docvortex 0.4.21__tar.gz → 0.4.23__tar.gz

{docvortex-0.4.21/src/docvortex.egg-info → docvortex-0.4.23}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: docvortex
-Version: 0.4.21
+Version: 0.4.23
 Summary: A fast, multi-format document parsing and conversion engine
 License-Expression: MIT
 Project-URL: Repository, https://github.com/myhloli/DocVortex
@@ -127,17 +127,19 @@ use `overwrite=True` in Python or `--overwrite` in the CLI to replace them.
 
 ## Supported formats
 
-### Native inputs · 15 formats
+### Native inputs · 17 formats
 
-| Document family | Formats |
+| Document family | File extensions |
 | --- | --- |
-| PDF with native text | PDF |
-| Word & rich text | DOC, DOCX, RTF |
-| Presentations | PPT, PPTX |
-| Spreadsheets | XLS, XLSX, CSV, TSV |
-| OpenDocument | ODT, ODS, ODP |
-| E-books & web documents | EPUB, HTML |
-| Open Fixed-layout Document | OFD |
+| PDF with native text | `.pdf` |
+| Word & rich text | `.doc`, `.docx`, `.rtf` |
+| Presentations | `.ppt`, `.pptx` |
+| Spreadsheets | `.xls`, `.xlsx`, `.csv`, `.tsv` |
+| OpenDocument | `.odt`, `.ods`, `.odp` |
+| E-books | `.epub` |
+| Web documents | `.html`, `.htm`, `.shtml` |
+| Web archives | `.mhtml`, `.mht` |
+| Open Fixed-layout Document | `.ofd` |
 
 ### Outputs · 7 formats
 

{docvortex-0.4.21 → docvortex-0.4.23}/README.md

@@ -73,17 +73,19 @@ use `overwrite=True` in Python or `--overwrite` in the CLI to replace them.
 
 ## Supported formats
 
-### Native inputs · 15 formats
+### Native inputs · 17 formats
 
-| Document family | Formats |
+| Document family | File extensions |
 | --- | --- |
-| PDF with native text | PDF |
-| Word & rich text | DOC, DOCX, RTF |
-| Presentations | PPT, PPTX |
-| Spreadsheets | XLS, XLSX, CSV, TSV |
-| OpenDocument | ODT, ODS, ODP |
-| E-books & web documents | EPUB, HTML |
-| Open Fixed-layout Document | OFD |
+| PDF with native text | `.pdf` |
+| Word & rich text | `.doc`, `.docx`, `.rtf` |
+| Presentations | `.ppt`, `.pptx` |
+| Spreadsheets | `.xls`, `.xlsx`, `.csv`, `.tsv` |
+| OpenDocument | `.odt`, `.ods`, `.odp` |
+| E-books | `.epub` |
+| Web documents | `.html`, `.htm`, `.shtml` |
+| Web archives | `.mhtml`, `.mht` |
+| Open Fixed-layout Document | `.ofd` |
 
 ### Outputs · 7 formats
 

{docvortex-0.4.21 → docvortex-0.4.23}/schemas/middle-2.0.json

@@ -765,6 +765,7 @@
             "tsv",
             "epub",
             "html",
+            "mhtml",
             "ofd",
             "odt",
             "ods",

{docvortex-0.4.21 → docvortex-0.4.23}/schemas/model-2.0.json

@@ -18,6 +18,7 @@
             "tsv",
             "epub",
             "html",
+            "mhtml",
             "ofd",
             "odt",
             "ods",

{docvortex-0.4.21 → docvortex-0.4.23}/src/docvortex/analyzers/native/html/converter.py

@@ -41,6 +41,10 @@ class HtmlConverter:
             raise HtmlResourceLimitError(f"HTML resource limit exceeded: max_html_bytes={MAX_HTML_BYTES}")
         document = parse_html_document(file_bytes, source_context)
         resources = HtmlResourceContext(document.source_context, base_href=document.base_href)
+        self._convert_document(document, resources)
+
+    def _convert_document(self, document: HtmlDocument, resources: HtmlResourceContext) -> None:
+        """复用已解析 DOM 和资源适配器，供 HTML 与网页归档共享正文投影。"""
         wire_result = decode_docvortex_html_wire(document.body, resources)
         if wire_result.blocks is not None:
             blocks = wire_result.blocks

{docvortex-0.4.21 → docvortex-0.4.23}/src/docvortex/analyzers/native/html/selector.py

@@ -331,6 +331,7 @@ def _is_containment_equivalent(first: _ScoredCandidate, second: _ScoredCandidate
 
 def _soft_prune(root: etree._Element) -> None:
     """在候选副本中删除确定的导航/表单和高噪声 token 子树。"""
+    _prune_article_comments(root)
     metrics_by_element = _collect_soft_prune_metrics(root)
     for element in list(root.iterdescendants()):
         if not isinstance(element.tag, str):
@@ -349,6 +350,36 @@ def _soft_prune(root: etree._Element) -> None:
             _drop_tree_preserve_tail(element)
 
 
+def _is_comment_region(element: etree._Element) -> bool:
+    """按完整词与驼峰边界识别评论容器，不匹配正文文本或任意子串。"""
+    value = f"{element.get('id') or ''} {element.get('class') or ''}"
+    value = re.sub(r"([a-z0-9])([A-Z])", r"\1 \2", value).casefold()
+    return bool(frozenset(_TOKEN_RE.split(value)) & {"comment", "comments"})
+
+
+def _prune_article_comments(root: etree._Element) -> None:
+    """仅在单篇独立文章明确存在时移除附属评论，保留论坛及讨论型正文。"""
+    articles = [
+        element
+        for element in root.iter()
+        if isinstance(element.tag, str)
+        and (local_name(element) == "article" or "articlebody" in (element.get("itemprop") or "").casefold().split())
+        and not any(_is_comment_region(parent) for parent in (element, *element.iterancestors()))
+        and len(_normalized_text(" ".join(element.itertext()))) >= _MIN_TEXT_CHARS
+    ]
+    # 嵌套 articleBody 与 article 属于同一正文；多个独立 article 按论坛处理。
+    outer_articles = [element for element in articles if not any(parent in articles for parent in element.iterancestors())]
+    if len(outer_articles) != 1:
+        return
+    article = outer_articles[0]
+    for element in list(root.iterdescendants()):
+        if not isinstance(element.tag, str) or not _is_comment_region(element):
+            continue
+        if element is article or element in article.iterancestors():
+            continue
+        _drop_tree_preserve_tail(element)
+
+
 def _collect_soft_prune_metrics(root: etree._Element) -> dict[etree._Element, _SoftPruneMetrics]:
     """单次后序遍历预计算每个元素的完整子树文本、链接和语义对象统计。"""
     elements = [element for element in root.iter() if isinstance(element.tag, str)]

docvortex-0.4.23/src/docvortex/analyzers/native/mhtml/__init__.py

@@ -0,0 +1,3 @@
+"""MHTML 网页归档的 MIME 解包与静态正文解析。"""
+
+__all__: list[str] = []

docvortex-0.4.23/src/docvortex/analyzers/native/mhtml/converter.py

@@ -0,0 +1,73 @@
+"""把 MHTML 主文档及归档资源投影为现有 HTML 语义块。"""
+
+from __future__ import annotations
+
+from typing import BinaryIO
+
+from docvortex.document.contracts import HtmlSourceContext
+from docvortex.document.mhtml import MAX_ARCHIVE_BYTES, MhtmlArchive
+from docvortex.result import Diagnostic
+from docvortex.schema import DocumentProperties
+
+from ..html.converter import HtmlConverter
+from ..html.document import HtmlDocument, parse_html_document
+from ..html.metadata import read_html_properties
+from .resources import MhtmlResourceContext
+
+
+def read_archive_properties(archive: MhtmlArchive) -> tuple[DocumentProperties, list[str]]:
+    """复用 HTML 声明元数据，仅用 MIME Subject 补充缺失标题。"""
+    properties, warnings = read_html_properties(archive.html, archive.source_context)
+    properties.title = properties.title or archive.subject
+    return properties, warnings
+
+
+class MhtmlConverter(HtmlConverter):
+    """通过一次解包生成正文、元数据和可选资源诊断。"""
+
+    def __init__(self) -> None:
+        """初始化与单次转换绑定的元数据及诊断。"""
+        super().__init__()
+        self.properties = DocumentProperties()
+        self.diagnostics: tuple[Diagnostic, ...] = ()
+
+    def convert(self, file_binary: BinaryIO, *, source_context: HtmlSourceContext | None = None) -> None:
+        """加载主 HTML 后调用共享投影，不合并 iframe 或其他 HTML 附件。"""
+        archive = MhtmlArchive(file_binary.read(MAX_ARCHIVE_BYTES + 1), source_context)
+        self.properties, warnings = read_archive_properties(archive)
+        document = parse_html_document(archive.html, archive.source_context)
+        if document.title is None and archive.subject:
+            from dataclasses import replace
+
+            document = replace(document, title=archive.subject)
+        resources = MhtmlResourceContext(archive, base_href=document.base_href)
+        _restore_picture_sources(document, archive)
+        self._convert_document(document, resources)
+        self.diagnostics = tuple(Diagnostic("read_metadata_failed", warning) for warning in warnings) + tuple(
+            archive.diagnostics
+        )
+
+
+def _restore_picture_sources(document: HtmlDocument, archive: MhtmlArchive) -> None:
+    """当回退图片未归档时采用已保存的 picture/srcset 候选，不发起外部请求。"""
+    for image in document.body.iter("img"):
+        if (image.get("src") or "").strip().casefold().startswith("data:"):
+            continue
+        if archive.find(image.get("src") or "", base_href=document.base_href) is not None:
+            continue
+        picture = next((parent for parent in image.iterancestors() if parent.tag == "picture"), None)
+        sources = list(picture.iter("source")) if picture is not None else []
+        for source in [*sources, image]:
+            # 浏览器归档中的网络候选按逗号分隔；data URI 无需做归档匹配。
+            candidates = (source.get("srcset") or "").split(",")
+            selected = next(
+                (
+                    fields[0]
+                    for candidate in candidates
+                    if (fields := candidate.split()) and archive.find(fields[0], base_href=document.base_href) is not None
+                ),
+                None,
+            )
+            if selected:
+                image.set("src", selected)
+                break

docvortex-0.4.23/src/docvortex/analyzers/native/mhtml/resources.py

@@ -0,0 +1,80 @@
+"""使用 MIME 内嵌图片和样式的 HTML 资源适配器。"""
+
+from __future__ import annotations
+
+from docvortex.content.markup import ResolvedMarkupImage
+from docvortex.document.mhtml import ArchivePart, MhtmlArchive, MhtmlParseError
+from docvortex.foundation._svg_raster import looks_like_svg_payload, serialize_svg_image
+
+from ..html.constants import MAX_HTML_IMAGE_BYTES
+from ..html.errors import HtmlResourceLimitError
+from ..html.resources import HtmlResourceContext, _image_data_uri
+
+
+class MhtmlResourceContext(HtmlResourceContext):
+    """仅从归档读取资源，复用 HTML 的链接、图片校验和字节预算。"""
+
+    def __init__(self, archive: MhtmlArchive, *, base_href: str | None = None) -> None:
+        """绑定归档并初始化以 MIME 部件为键的去重缓存。"""
+        super().__init__(archive.source_context, base_href=base_href)
+        self.archive = archive
+        self._archive_images: dict[ArchivePart, str | None] = {}
+        self._archive_stylesheets: dict[ArchivePart, str | None] = {}
+
+    def _payload(self, part: ArchivePart, reference: str) -> bytes | None:
+        """可选部件损坏时记录诊断；资源限额错误仍向调用方传播。"""
+        try:
+            return self.archive.decode(part)
+        except MhtmlParseError:
+            self.archive.report("mhtml_resource_invalid", reference)
+            return None
+
+    def resolve_image(self, source: str, *, alt: str = "") -> ResolvedMarkupImage | None:
+        """优先恢复归档图片，缺失时沿用安全外链或替代文本降级。"""
+        if not source.strip() or source.strip().casefold().startswith("data:"):
+            return super().resolve_image(source, alt=alt)
+        part = self.archive.find(source, base_href=self.base_href)
+        if part is None:
+            self.archive.report("mhtml_resource_missing", source)
+            return super().resolve_image(source, alt=alt)
+        if part not in self._archive_images:
+            payload = self._payload(part, source)
+            data_uri = None
+            if payload is not None:
+                if len(payload) > MAX_HTML_IMAGE_BYTES:
+                    raise HtmlResourceLimitError(f"MHTML image exceeds max_html_image_bytes={MAX_HTML_IMAGE_BYTES}")
+                if looks_like_svg_payload(payload):
+                    data_uri = serialize_svg_image(payload)
+                else:
+                    data_uri = _image_data_uri(payload)
+                if data_uri is not None:
+                    self._charge_image_bytes(len(payload))
+                else:
+                    self.archive.report("mhtml_resource_invalid", source)
+            self._archive_images[part] = data_uri
+        if data_uri := self._archive_images[part]:
+            return ResolvedMarkupImage(image_base64=data_uri, alt=alt)
+        return super().resolve_image(source, alt=alt)
+
+    def load_stylesheet(self, href: str) -> str | None:
+        """按 HTML 源顺序读取归档 CSS，仅解释现有静态样式子集。"""
+        part = self.archive.find(href, base_href=self.base_href)
+        if part is None:
+            self.archive.report("mhtml_resource_missing", href)
+            return None
+        if part in self._archive_stylesheets:
+            return self._archive_stylesheets[part]
+        payload = self._payload(part, href)
+        text = None
+        if payload is not None:
+            if part.message.get_content_type() != "text/css":
+                self.archive.report("mhtml_resource_invalid", href)
+            else:
+                self._charge_stylesheet_bytes(len(payload))
+                try:
+                    text = payload.decode(part.message.get_content_charset() or "utf-8-sig", errors="replace")
+                except LookupError:
+                    self.archive.report("mhtml_resource_invalid", href)
+                    text = payload.decode("utf-8-sig", errors="replace")
+        self._archive_stylesheets[part] = text
+        return text

{docvortex-0.4.21 → docvortex-0.4.23}/src/docvortex/analyzers/native/office/docx/tables.py

@@ -185,7 +185,8 @@ class _DocxTables:
         """提取与 Mammoth 渲染规则一致的 XML 表格轻量签名。
 
         按文档顺序收集 w:t 及特殊字符元素，并且只处理 WordprocessingML
-        命名空间，以排除 Mammoth 不渲染的 OMML m:t 公式文本。
+        命名空间，以排除 Mammoth 不渲染的 OMML m:t 公式文本。纵向合并的
+        continuation 单元格会被 Mammoth 折叠掉，因此其整个子树也不参与文本签名。
         """
         w_ns = _DocxConstants._BLIP_NAMESPACES["w"]
         char_tags = (
@@ -194,7 +195,22 @@ class _DocxTables:
             f"{{{w_ns}}}softHyphen",
             f"{{{w_ns}}}sym",
         )
-        text = "".join(_DocxTables._xml_table_char_fragment(node) for node in xml_table.iter() if node.tag in char_tags)
+        continuation_nodes = set()
+        for cell in xml_table.iter(f"{{{w_ns}}}tc"):
+            properties = cell.find(f"{{{w_ns}}}tcPr")
+            vmerge = properties.find(f"{{{w_ns}}}vMerge") if properties is not None else None
+            if vmerge is None:
+                continue
+            # Mammoth 将缺少 val 或 val=continue 的单元格视为 continuation，
+            # 并在计算 rowspan 时丢弃该单元格的全部内容。
+            if vmerge.get(f"{{{w_ns}}}val") in (None, "continue"):
+                continuation_nodes.update(cell.iter())
+
+        text = "".join(
+            _DocxTables._xml_table_char_fragment(node)
+            for node in xml_table.iter()
+            if node not in continuation_nodes and node.tag in char_tags
+        )
         return {
             "row_count": len(xml_table.xpath('.//*[local-name()="tr"]')),
             "cell_count": len(xml_table.xpath('.//*[local-name()="tc"]')),

{docvortex-0.4.21 → docvortex-0.4.23}/src/docvortex/analyzers/native/office/equation/image.py

@@ -10,7 +10,7 @@ import struct
 from loguru import logger
 
 from ..errors import LegacyOfficeResourceLimitError
-from ..limits import MAX_ASSET_TOTAL_BYTES, MAX_ENTRY_BYTES, MAX_PICTURE_RECORDS
+from ..limits import MAX_ENTRY_BYTES, MAX_EQUATION_CANDIDATE_TOTAL_BYTES, MAX_PICTURE_RECORDS, MAX_RECORDS
 from .mtef import decode_mtef
 
 _PLACEABLE_WMF_MAGIC = b"\xd7\xcd\xc6\x9a"
@@ -271,14 +271,18 @@ def _wmf_mtef_candidates(image_data: bytes) -> tuple[list[bytes], bool]:
 def _gif_subblocks(
     image_data: bytes,
     cursor: int,
-    counter: list[int],
-) -> tuple[bytes, int]:
-    """读取以零长度块终止的 GIF sub-block 序列。"""
+    *,
+    collect_payload: bool,
+    subblock_counter: list[int],
+) -> tuple[bytes | None, int]:
+    """读取 GIF sub-block 序列；仅对已识别的公式扩展保留 payload。"""
 
     chunks: list[bytes] = []
     total = 0
     while True:
-        _charge_picture_record(counter)
+        subblock_counter[0] += 1
+        if subblock_counter[0] > MAX_RECORDS:
+            raise LegacyOfficeResourceLimitError(f"GIF sub-block count exceeds max_records={MAX_RECORDS}")
         if cursor >= len(image_data):
             raise _ImageEquationError("GIF sub-block length is truncated")
         size = image_data[cursor]
@@ -290,8 +294,9 @@ def _gif_subblocks(
             raise _ImageEquationError("GIF sub-block data is truncated")
         total += size
         if total > MAX_ENTRY_BYTES:
-            raise LegacyOfficeResourceLimitError(f"GIF extension exceeds max_entry_bytes={MAX_ENTRY_BYTES}")
-        chunks.append(image_data[cursor:end])
+            raise LegacyOfficeResourceLimitError(f"GIF sub-block payload exceeds max_entry_bytes={MAX_ENTRY_BYTES}")
+        if collect_payload:
+            chunks.append(image_data[cursor:end])
         cursor = end
 
 
@@ -310,6 +315,7 @@ def _gif_mtef_candidates(image_data: bytes) -> tuple[list[bytes], bool]:
     candidates: list[bytes] = []
     recognized = False
     counter = [0]
+    subblock_counter = [0]
     saw_trailer = False
     while cursor < len(image_data):
         _charge_picture_record(counter)
@@ -328,7 +334,12 @@ def _gif_mtef_candidates(image_data: bytes) -> tuple[list[bytes], bool]:
             if cursor >= len(image_data):
                 raise _ImageEquationError("GIF image data is truncated")
             cursor += 1  # LZW minimum code size
-            _image_payload, cursor = _gif_subblocks(image_data, cursor, counter)
+            _image_payload, cursor = _gif_subblocks(
+                image_data,
+                cursor,
+                collect_payload=False,
+                subblock_counter=subblock_counter,
+            )
             continue
         if marker != 0x21:
             raise _ImageEquationError("GIF block marker is invalid")
@@ -337,7 +348,12 @@ def _gif_mtef_candidates(image_data: bytes) -> tuple[list[bytes], bool]:
         extension_label = image_data[cursor]
         cursor += 1
         if extension_label != 0xFF:
-            _payload, cursor = _gif_subblocks(image_data, cursor, counter)
+            _payload, cursor = _gif_subblocks(
+                image_data,
+                cursor,
+                collect_payload=False,
+                subblock_counter=subblock_counter,
+            )
             continue
         if cursor >= len(image_data):
             raise _ImageEquationError("GIF application block size is truncated")
@@ -348,12 +364,24 @@ def _gif_mtef_candidates(image_data: bytes) -> tuple[list[bytes], bool]:
             raise _ImageEquationError("GIF application identifier is truncated")
         application = image_data[cursor:application_end]
         cursor = application_end
-        payload, cursor = _gif_subblocks(image_data, cursor, counter)
         if application_size != 11:
+            _payload, cursor = _gif_subblocks(
+                image_data,
+                cursor,
+                collect_payload=False,
+                subblock_counter=subblock_counter,
+            )
             continue
         application_id = application[:8]
         authentication = application[8:11]
-        if application_id == b"MathType" and authentication == b"001":
+        is_mathtype = application_id == b"MathType" and authentication == b"001"
+        payload, cursor = _gif_subblocks(
+            image_data,
+            cursor,
+            collect_payload=is_mathtype,
+            subblock_counter=subblock_counter,
+        )
+        if is_mathtype and payload is not None:
             recognized = True
             candidates.append(payload)
         # MathType/002 是 baseline，必须明确忽略。
@@ -406,6 +434,12 @@ class OfficeImageEquationDecoder:
     _cache: dict[tuple[str, bytes], str | None] = field(default_factory=dict)
     _warned: set[tuple[str, bytes]] = field(default_factory=set)
 
+    @property
+    def candidate_total_bytes(self) -> int:
+        """返回已计入预算的公式 candidate 总字节数。"""
+
+        return self.total_bytes
+
     def decode(
         self,
         image_data: object | None,
@@ -421,23 +455,23 @@ class OfficeImageEquationDecoder:
         if image_format is None:
             return None
         if len(image_data) > MAX_ENTRY_BYTES:
-            raise LegacyOfficeResourceLimitError(f"image equation payload exceeds max_entry_bytes={MAX_ENTRY_BYTES}")
+            raise LegacyOfficeResourceLimitError(f"office image exceeds max_entry_bytes={MAX_ENTRY_BYTES}")
         digest = hashlib.sha256(image_data).digest()
         cache_key = (image_format, digest)
         if cache_key in self._cache:
             return self._cache[cache_key]
-        if self.total_bytes + len(image_data) > MAX_ASSET_TOTAL_BYTES:
-            raise LegacyOfficeResourceLimitError(
-                f"image equation payloads exceed max_asset_total_bytes={MAX_ASSET_TOTAL_BYTES}"
-            )
-        self.total_bytes += len(image_data)
-
         recognized = False
         try:
             if image_format == "wmf":
                 candidates, recognized = _wmf_mtef_candidates(image_data)
             else:
                 candidates, recognized = _gif_mtef_candidates(image_data)
+            candidate_bytes = sum(len(candidate) for candidate in candidates)
+            if self.total_bytes + candidate_bytes > MAX_EQUATION_CANDIDATE_TOTAL_BYTES:
+                raise LegacyOfficeResourceLimitError(
+                    f"image equation candidates exceed max_equation_candidate_total_bytes={MAX_EQUATION_CANDIDATE_TOTAL_BYTES}"
+                )
+            self.total_bytes += candidate_bytes
             latex = _select_candidate_latex(candidates)
         except LegacyOfficeResourceLimitError:
             raise

{docvortex-0.4.21 → docvortex-0.4.23}/src/docvortex/analyzers/native/office/limits.py

@@ -5,6 +5,8 @@ from typing import Final
 MAX_ENTRY_BYTES: Final = 128 * 1024 * 1024
 MAX_TOTAL_BYTES: Final = 512 * 1024 * 1024
 MAX_ASSET_TOTAL_BYTES: Final = 128 * 1024 * 1024
+# 公式探测单独计费，避免普通图片耗尽嵌入素材预算；初始值与素材预算保持一致。
+MAX_EQUATION_CANDIDATE_TOTAL_BYTES: Final = 128 * 1024 * 1024
 MAX_GRID_SLOTS: Final = 4_000_000
 MAX_RECORD_DEPTH: Final = 64
 MAX_RECORDS: Final = 16_000_000

{docvortex-0.4.21 → docvortex-0.4.23}/src/docvortex/analyzers/native/office/metadata.py

@@ -5,6 +5,7 @@ from __future__ import annotations
 from io import BytesIO
 from datetime import datetime, timezone
 import posixpath
+from pathlib import Path
 from zipfile import ZipFile
 
 import lxml.etree as etree
@@ -12,7 +13,7 @@ import lxml.etree as etree
 from ....document.properties import property_count, property_date, property_text, property_values
 from ....errors import DocumentError
 from ....schema import DocumentProperties
-from .limits import MAX_ENTRY_BYTES, MAX_TOTAL_BYTES
+from .limits import MAX_ENTRY_BYTES
 
 _DC = "{http://purl.org/dc/elements/1.1/}"
 _DCT = "{http://purl.org/dc/terms/}"
@@ -43,14 +44,15 @@ def _element_value(root: etree._Element | None, tag: str) -> str | None:
     return None
 
 
-def read_ooxml_properties(data: bytes, suffix: str) -> tuple[DocumentProperties, list[str]]:
+def read_ooxml_properties(data: bytes | str | Path, suffix: str) -> tuple[DocumentProperties, list[str]]:
     """读取 OOXML 核心属性和结构计数，坏的可选属性不阻止其他读取。"""
     properties = DocumentProperties()
     warnings: list[str] = []
-    with ZipFile(BytesIO(data)) as package:
+    package_source = BytesIO(data) if isinstance(data, bytes) else data
+    with ZipFile(package_source) as package:
         infos = package.infolist()
-        if len(infos) > 100_000 or sum(info.file_size for info in infos) > MAX_TOTAL_BYTES:
-            raise DocumentError("resource_limit", "OOXML metadata package exceeds resource limits")
+        if len(infos) > 100_000:
+            raise DocumentError("resource_limit", "OOXML metadata package has too many members")
         if len({info.filename for info in infos}) != len(infos):
             raise DocumentError("open_failed", "Duplicate OOXML package members")
         if _xml_part(package, "[Content_Types].xml") is None:

{docvortex-0.4.21 → docvortex-0.4.23}/src/docvortex/analyzers/native/pdf/models.py

@@ -2,6 +2,7 @@
 
 from __future__ import annotations
 
+from collections.abc import Iterable, Iterator, MutableSet
 from dataclasses import dataclass, field
 from typing import TYPE_CHECKING, Any, Literal
 
@@ -13,6 +14,102 @@ if TYPE_CHECKING:
     from .inline.types import PDFTextScriptLine
 
 
+class _SharedLineIndexSet(MutableSet[int]):
+    """共享只读基底，仅为单个候选记录增删差集，避免复制大型行号集合。"""
+
+    __slots__ = ("base", "added", "removed")
+
+    def __init__(
+        self,
+        base: frozenset[int],
+        values: Iterable[int] = (),
+    ) -> None:
+        """以共享基底和候选已有成员初始化集合。"""
+
+        self.base = base
+        self.removed: set[int] = set()
+        if isinstance(values, _SharedLineIndexSet) and values.base is base:
+            # 与共享基底求并集时，基底成员天然全部可见，只需保留额外成员。
+            self.added = set(values.added)
+        elif isinstance(values, (set, frozenset)):
+            self.added = values.difference(base)
+        else:
+            self.added = {value for value in values if value not in base}
+
+    @classmethod
+    def from_exact_values(
+        cls,
+        base: frozenset[int],
+        values: Iterable[int],
+    ) -> _SharedLineIndexSet:
+        """以共享基底精确表示给定集合，而不是默认并入全部基底成员。"""
+
+        visible = set(values)
+        instance = cls.__new__(cls)
+        instance.base = base
+        instance.added = visible.difference(base)
+        instance.removed = set(base.difference(visible))
+        return instance
+
+    def __contains__(self, value: object) -> bool:
+        """按基底、删除集和新增集判断成员是否可见。"""
+
+        if value in self.added:
+            return True
+        return value in self.base and value not in self.removed
+
+    def __iter__(self) -> Iterator[int]:
+        """迭代当前可见成员，不展开持久化副本。"""
+
+        for value in self.base:
+            if value not in self.removed:
+                yield value
+        yield from self.added
+
+    def __len__(self) -> int:
+        """返回当前可见成员数量。"""
+
+        return len(self.base) - len(self.removed) + len(self.added)
+
+    def add(self, value: int) -> None:
+        """加入成员；基底已有成员只需撤销删除标记。"""
+
+        if value in self.base:
+            self.removed.discard(value)
+            return
+        self.added.add(value)
+
+    def discard(self, value: int) -> None:
+        """删除成员；基底成员以差集标记表示。"""
+
+        if value in self.added:
+            self.added.discard(value)
+            return
+        if value in self.base:
+            self.removed.add(value)
+
+    def update(self, values: Iterable[int]) -> None:
+        """并入成员；同共享基底的候选直接合并差集。"""
+
+        if isinstance(values, _SharedLineIndexSet) and values.base is self.base:
+            self.removed.intersection_update(values.removed)
+            for value in values.added:
+                self.add(value)
+            return
+        for value in values:
+            self.add(value)
+
+    def difference_update(self, values: Iterable[int]) -> None:
+        """移除给定成员，并保持共享基底不变。"""
+
+        if isinstance(values, (set, frozenset)):
+            self.added.difference_update(values)
+            self.removed.update(self.base.intersection(values))
+            return
+        for value in values:
+            self.discard(value)
+
+
 @dataclass(slots=True)
 class _LineItem:
     """保存单个可视文本行及其 source/ink/canonical 几何。"""
@@ -122,7 +219,7 @@ class _TableCandidate:
     angle: int
     score: float
     core_bbox: BBox | None = None
-    line_indices: set[int] = field(default_factory=set)
+    line_indices: MutableSet[int] = field(default_factory=set)
     annotations: list[_TableAnnotation] = field(default_factory=list)