文档解析器：支持纯文本与 PDF（pypdf）。
| 20 | |
| 21 | |
| 22 | class DocumentParser: |
| 23 | """文档解析器：支持纯文本与 PDF（pypdf）。""" |
| 24 | |
| 25 | def __init__(self, max_chars: int = 500_000) -> None: |
| 26 | self._max_chars = max_chars |
| 27 | |
| 28 | def parse_file(self, path: Path, mime_type: str | None = None) -> ParsedDocument: |
| 29 | """从文件路径解析。""" |
| 30 | mime = mime_type or "application/octet-stream" |
| 31 | suffix = path.suffix.lower() |
| 32 | if suffix == ".txt" or mime.startswith("text/"): |
| 33 | raw = path.read_text(encoding="utf-8", errors="ignore") |
| 34 | return ParsedDocument(text=raw[: self._max_chars], mime_type=mime, meta={}) |
| 35 | |
| 36 | if suffix == ".pdf" or mime == "application/pdf": |
| 37 | return self._parse_pdf(path.read_bytes(), mime) |
| 38 | |
| 39 | logger.warning("未知类型，按二进制忽略解析 path={}", path) |
| 40 | return ParsedDocument(text="", mime_type=mime, meta={"warning": "unsupported"}) |
| 41 | |
| 42 | def parse_bytes(self, data: bytes, filename: str, mime_type: str | None) -> ParsedDocument: |
| 43 | """从内存字节解析。""" |
| 44 | mime = mime_type or "application/octet-stream" |
| 45 | lower = filename.lower() |
| 46 | if lower.endswith(".txt") or (mime.startswith("text/") and mime != "text/html"): |
| 47 | text = data.decode("utf-8", errors="ignore") |
| 48 | return ParsedDocument(text=text[: self._max_chars], mime_type=mime, meta={}) |
| 49 | |
| 50 | if lower.endswith(".pdf") or mime == "application/pdf": |
| 51 | return self._parse_pdf(data, mime) |
| 52 | |
| 53 | return ParsedDocument(text="", mime_type=mime, meta={"warning": "unsupported"}) |
| 54 | |
| 55 | def _parse_pdf(self, data: bytes, mime: str) -> ParsedDocument: |
| 56 | """使用 pypdf 抽取文本。""" |
| 57 | try: |
| 58 | from pypdf import PdfReader |
| 59 | except ImportError as exc: |
| 60 | logger.exception("未安装 pypdf: {}", exc) |
| 61 | raise |
| 62 | |
| 63 | import io |
| 64 | |
| 65 | reader = PdfReader(io.BytesIO(data)) |
| 66 | parts: list[str] = [] |
| 67 | for page in reader.pages: |
| 68 | try: |
| 69 | parts.append(page.extract_text() or "") |
| 70 | except Exception as exc: |
| 71 | logger.warning("单页 PDF 抽取失败: {}", exc) |
| 72 | |
| 73 | text = "\n".join(parts)[: self._max_chars] |
| 74 | return ParsedDocument(text=text, mime_type=mime, meta={"pages": str(len(reader.pages))}) |