"""简历两阶段并行提取

第一阶段：5路并行提取主表短字段 + 各子表标识名（极快，输出极短）。
第二阶段：N+1路并行提取每条子表记录的短字段 + description/summary 的「行号区间」。
description/summary 不再由 AI 照抄原文，AI 只返回行号区间字符串，由代码从原文单元数组切片还原。
最终组装为与原方案完全一致的 dict 结构（description 为 list[str]），上下游无感知。
"""

import asyncio
import re
import time

from langchain_core.output_parsers import StrOutputParser
from langchain_core.prompts import ChatPromptTemplate

from app.ai.model_config import ResumeExtractorModel
from app.ai.resume_extractor.prompts import (
    OVERVIEW_PROFILE_PROMPT, OVERVIEW_EDUCATION_PROMPT, OVERVIEW_WORK_PROMPT,
    OVERVIEW_PROJECT_PROMPT, OVERVIEW_COMPETITION_PROMPT,
    DETAIL_PROFILE_PROMPT, DETAIL_EDUCATION_PROMPT, DETAIL_WORK_PROMPT,
    DETAIL_INTERNSHIP_PROMPT, DETAIL_PROJECT_PROMPT, DETAIL_COMPETITION_PROMPT,
)
from app.core.logger import log
from app.tool.json_helper import parse_llm_json


# ==================== LLM 调用工具 ====================

def _build_chain(prompt: str):
    """构建提取链：prompt → LLM → 文本输出"""
    return ChatPromptTemplate.from_messages([("system", prompt), ("human", "{text}")]) | ResumeExtractorModel.PARSE | StrOutputParser()


async def _safe_invoke(chain, inp: dict, label: str):
    """单个链调用，记录耗时，失败返回空"""
    start = time.perf_counter()
    try:
        raw = await chain.ainvoke(inp)
        log.info(f"AI提取[{label}]完成，耗时: {time.perf_counter() - start:.2f}s")
        return parse_llm_json(raw)
    except Exception as e:
        log.warning(f"AI提取[{label}]失败，耗时: {time.perf_counter() - start:.2f}s，错误: {e}")
        return None


def _number_segments(segments: list[str]) -> str:
    """将单元数组构造为带行号文本：每行 `[行号] 内容`，行号 0 开始"""
    return "\n".join(f"[{i}] {seg}" for i, seg in enumerate(segments))


# 条目/序号行的起始特征：● ○ • · ▪ ■ ‣ ◆ 等项目符号，或 "1." "2、" "3，" "(4)" "①" 等序号。
# 视觉行解析下，一条 bullet 常被折成多行；只有「条目起始行」才另起段落，其余行视为折行续接。
_LIST_ITEM_RE = re.compile(
    r"^\s*(?:"
    r"[●○◦•·∙▪■□‣◆◇►▶*]"                    # 项目符号
    r"|[-–—]\s"                                # 连字符 + 空格（markdown 风格）
    r"|\d+\s*[.．、，)）]"                       # 阿拉伯数字 + 标点：1. / 2、 / 3，/ 4)
    r"|[（(]\s*\d+\s*[）)]"                      # 括号数字：(1) （2）
    r"|[①②③④⑤⑥⑦⑧⑨⑩⑪⑫⑬⑭⑮⑯⑰⑱⑲⑳]"          # 圈数字
    r")"
)


def _split_into_paragraphs(lines: list[str]) -> list[str]:
    """将一个行号区间内的多行拆分为段落列表：

    - 以项目符号/序号开头的行 → 另起一个新段落；
    - 其余行 → 视为上一段落的折行续接，直连拼接（无分隔符）。
    这样即使 AI 把多条 bullet 塞进同一个区间，也能按条目正确换行。
    """
    paragraphs: list[str] = []
    current = ""
    for line in lines:
        if not current:
            current = line
        elif _LIST_ITEM_RE.match(line):
            paragraphs.append(current)
            current = line
        else:
            current += line
    if current:
        paragraphs.append(current)
    return [p for p in paragraphs if p.strip()]


def _slice_ranges(range_str, segments: list[str]) -> list[str]:
    """按行号区间字符串切片还原段落：逗号分段，段内按条目符号/折行拆分。

    range_str 形如 "3-6,7-9" / "5"；非法或越界 token 静默跳过/clamp。
    返回段落数组（每个逗号段可能因含多条 bullet 而进一步拆成多个段落）。
    """
    if not range_str or not isinstance(range_str, str):
        return []
    n = len(segments)
    paragraphs: list[str] = []
    for token in range_str.split(","):
        token = token.strip()
        if not token:
            continue
        if "-" in token:
            a, _, b = token.partition("-")
            a, b = a.strip(), b.strip()
            if not a.isdigit() or not b.isdigit():
                continue
            start, end = int(a), int(b)
        else:
            if not token.isdigit():
                continue
            start = end = int(token)
        if start > end:
            start, end = end, start
        start = max(0, start)
        end = min(n - 1, end)
        if start > n - 1:
            continue
        paragraphs.extend(_split_into_paragraphs(segments[start:end + 1]))
    return paragraphs


# ==================== 第一阶段：概览 ====================

_overview_profile_chain = _build_chain(OVERVIEW_PROFILE_PROMPT)
_overview_education_chain = _build_chain(OVERVIEW_EDUCATION_PROMPT)
_overview_work_chain = _build_chain(OVERVIEW_WORK_PROMPT)
_overview_project_chain = _build_chain(OVERVIEW_PROJECT_PROMPT)
_overview_competition_chain = _build_chain(OVERVIEW_COMPETITION_PROMPT)


async def _extract_overview(numbered_text: str) -> dict:
    """第一阶段：5路并行提取概览信息"""
    inp = {"text": numbered_text}
    profile, edu_names, work_names, proj_names, comp_names = await asyncio.gather(
        _safe_invoke(_overview_profile_chain, inp, "概览-个人信息"),
        _safe_invoke(_overview_education_chain, inp, "概览-教育"),
        _safe_invoke(_overview_work_chain, inp, "概览-工作实习"),
        _safe_invoke(_overview_project_chain, inp, "概览-项目"),
        _safe_invoke(_overview_competition_chain, inp, "概览-竞赛"),
    )
    return {
        "profile": profile if isinstance(profile, dict) else {},
        "education": edu_names if isinstance(edu_names, list) else [],
        "work": work_names.get("work", []) if isinstance(work_names, dict) else [],
        "internship": work_names.get("internship", []) if isinstance(work_names, dict) else [],
        "project": proj_names if isinstance(proj_names, list) else [],
        "competition": comp_names if isinstance(comp_names, list) else [],
    }


# ==================== 第二阶段：详情 ====================

# 子表模块统一配置（单一数据源）：(模块键, 详情 prompt 模板, 日志标签)
# 概览/详情/组装/日志均复用此清单，新增子表只需在此追加一行。
_DETAIL_MODULES: tuple[tuple[str, str, str], ...] = (
    ("education", DETAIL_EDUCATION_PROMPT, "教育"),
    ("work", DETAIL_WORK_PROMPT, "工作"),
    ("internship", DETAIL_INTERNSHIP_PROMPT, "实习"),
    ("project", DETAIL_PROJECT_PROMPT, "项目"),
    ("competition", DETAIL_COMPETITION_PROMPT, "竞赛"),
)
_SUB_MODULES: tuple[str, ...] = tuple(m[0] for m in _DETAIL_MODULES)


async def _extract_detail(prompt_tpl: str, name: str, numbered_text: str, label: str) -> dict | None:
    """单条子表记录详情提取：用 name 替换 prompt 中的 {name}"""
    chain = _build_chain(prompt_tpl.replace("{name}", name))
    return await _safe_invoke(chain, {"text": numbered_text}, label)


async def _extract_all_details(overview: dict, numbered_text: str) -> dict:
    """第二阶段：根据概览结果，N+1路并行提取所有子表记录详情 + 个人信息补充"""
    # 第 0 路固定为 profile 补充（skills/certificates/summaryRange），其余按子表记录展开
    tasks = [_extract_detail(DETAIL_PROFILE_PROMPT, "", numbered_text, "详情-个人信息补充")]
    task_modules = ["profile_extra"]
    for module, prompt_tpl, label in _DETAIL_MODULES:
        for name in overview[module]:
            tasks.append(_extract_detail(prompt_tpl, name, numbered_text, f"详情-{label}-{name}"))
            task_modules.append(module)

    results = await asyncio.gather(*tasks)
    details: dict[str, list] = {"profile_extra": [], **{m: [] for m in _SUB_MODULES}}
    for module, result in zip(task_modules, results):
        details[module].append(result if isinstance(result, dict) else {})
    return details


# ==================== 组装 ====================

def _assemble(overview: dict, details: dict, segments: list[str]) -> dict:
    """将两阶段结果组装为与原方案一致的 dict 结构（description 还原为 list[str]）"""
    profile = overview["profile"]
    profile_extra = details.get("profile_extra", [{}])[0] if details.get("profile_extra") else {}
    profile["skills"] = (profile_extra.get("skills") or [])[:5]
    profile["certificates"] = profile_extra.get("certificates") or []
    summary_paras = _slice_ranges(profile_extra.get("summaryRange"), segments)
    profile["summary"] = "\n".join(summary_paras) if summary_paras else None

    result = dict(profile)
    for module in _SUB_MODULES:
        items = []
        for item in details.get(module, []):
            item["description"] = _slice_ranges(item.get("descriptionRange"), segments)
            item.pop("descriptionRange", None)
            items.append(item)
        result[module] = items
    return result


# ==================== 入口 ====================

async def extract_all(segments: list[str]) -> dict:
    """两阶段并行提取简历，返回与原方案完全一致的结构化数据

    segments: 文件解析后的「文本单元数组」（PDF按块/docx·txt按行）。
    """
    numbered_text = _number_segments(segments)

    log.info("第一阶段：5路并行概览提取")
    overview = await _extract_overview(numbered_text)
    log.info(
        "概览完成 - " + " ".join(f"{label}:{len(overview[module])}" for module, _, label in _DETAIL_MODULES)
    )

    total = sum(len(overview[m]) for m in _SUB_MODULES)
    log.info(f"第二阶段：{total + 1}路并行详情提取")
    details = await _extract_all_details(overview, numbered_text)

    result = _assemble(overview, details, segments)
    log.info("两阶段提取完成，数据组装完毕")
    return result