"""宽松 JSON 提取：模型常把 LaTeX 的 \\ 写进 JSON，需要修复非法转义。"""
from __future__ import annotations

import json
import re


VALID_JSON_ESCAPES = set('"\\/bfnrtu')


def _repair_json_escapes(content: str) -> str:
    """把字符串里未被正确转义的单个反斜杠补成双反斜杠。"""
    result: list[str] = []
    index = 0
    in_string = False
    length = len(content)
    while index < length:
        char = content[index]
        if not in_string:
            result.append(char)
            if char == '"':
                in_string = True
            index += 1
            continue
        if char != "\\":
            result.append(char)
            if char == '"':
                in_string = False
            index += 1
            continue

        run_end = index
        while run_end < length and content[run_end] == "\\":
            run_end += 1
        run_length = run_end - index
        next_char = content[run_end] if run_end < length else ""
        if (
            run_length % 2 == 1
            and next_char
            and next_char not in VALID_JSON_ESCAPES
        ):
            result.append("\\" * (run_length + 1))
        else:
            result.append("\\" * run_length)
        index = run_end
    return "".join(result)


def extract_json_lax(content: str) -> dict:
    text = content.strip()
    if text.startswith("```"):
        text = re.sub(r"^```(?:json)?\s*", "", text)
        text = re.sub(r"\s*```$", "", text)
    end = text.rfind("}")
    if end <= 0:
        raise ValueError("模型没有返回 JSON，请重试或调整模型")
    starts = [
        match.start()
        for match in re.finditer(r'\{"(questions|chapters)":', text)
    ]
    starts.extend(match.start() for match in re.finditer(r"\{", text))
    starts = sorted(set(starts))
    # 限制扫描数量，避免模型把大量思考文本误当成 JSON
    for start in starts:
        if start >= end:
            break
        payload = text[start : end + 1]
        try:
            return json.loads(payload, strict=False)
        except json.JSONDecodeError:
            repaired = _repair_json_escapes(payload)
            try:
                return json.loads(repaired, strict=False)
            except json.JSONDecodeError:
                continue
    # 没有可解析候选时，给出最后一次错误细节
    payload = text[starts[0] : end + 1] if starts else text
    try:
        json.loads(payload, strict=False)
    except json.JSONDecodeError as exc:
        raise ValueError(
            "模型返回的 JSON 格式不正确"
            f"（{exc.msg}，位置 {exc.pos}），请重试或调整模型"
        ) from exc
    raise ValueError("模型返回的 JSON 格式不正确，请重试或调整模型")