scripts/rag_query.py

"""Query the SCUT-SS RAG library.



    python scripts/rag_query.py "软件测试的黑盒测试有哪些方法"

    python scripts/rag_query.py "数据库范式" -k 5 --course "数据库系统"

    python scripts/rag_query.py "解释一下 ATAM" --ask          # needs LLM env vars



LLM env vars (optional, any OpenAI-compatible endpoint):

    OPENAI_API_KEY / OPENAI_BASE_URL / OPENAI_MODEL

"""

import argparse
import json
import sys
import textwrap
from pathlib import Path

sys.path.insert(0, str(Path(__file__).resolve().parent))
import llm  # noqa: E402
from vectorstore import Index  # noqa: E402

ROOT = Path(__file__).resolve().parent.parent
DEFAULT_INDEX = ROOT / "vectorstore"

PROMPT = """你是华南理工大学软件学院的学习助手。请只依据下面提供的资料回答问题；

资料里没有的内容请明确说明"资料中没有提到"，不要编造。回答用中文，尽量给出要点式结构，

并在关键结论后标注来源（用 [1] [2] 对应资料编号）。



资料：

{context}



问题：{question}

"""


def retrieve(index, query, k, course=None):
    hits = index.search(query, k=max(k * 6, 20))
    if course:
        hits = [h for h in hits if course in (h[0].get("course") or "")] or hits
    return hits[:k]


def build_context(hits):
    parts = []
    for i, (chunk, score) in enumerate(hits, 1):
        parts.append(
            f"[{i}] 来源：{chunk['path']}｜{chunk.get('section', '')}\nURL: {chunk.get('url', '')}\n{chunk['text']}\n"
        )
    return "\n".join(parts)


def ask_llm(prompt, provider=None, base_url=None, api_key=None, model=None):
    """Call a cloud model (any OpenAI-compatible endpoint). Returns (answer, conf) or (None, err)."""
    try:
        answer, conf = llm.chat(
            [
                {"role": "system", "content": "你是一个严谨的课程学习助手。"},
                {"role": "user", "content": prompt},
            ],
            provider=provider,
            base_url=base_url,
            api_key=api_key,
            model=model,
        )
        return answer, conf
    except Exception as exc:  # noqa: BLE001
        return None, llm.http_error_hint(exc)


def main():
    ap = argparse.ArgumentParser()
    ap.add_argument("question")
    ap.add_argument("-k", type=int, default=5)
    ap.add_argument("--course", default=None, help="limit to one course, e.g. 软件测试")
    ap.add_argument("--index", default=str(DEFAULT_INDEX))
    ap.add_argument("--ask", action="store_true", help="call a cloud LLM with the retrieved context")
    ap.add_argument("--provider", default=None, help="one of: " + ", ".join(llm.available_presets()))
    ap.add_argument("--base-url", default=None)
    ap.add_argument("--model", default=None)
    ap.add_argument(
        "--save-context",
        default=None,
        metavar="FILE",
        help="dump question+context to a Markdown file so an external AI (IDE/chat) can read it",
    )
    ap.add_argument("--json", action="store_true")
    args = ap.parse_args()

    for _s in (sys.stdout, sys.stderr):
        try:
            _s.reconfigure(encoding="utf-8", errors="replace")
        except Exception:  # noqa: BLE001
            pass

    index = Index(args.index)
    hits = retrieve(index, args.question, args.k, args.course)

    if args.json:
        print(json.dumps([{"score": s, **c} for c, s in hits], ensure_ascii=False, indent=2))
        return

    context = build_context(hits)

    if args.save_context:
        out = Path(args.save_context)
        out.parent.mkdir(parents=True, exist_ok=True)
        out.write_text(
            "# 请基于以下资料回答问题\n\n"
            f"**问题**：{args.question}\n\n"
            "**要求**：只依据下列资料作答；资料未覆盖的内容请明确说明「资料中没有提到」；"
            "关键结论后标注来源编号 [1] [2]。\n\n"
            "## 检索到的资料\n\n" + context,
            encoding="utf-8",
        )
        print(f"上下文已写入 {out} —— 在 IDE / 聊天窗口里 @ 这个文件，让外部模型回答即可。")

    if args.ask:
        answer, conf = ask_llm(
            PROMPT.format(context=context, question=args.question),
            provider=args.provider,
            base_url=args.base_url,
            model=args.model,
        )
        if answer is None:
            print(f"调用云端模型失败：{conf}\n下面是检索到的资料，可用 --save-context 导出后交给外部模型：\n")
        else:
            print(f"[{conf.get('provider') or 'custom'} | {conf['model']} @ {conf['base_url']}]\n")
            print(answer)
            print("\n" + "=" * 60 + "\n")

    print(f"检索到 {len(hits)} 条资料（backend={index.meta['backend']}）：\n")
    for i, (chunk, score) in enumerate(hits, 1):
        head = f"[{i}] {chunk['path']}" + (f" · {chunk['section']}" if chunk.get("section") else "")
        print(head)
        print(f"    score={score:.4f}  {chunk.get('url','')}")
        print(textwrap.indent(textwrap.shorten(chunk["text"], 300, placeholder=" ..."), "    "))
        print()


if __name__ == "__main__":
    main()