app.py

import os
import gradio as gr
from gradio import ChatMessage
from typing import Iterator, List, Dict, Tuple, Any
import google.generativeai as genai
from huggingface_hub import HfApi
import requests
import re
import traceback

# HuggingFace 관련 API 키 (스페이스 분석 용)
HF_TOKEN = os.getenv("HF_TOKEN")
hf_api = HfApi(token=HF_TOKEN)

# Gemini 2.0 Flash Thinking 모델 관련 API 키 및 클라이언트 (LLM 용)
GEMINI_API_KEY = os.getenv("GEMINI_API_KEY")
genai.configure(api_key=GEMINI_API_KEY)
model = genai.GenerativeModel("gemini-2.0-flash-thinking-exp-01-21")

def get_headers():
    if not HF_TOKEN:
        raise ValueError("Hugging Face token not found in environment variables")
    return {"Authorization": f"Bearer {HF_TOKEN}"}

def get_file_content(space_id: str, file_path: str) -> str:
    file_url = f"https://huggingface.co/spaces/{space_id}/raw/main/{file_path}"
    try:
        response = requests.get(file_url, headers=get_headers())
        if response.status_code == 200:
            return response.text
        else:
            return f"File not found or inaccessible: {file_path}"
    except requests.RequestException:
        return f"Error fetching content for file: {file_path}"

def get_space_structure(space_id: str) -> Dict:
    try:
        files = hf_api.list_repo_files(repo_id=space_id, repo_type="space")
        tree = {"type": "directory", "path": "", "name": space_id, "children": []}
        for file in files:
            path_parts = file.split('/')
            current = tree
            for i, part in enumerate(path_parts):
                if i == len(path_parts) - 1:  # 파일
                    current["children"].append({"type": "file", "path": file, "name": part})
                else:
                    found = False
                    for child in current["children"]:
                        if child["type"] == "directory" and child["name"] == part:
                            current = child
                            found = True
                            break
                    if not found:
                        new_dir = {"type": "directory", "path": '/'.join(path_parts[:i+1]), "name": part, "children": []}
                        current["children"].append(new_dir)
                        current = new_dir
        return tree
    except Exception as e:
        print(f"Error in get_space_structure: {str(e)}")
        return {"error": f"API request error: {str(e)}"}

def format_tree_structure(tree_data: Dict, indent: str = "") -> str:
    if "error" in tree_data:
        return tree_data["error"]
    formatted = f"{indent}{'📁' if tree_data.get('type') == 'directory' else '📄'} {tree_data.get('name', 'Unknown')}\n"
    if tree_data.get("type") == "directory":
        for child in sorted(tree_data.get("children", []), key=lambda x: (x.get("type", "") != "directory", x.get("name", ""))):
            formatted += format_tree_structure(child, indent + "  ")
    return formatted

def adjust_lines_for_code(code_content: str, min_lines: int = 10, max_lines: int = 100) -> int:
    num_lines = len(code_content.split('\n'))
    return min(max(num_lines, min_lines), max_lines)

def analyze_space(url: str, progress=gr.Progress()):
    try:
        space_id = url.split('spaces/')[-1]
        if not re.match(r'^[\w.-]+/[\w.-]+$', space_id):
            raise ValueError(f"Invalid Space ID format: {space_id}")

        progress(0.1, desc="파일 구조 분석 중...")
        tree_structure = get_space_structure(space_id)
        if "error" in tree_structure:
            raise ValueError(tree_structure["error"])
        tree_view = format_tree_structure(tree_structure)

        progress(0.3, desc="app.py 내용 가져오는 중...")
        app_content = get_file_content(space_id, "app.py")

        progress(0.5, desc="코드 요약 중...")
        summary = summarize_code(app_content)

        progress(0.7, desc="코드 분석 중...")
        analysis = analyze_code(app_content)

        progress(0.9, desc="사용법 설명 생성 중...")
        usage = explain_usage(app_content)

        lines_for_app_py = adjust_lines_for_code(app_content)
        progress(1.0, desc="완료")

        return app_content, tree_view, tree_structure, space_id, summary, analysis, usage, lines_for_app_py

    except Exception as e:
        print(f"Error in analyze_space: {str(e)}")
        print(traceback.format_exc())
        return f"오류가 발생했습니다: {str(e)}", "", None, "", "", "", "", 10


# --------------------------------------------------
# Gemini 2.0 Flash Thinking 모델 (LLM) 함수들
# --------------------------------------------------
from gradio import ChatMessage

def format_chat_history(messages: List[ChatMessage]) -> List[Dict]:
    """
    ChatMessage 목록을 Gemini 모델이 이해할 수 있는 형식으로 변환
    (Thinking 메타데이터가 있는 메시지는 무시)
    """
    formatted = []
    for m in messages:
        if hasattr(m, "metadata") and m.metadata:  # 'Thinking' 메시지는 무시
            continue
        role = "assistant" if m.role == "assistant" else "user"
        formatted.append({"role": role, "parts": [m.content or ""]})
    return formatted

import google.generativeai as genai

def gemini_chat_completion(system_message: str, user_message: str, max_tokens: int = 200, temperature: float = 0.7) -> str:
    init_msgs = [
        ChatMessage(role="system", content=system_message),
        ChatMessage(role="user", content=user_message)
    ]
    chat_history = format_chat_history(init_msgs)
    chat = model.start_chat(history=chat_history)
    final = ""
    try:
        for chunk in chat.send_message(user_message, stream=True):
            parts = chunk.candidates[0].content.parts
            if len(parts) == 2:
                final += parts[1].text
            else:
                final += parts[0].text
        return final.strip()
    except Exception as e:
        return f"LLM 호출 중 오류 발생: {str(e)}"


def summarize_code(app_content: str):
    system_msg = "당신은 Python 코드를 분석하고 요약하는 AI 조수입니다. 주어진 코드를 3줄 이내로 간결하게 요약해주세요."
    user_msg = f"다음 Python 코드를 3줄 이내로 요약해주세요:\n\n{app_content}"
    try:
        return gemini_chat_completion(system_msg, user_msg, max_tokens=200, temperature=0.7)
    except Exception as e:
        return f"요약 생성 중 오류 발생: {str(e)}"

def analyze_code(app_content: str):
    system_msg = (
        "You are a deep thinking AI. You may use extremely long chains of thought to deeply consider the problem "
        "and deliberate with yourself via systematic reasoning processes to help come to a correct solution prior to answering. "
        "You should enclose your thoughts and internal monologue inside tags, and then provide your solution or response to the problem. "
        "당신은 Python 코드를 분석하는 AI 조수입니다. 주어진 코드를 분석하여 서비스의 효용성과 활용 측면에서 다음 항목에 대해 설명해주세요:\n"
        "A. 배경 및 필요성\n"
        "B. 기능적 효용성 및 가치\n"
        "C. 특장점\n"
        "D. 적용 대상 및 타겟\n"
        "E. 기대효과\n"
        "기존 및 유사 프로젝트와 비교하여 분석해주세요. Markdown 형식으로 출력하세요."
    )
    user_msg = f"다음 Python 코드를 분석해주세요:\n\n{app_content}"
    try:
        return gemini_chat_completion(system_msg, user_msg, max_tokens=1000, temperature=0.7)
    except Exception as e:
        return f"분석 생성 중 오류 발생: {str(e)}"

def explain_usage(app_content: str):
    system_msg = (
        "You are a deep thinking AI. You may use extremely long chains of thought to deeply consider the problem "
        "and deliberate with yourself via systematic reasoning processes to help come to a correct solution prior to answering. "
        "You should enclose your thoughts and internal monologue inside tags, and then provide your solution or response to the problem. "
        "당신은 Python 코드를 분석하여 사용법을 설명하는 AI 조수입니다. 주어진 코드를 바탕으로 마치 화면을 보는 것처럼 사용법을 상세히 설명해주세요. Markdown 형식으로 출력하세요."
    )
    user_msg = f"다음 Python 코드의 사용법을 설명해주세요:\n\n{app_content}"
    try:
        return gemini_chat_completion(system_msg, user_msg, max_tokens=800, temperature=0.7)
    except Exception as e:
        return f"사용법 설명 생성 중 오류 발생: {str(e)}"

def stream_gemini_response(user_message: str, conversation_state: List[ChatMessage]) -> Iterator[List[ChatMessage]]:
    """
    Gemini에 스트리밍 요청. 
    user_message가 비어 있으면, 아무것도 하지 않고 return.
    """
    if not user_message.strip():
        # 사용자 메시지가 빈 경우: 아무 메시지도 추가 안 함, LLM 호출도 안 함.
        return  # yield도 없이 그냥 종료

    print(f"\n=== New Request ===\nUser message: {user_message}")
    chat_history = format_chat_history(conversation_state)
    chat = model.start_chat(history=chat_history)
    response = chat.send_message(user_message, stream=True)

    thought_buffer = ""
    response_buffer = ""
    thinking_complete = False

    conversation_state.append(
        ChatMessage(
            role="assistant",
            content="",
            metadata={"title": "⚙️ Thinking: *The thoughts produced by the model are experimental"}
        )
    )

    try:
        for chunk in response:
            parts = chunk.candidates[0].content.parts
            current_chunk = parts[0].text

            if len(parts) == 2 and not thinking_complete:
                thought_buffer += current_chunk
                print(f"\n=== Complete Thought ===\n{thought_buffer}")
                conversation_state[-1] = ChatMessage(
                    role="assistant",
                    content=thought_buffer,
                    metadata={"title": "⚙️ Thinking: *The thoughts produced by the model are experimental"}
                )
                yield conversation_state

                response_buffer = parts[1].text
                print(f"\n=== Starting Response ===\n{response_buffer}")
                conversation_state.append(
                    ChatMessage(role="assistant", content=response_buffer)
                )
                thinking_complete = True

            elif thinking_complete:
                response_buffer += current_chunk
                print(f"\n=== Response Chunk ===\n{current_chunk}")
                conversation_state[-1] = ChatMessage(
                    role="assistant",
                    content=response_buffer
                )
            else:
                thought_buffer += current_chunk
                print(f"\n=== Thinking Chunk ===\n{current_chunk}")
                conversation_state[-1] = ChatMessage(
                    role="assistant",
                    content=thought_buffer,
                    metadata={"title": "⚙️ Thinking: *The thoughts produced by the model are experimental"}
                )
            yield conversation_state

        print(f"\n=== Final Response ===\n{response_buffer}")

    except Exception as e:
        print(f"\n=== Error ===\n{str(e)}")
        conversation_state.append(
            ChatMessage(
                role="assistant",
                content=f"I apologize, but encountered an error: {str(e)}"
            )
        )
        yield conversation_state

def convert_for_messages_format(messages: List[ChatMessage]) -> List[Dict[str, str]]:
    """
    ChatMessage 리스트 -> [{"role":"...", "content":"..."}] 목록
    """
    output = []
    for msg in messages:
        output.append({"role": msg.role, "content": msg.content})
    return output

def user_submit_message(msg: str, conversation_state: List[ChatMessage]):
    conversation_state.append(ChatMessage(role="user", content=msg))
    # 입력창 비움
    return "", conversation_state

def respond_wrapper(message: str, conversation_state: List[ChatMessage], max_tokens, temperature, top_p):
    for updated_messages in stream_gemini_response(message, conversation_state):
        yield "", convert_for_messages_format(updated_messages)

def create_ui():
    try:
        css = """
        footer {visibility: hidden;}
        """

        with gr.Blocks(css=css) as demo:
            gr.Markdown("# MOUSE: Space Research Thinking")

            with gr.Tabs():
                with gr.TabItem("분석"):
                    with gr.Row():
                        with gr.Column():
                            url_input = gr.Textbox(label="HuggingFace Space URL")
                            analyze_button = gr.Button("분석")

                            summary_output = gr.Markdown(label="요약")
                            analysis_output = gr.Markdown(label="분석")
                            usage_output = gr.Markdown(label="사용법")
                            tree_view_output = gr.Textbox(label="파일 구조", lines=20)

                        with gr.Column():
                            code_tabs = gr.Tabs()
                            with code_tabs:
                                with gr.TabItem("app.py"):
                                    app_py_content = gr.Code(
                                        language="python",
                                        label="app.py",
                                        lines=50
                                    )
                                with gr.TabItem("requirements.txt"):
                                    requirements_content = gr.Textbox(
                                        label="requirements.txt",
                                        lines=50
                                    )

                with gr.TabItem("AI 코드챗"):
                    gr.Markdown("## 예제를 입력 또는 소스 코드를 붙여넣고 질문하세요")

                    chatbot = gr.Chatbot(
                        label="대화",
                        height=400,
                        type="messages"  
                    )

                    msg = gr.Textbox(
                        label="메시지", 
                        placeholder="메시지를 입력하세요..."
                    )

                    max_tokens = gr.Slider(
                        minimum=1, maximum=8000, 
                        value=4000, label="Max Tokens", 
                        visible=False
                    )
                    temperature = gr.Slider(
                        minimum=0, maximum=1, 
                        value=0.7, label="Temperature", 
                        visible=False
                    )
                    top_p = gr.Slider(
                        minimum=0, maximum=1, 
                        value=0.9, label="Top P", 
                        visible=False
                    )
                    
                    examples = [
                        ["상세한 사용 방법을 4000 토큰 이상 상세히 설명"],
                        ["FAQ 20건을 4000 토큰 이상 작성"],
                        ["기술 차별점, 강점을 중심으로 4000 토큰 이상 설명"],
                        ["특허 출원에 활용 가능한 혁신 아이디어를 4000 토큰 이상 작성"],
                        ["논문 형식으로 4000 토큰 이상 작성"],
                        ["계속 이어서 답변하라"]
                    ]
                    gr.Examples(examples, inputs=msg)

                    conversation_state = gr.State([])

                    msg.submit(
                        user_submit_message, 
                        inputs=[msg, conversation_state], 
                        outputs=[msg, conversation_state],
                        queue=False
                    ).then(
                        respond_wrapper, 
                        inputs=[msg, conversation_state, max_tokens, temperature, top_p], 
                        outputs=[msg, chatbot],
                    )

                with gr.TabItem("Recommended Best"):
                    gr.Markdown(
                        "Discover recommended HuggingFace Spaces [here](https://huggingface.co/spaces/openfree/Korean-Leaderboard)."
                    )

            # 분석 탭 로직
            space_id_state = gr.State()
            tree_structure_state = gr.State()
            app_py_content_lines = gr.State()

            analyze_button.click(
                analyze_space,
                inputs=[url_input],
                outputs=[
                    app_py_content, 
                    tree_view_output, 
                    tree_structure_state, 
                    space_id_state, 
                    summary_output, 
                    analysis_output, 
                    usage_output, 
                    app_py_content_lines
                ]
            ).then(
                lambda space_id: get_file_content(space_id, "requirements.txt"),
                inputs=[space_id_state],
                outputs=[requirements_content]
            ).then(
                lambda lines: gr.update(lines=lines),
                inputs=[app_py_content_lines],
                outputs=[app_py_content]
            )

        return demo

    except Exception as e:
        print(f"Error in create_ui: {str(e)}")
        print(traceback.format_exc())
        raise

if __name__ == "__main__":
    try:
        print("Starting HuggingFace Space Analyzer...")
        demo = create_ui()
        print("UI created successfully.")
        print("Configuring Gradio queue...")
        demo.queue()
        print("Gradio queue configured.")
        print("Launching Gradio app...")
        demo.launch(
            server_name="0.0.0.0",
            server_port=7860,
            share=False,
            debug=True,
            show_api=False
        )
        print("Gradio app launched successfully.")
    except Exception as e:
        print(f"Error in main: {str(e)}")
        print("Detailed error information:")
        print(traceback.format_exc())
        raise