utils/thinking_parser.py

"""<think> 태그 파싱 유틸리티"""

import re
from typing import Tuple, Optional


def parse_thinking_response(response: str) -> Tuple[Optional[str], str]:
    """
    응답에서 <think> 태그 추출

    Returns:
        (thinking_content, clean_response)
    """
    if not response:
        return None, ""

    # <think>...</think> 패턴 매칭
    pattern = r'<think>(.*?)</think>'
    match = re.search(pattern, response, re.DOTALL)

    if match:
        thinking = match.group(1).strip()
        # <think> 태그 제거한 클린 응답
        clean = re.sub(pattern, '', response, flags=re.DOTALL).strip()
        return thinking, clean

    return None, response


def format_thinking_for_display(thinking: str) -> str:
    """Thinking 내용을 마크다운으로 포맷팅"""
    if not thinking:
        return "*No thinking process*"

    # 6단계 구조 하이라이트 (있는 경우)
    sections = [
        "[상황분석]", "[관계단계]", "[캐릭터스타일]",
        "[밀당결정]", "[금지패턴체크]", "[응답설계]"
    ]

    formatted = thinking
    for section in sections:
        formatted = formatted.replace(
            section,
            f"**{section}**"
        )

    return formatted


def extract_response_only(full_response: str) -> str:
    """Thinking 제거하고 응답만 추출"""
    _, clean = parse_thinking_response(full_response)
    return clean


def has_thinking_tag(response: str) -> bool:
    """응답에 <think> 태그가 있는지 확인"""
    pattern = r'<think>.*?</think>'
    return bool(re.search(pattern, response, re.DOTALL))


def get_thinking_stats(response: str) -> dict:
    """Thinking 관련 통계"""
    thinking, clean = parse_thinking_response(response)

    return {
        "has_thinking": thinking is not None,
        "thinking_length": len(thinking) if thinking else 0,
        "response_length": len(clean),
        "total_length": len(response),
    }