Bearer {api_key}", "Content-Type": "application/json"}

    try:
        with httpx.Client(timeout=settings.llm_request_timeout_seconds) as client:
            res = client.post(url, json=payload, headers=headers)
            res.raise_for_status()
            data = res.json()
            content = data["choices"][0]["message"]["content"]
    except Exception as e:
        return False, 0.0, f"DeepSeek调用失败: {e}"

    obj = _extract_json_from_text(content)
    if not isinstance(obj, dict):
        return False, 0.0, "DeepSeek返回非JSON"

    is_correct = obj.get("is_correct", False)
    if isinstance(is_correct, str):
        is_correct = is_correct.strip().lower() in {"true", "1", "yes", "y"}

    score = obj.get("score", 0.0)
    try:
        score_f = float(score)
    except Exception:
        score_f = 0.0
    score_f = max(0.0, min(1.0, score_f))

    reason = str(obj.get("reason", "")).strip()
    return bool(is_correct), score_f, reason

def _grade_attempt(ex: Exercise, student_answer: str) -> tuple[bool, float]:
    """判题入口：先规则匹配标答，无法判定时再调用大模型返回 JSON。"""
    # 规则判题（教师提供 correct_answer 或题干内标准答案）
    rule_ok, rule_score = _try_rule_based_grade(ex, student_answer)
    if rule_ok is not None:
        return bool(rule_ok), float(rule_score or 0.0)

    is_correct, score, _reason = _judge_with_deepseek(ex, student_answer)
    return bool(is_correct), float(score)

@router.post("/exercises", response_model=ExercisePublic)
def create_exercise(
    payload: ExerciseCreateRequest,
    session: Session = Depends(get_session),
    neo4j: Neo4jClient = Depends(get_neo4j_client),
    _=Depends(require_role(UserRole.teacher)),
):
    ex = Exercise(**payload.model_dump())

    if getattr(ex, "weight", None) is None:
        ex.weight = 0
    ex.resolution = normalize_resolution_for_store(getattr(ex, "resolution", None))

    enrich_exercise_from_neo4j(ex, neo4j)

    session.add(ex)
    session.commit()
    session.refresh(ex)
    return ex

@router.get("/exercises", response_model=list[ExercisePublic])
def list_exercises(knowledge_code: str | None = None, session: Session = Depends(get_session)):
    exs = list(session.exec(select(Exercise)).all())
    if not knowledge_code:
        return exs
    codes = set(split_knowledge_codes(knowledge_code))
    if not codes:
        return exs
    out: list[Exercise] = []
    for ex in exs:
        ex_codes = set(split_knowledge_codes(ex.knowledge_code))
        if ex_codes & codes:
            out.append(ex)
    return out

@router.post("/exercises/batch", response_model=list[ExercisePublic])
def get_exercises_batch(payload: ExerciseBatchRequest, session: Session = Depends(get_session)):
    """按 ID 列表批量返回习题，保持请求顺序；缺失 ID 时返回 404。"""
    if not payload.ids:
        return []

    seen: set[str] = set()
    ordered_ids: list[UUID] = []
    for eid in payload.ids:
        key = str(eid)
        if key in seen:
            continue
        seen.add(key)
        ordered_ids.append(eid)

    rows = list(session.exec(select(Exercise).where(Exercise.id.in_(ordered_ids))).all())
    by_id = {ex.id: ex for ex in rows}

    out: list[Exercise] = []
    for eid in ordered_ids:
        ex = by_id.get(eid)
        if not ex:
            raise HTTPException(status_code=404, detail=f"Exercise not found: {eid}")
        out.append(ex)
    return out

@router.get("/exercises/{exercise_id}", response_model=ExercisePublic)
def get_exercise(exercise_id: UUID, session: Session = Depends(get_session)):
    ex = session.get(Exercise, exercise_id)
    if not ex:
        raise HTTPException(status_code=404, detail="Exercise not found")
    return ex

@router.post("/exercises/{exercise_id}/attempts", response_model=AttemptPublic)
def submit_attempt(
    exercise_id: UUID,
    payload: AttemptSubmitRequest,
    session: Session = Depends(get_session),
    current_user: User = Depends(get_current_user),
):
    """学生提交作答：判题 → 写作答记录 → 更新题目 weight → 错误时写入错题本。"""
    ex = session.get(Exercise, exercise_id)
    if not ex:
        raise HTTPException(status_code=404, detail="Exercise not found")

    is_correct, score = _grade_attempt(ex, payload.answer)

    # 个性化权重：错题升权、做对降权（最低 0），驱动后续推荐优先级。
    cur_w = int(getattr(ex, "weight", 0) or 0)
    if is_correct is False:
        ex.weight = cur_w + 1
    else:
        ex.weight = max(0, cur_w - 1)

    attempt = ExerciseAttempt(
        user_id=current_user.id,
        exercise_id=exercise_id,
        answer=payload.answer,
        is_correct=is_correct,
        score=score,
    )
    session.add(attempt)

    # 将本次错误按知识点写入错题本，便于个性推荐与学习分析统计。
    if is_correct is False:
        correct_answer = _extract_expected_answer(ex)
        codes = split_knowledge_codes(ex.knowledge_code)
        for code in codes:
            session.add(
                WrongBookItem(
                    user_id=current_user.id,
                    exercise_id=exercise_id,
                    knowledge_code=code,
                    wrong_answer=payload.answer,
                    correct_answer=correct_answer,
                )
            )

    session.commit()
    session.refresh(attempt)
    return attempt

@router.get("/attempts/me", response_model=list[AttemptPublic])
def my_attempts(session: Session = Depends(get_session), current_user: User = Depends(get_current_user)):
    return list(session.exec(select(ExerciseAttempt).where(ExerciseAttempt.user_id == current_user.id)).all())