docvortex 0.4.9__tar.gz → 0.4.11__tar.gz

{docvortex-0.4.9/src/docvortex.egg-info → docvortex-0.4.11}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: docvortex
-Version: 0.4.9
+Version: 0.4.11
 Summary: A fast, multi-format document parsing and conversion engine
 License-Expression: MIT
 Project-URL: Repository, https://github.com/myhloli/DocVortex
@@ -82,8 +82,7 @@ then exports the result in the formats your workflow needs.
 - **Portable results** — save document structure and image assets in a Bundle, then export again without the source file.
 - **Composable APIs** — use the complete pipeline or integrate analysis, postprocessing and rendering separately.
 
-Native parsing works without an OCR or VLM inference service. Use DocVortex
-directly through its CLI or Python SDK, independently of MinerU.
+Native parsing works without an OCR or VLM inference service. Use DocVortex directly through its CLI or Python SDK.
 
 ![DocVortex pipeline: native documents become a unified representation, then Markdown, HTML, LaTeX, DOCX, EPUB, PDF or structured JSON.](https://gcore.jsdelivr.net/gh/myhloli/DocVortex@main/docs/images/docvortex-overview.jpg)
 

{docvortex-0.4.9 → docvortex-0.4.11}/README.md

@@ -28,8 +28,7 @@ then exports the result in the formats your workflow needs.
 - **Portable results** — save document structure and image assets in a Bundle, then export again without the source file.
 - **Composable APIs** — use the complete pipeline or integrate analysis, postprocessing and rendering separately.
 
-Native parsing works without an OCR or VLM inference service. Use DocVortex
-directly through its CLI or Python SDK, independently of MinerU.
+Native parsing works without an OCR or VLM inference service. Use DocVortex directly through its CLI or Python SDK.
 
 ![DocVortex pipeline: native documents become a unified representation, then Markdown, HTML, LaTeX, DOCX, EPUB, PDF or structured JSON.](https://gcore.jsdelivr.net/gh/myhloli/DocVortex@main/docs/images/docvortex-overview.jpg)
 

{docvortex-0.4.9 → docvortex-0.4.11}/src/docvortex/analyzers/native/epub/xhtml.py

@@ -35,6 +35,7 @@ from docvortex.content.markup.projector import (
 
 from ....content.spans import normalize_span_dicts
 from ....foundation._image_payload import parse_image_data_uri_strict
+from ....foundation._svg_raster import serialize_svg_image
 from .._shared.hyperlink import sanitize_hyperlink_target
 from .constants import IMAGE_MEDIA_BY_EXTENSION, SVG_MEDIA_TYPE
 from .package import EpubPackage
@@ -275,14 +276,22 @@ class _EpubMarkupContext:
         )
 
     def resolve_image(self, source: str, *, alt: str = "") -> ResolvedMarkupImage | None:
-        """读取并严格校验一个 EPUB 包内栅格图片引用。"""
+        """读取并严格校验一个 EPUB 包内图片引用，SVG 光栅化为 PNG。"""
         target = self.package.resolve_reference(source, base_part=self.chapter_path)
         if target is None:
             return ResolvedMarkupImage(alt=alt) if alt else None
         media_type = (self.package.content_type_for(target.path) or "").casefold()
         extension = target.path.rsplit(".", 1)[-1].casefold() if "." in target.path else ""
         media_type = media_type or IMAGE_MEDIA_BY_EXTENSION.get(extension, "")
-        if not media_type.startswith("image/") or media_type == SVG_MEDIA_TYPE:
+        if media_type == SVG_MEDIA_TYPE:
+            payload = self.package.read_part(target.path, asset=True)
+            if payload is None:
+                return ResolvedMarkupImage(alt=alt) if alt else None
+            data_uri = serialize_svg_image(payload)
+            if data_uri is None:
+                return ResolvedMarkupImage(alt=alt) if alt else None
+            return ResolvedMarkupImage(image_base64=data_uri, alt=alt)
+        if not media_type.startswith("image/"):
             return ResolvedMarkupImage(alt=alt) if alt else None
         payload = self.package.read_part(target.path, asset=True)
         if payload is None:

{docvortex-0.4.9 → docvortex-0.4.11}/src/docvortex/analyzers/native/html/constants.py

@@ -8,6 +8,7 @@ MAX_HTML_NODES: Final = 2_000_000
 MAX_HTML_DEPTH: Final = 256
 MAX_HTML_IMAGE_BYTES: Final = 20 * 1024 * 1024
 MAX_HTML_IMAGE_TOTAL_BYTES: Final = 128 * 1024 * 1024
+MAX_REMOTE_IMAGE_FETCH_SECONDS: Final = 10
 MAX_HTML_STYLESHEET_BYTES: Final = 5 * 1024 * 1024
 MAX_HTML_STYLESHEET_TOTAL_BYTES: Final = 20 * 1024 * 1024
 MAX_HTML_RENDERED_BYTES: Final = 256 * 1024 * 1024
@@ -22,4 +23,5 @@ __all__ = [
     "MAX_HTML_RENDERED_BYTES",
     "MAX_HTML_STYLESHEET_BYTES",
     "MAX_HTML_STYLESHEET_TOTAL_BYTES",
+    "MAX_REMOTE_IMAGE_FETCH_SECONDS",
 ]

{docvortex-0.4.9 → docvortex-0.4.11}/src/docvortex/analyzers/native/html/resources.py

@@ -6,6 +6,7 @@ import base64
 from pathlib import Path
 from typing import Protocol
 from urllib.parse import SplitResult, unquote, urljoin, urlsplit, urlunsplit
+from urllib.request import urlopen
 
 from lxml import etree  # type: ignore[reportMissingImports]
 
@@ -13,12 +14,14 @@ from docvortex.content.markup import ResolvedMarkupImage
 from docvortex.document.contracts import HtmlSourceContext
 
 from ....foundation._image_payload import parse_image_data_uri_strict, validate_remote_image_url
+from ....foundation._svg_raster import looks_like_svg_payload, serialize_svg_image
 from .._shared.hyperlink import sanitize_hyperlink_target
 from .constants import (
     MAX_HTML_IMAGE_BYTES,
     MAX_HTML_IMAGE_TOTAL_BYTES,
     MAX_HTML_STYLESHEET_BYTES,
     MAX_HTML_STYLESHEET_TOTAL_BYTES,
+    MAX_REMOTE_IMAGE_FETCH_SECONDS,
 )
 from .errors import HtmlResourceLimitError
 
@@ -65,6 +68,7 @@ class HtmlResourceContext:
         self._image_bytes = 0
         self._stylesheet_bytes = 0
         self._image_cache: dict[Path, str] = {}
+        self._remote_image_cache: dict[str, str] = {}
         self._data_image_cache: dict[str, ResolvedMarkupImage] = {}
         self._stylesheet_cache: dict[Path, str] = {}
         self._local_root = source_context.local_resource_root.resolve() if source_context.local_resource_root else None
@@ -157,14 +161,14 @@ class HtmlResourceContext:
         return sanitize_hyperlink_target(normalized, allow_relative=True, allow_fragment=True)
 
     def resolve_image(self, source: str, *, alt: str = "") -> ResolvedMarkupImage | None:
-        """按 data URI、本地安全文件、远程 URL 的顺序解析图片。"""
+        """按 data URI、远程 URL、本地安全文件的顺序解析图片，远程图片仅在来源显式开启下载时内嵌。"""
         normalized = (source or "").strip()
         if not normalized:
             return ResolvedMarkupImage(alt=alt) if alt else None
         if normalized.casefold().startswith("data:"):
             return self._resolve_data_image(normalized, alt=alt)
         if remote_url := self._resolve_remote_image_url(normalized):
-            return ResolvedMarkupImage(image_url=remote_url, alt=alt)
+            return self._resolve_remote_image(remote_url, alt=alt)
         local_path = self._resolve_local_path(normalized)
         if local_path is None or not local_path.is_file():
             return ResolvedMarkupImage(alt=alt) if alt else None
@@ -175,7 +179,10 @@ class HtmlResourceContext:
             payload = source_file.read(MAX_HTML_IMAGE_BYTES + 1)
         if len(payload) > MAX_HTML_IMAGE_BYTES:
             raise HtmlResourceLimitError(f"HTML image exceeds max_html_image_bytes={MAX_HTML_IMAGE_BYTES}")
-        data_uri = _image_data_uri(payload)
+        if local_path.suffix.casefold() == ".svg" or looks_like_svg_payload(payload):
+            data_uri = serialize_svg_image(payload)
+        else:
+            data_uri = _image_data_uri(payload)
         if data_uri is None:
             return ResolvedMarkupImage(alt=alt) if alt else None
         self._charge_image_bytes(len(payload))
@@ -225,7 +232,7 @@ class HtmlResourceContext:
         return self.anchors.note_anchor(note) if self.anchors else None
 
     def _resolve_data_image(self, data_uri: str, *, alt: str) -> ResolvedMarkupImage | None:
-        """严格解析 data URI，并执行单图与累计图片预算。"""
+        """严格解析 data URI 并执行图片预算，SVG 光栅化为 PNG 后内嵌。"""
         if cached := self._data_image_cache.get(data_uri):
             return ResolvedMarkupImage(image_base64=cached.image_base64, alt=alt)
         if len(data_uri) > MAX_HTML_IMAGE_BYTES * 2:
@@ -235,7 +242,15 @@ class HtmlResourceContext:
         except ValueError:
             return ResolvedMarkupImage(alt=alt) if alt else None
         if extension == "svg":
-            return ResolvedMarkupImage(alt=alt) if alt else None
+            if len(payload) > MAX_HTML_IMAGE_BYTES:
+                raise HtmlResourceLimitError(f"HTML image exceeds max_html_image_bytes={MAX_HTML_IMAGE_BYTES}")
+            raster_uri = serialize_svg_image(payload)
+            if raster_uri is None:
+                return ResolvedMarkupImage(alt=alt) if alt else None
+            self._charge_image_bytes(len(payload))
+            resolved = ResolvedMarkupImage(image_base64=raster_uri, alt=alt)
+            self._data_image_cache[data_uri] = resolved
+            return resolved
         if len(payload) > MAX_HTML_IMAGE_BYTES:
             raise HtmlResourceLimitError(f"HTML image exceeds max_html_image_bytes={MAX_HTML_IMAGE_BYTES}")
         self._charge_image_bytes(len(payload))
@@ -243,6 +258,41 @@ class HtmlResourceContext:
         self._data_image_cache[data_uri] = resolved
         return resolved
 
+    def _resolve_remote_image(self, remote_url: str, *, alt: str) -> ResolvedMarkupImage:
+        """远程图片默认保持外链，来源显式开启下载时按缓存下载内嵌。"""
+        if not self.source_context.fetch_remote_images:
+            return ResolvedMarkupImage(image_url=remote_url, alt=alt)
+        if cached := self._remote_image_cache.get(remote_url):
+            return ResolvedMarkupImage(image_base64=cached, alt=alt)
+        data_uri = self._fetch_remote_image(remote_url)
+        if data_uri is None:
+            return ResolvedMarkupImage(image_url=remote_url, alt=alt)
+        self._remote_image_cache[remote_url] = data_uri
+        return ResolvedMarkupImage(image_base64=data_uri, alt=alt)
+
+    def _fetch_remote_image(self, remote_url: str) -> str | None:
+        """在单图上限与超时约束内下载远程图片并复核签名，网络失败返回 None 退回外链。"""
+        try:
+            with urlopen(remote_url, timeout=MAX_REMOTE_IMAGE_FETCH_SECONDS) as response:
+                chunks: list[bytes] = []
+                remaining = MAX_HTML_IMAGE_BYTES + 1
+                while remaining > 0:
+                    chunk = response.read(remaining)
+                    if not chunk:
+                        break
+                    chunks.append(chunk)
+                    remaining -= len(chunk)
+        except (OSError, ValueError):
+            return None
+        payload = b"".join(chunks)
+        if len(payload) > MAX_HTML_IMAGE_BYTES:
+            raise HtmlResourceLimitError(f"HTML image exceeds max_html_image_bytes={MAX_HTML_IMAGE_BYTES}")
+        data_uri = _image_data_uri(payload)
+        if data_uri is None:
+            return None
+        self._charge_image_bytes(len(payload))
+        return data_uri
+
     def _charge_image_bytes(self, byte_count: int) -> None:
         """累计实际保留的图片字节，并在超限时终止整份文档。"""
         self._image_bytes += byte_count

{docvortex-0.4.9 → docvortex-0.4.11}/src/docvortex/analyzers/native/office/image.py

@@ -11,6 +11,7 @@ from loguru import logger
 from metafile_render import MetafileError, MetafileResourceLimitError, render_metafile
 from PIL import Image, ImageDraw, ImageFont, UnidentifiedImageError
 
+from ....foundation._svg_raster import is_svg_image_part, looks_like_svg_payload, serialize_svg_image
 from ....foundation.image_encoding import image_to_b64str
 
 VECTOR_IMAGE_FORMATS = frozenset({"WMF", "EMF"})
@@ -267,6 +268,9 @@ def serialize_office_image(
             return rendered
         return serialize_vector_part_with_placeholder(part_name, content_type)
 
+    if is_svg_image_part(part_name, content_type) or looks_like_svg_payload(image_data):
+        return serialize_svg_image(image_data)
+
     try:
         pil_image = Image.open(BytesIO(image_data))
         if is_vector_image(pil_image):

{docvortex-0.4.9 → docvortex-0.4.11}/src/docvortex/analyzers/native/office/odf/styles.py

@@ -260,11 +260,15 @@ class OdfStyles:
             current = definition.parent or ""
         return False
 
-    def list_level(self, style_name: str | None, depth: int) -> ListLevel:
-        """返回指定列表深度的定义，不存在时使用无序默认值。"""
+    def list_level(self, style_name: str | None, depth: int) -> ListLevel | None:
+        """返回指定列表深度的定义；样式已定义但无任何层级时返回 None（无可见标记）。"""
         if not style_name:
             return ListLevel()
-        levels = self._list_styles.get(style_name, {})
+        levels = self._list_styles.get(style_name)
+        if levels is None:
+            return ListLevel()
+        if not levels:
+            return None
         return levels.get(depth, ListLevel())
 
     def table_is_visible(self, style_name: str | None) -> bool:

{docvortex-0.4.9 → docvortex-0.4.11}/src/docvortex/analyzers/native/office/odf/table.py

@@ -169,7 +169,6 @@ def parse_table_grid(
     duplicated_text_bytes = 0
     row_index = 0
     pending_empty_rows = 0
-    pending_empty_width = 0
 
     def ensure_row(target_row: int, width: int = 0) -> list[GridCell | None]:
         """在兼容独立表格调用的同时应用可选文档级预留。"""
@@ -179,7 +178,6 @@ def parse_table_grid(
 
     for row_element, header in _iter_rows(table):
         row_repeat = _positive_int(row_element.get(qname("table", "number-rows-repeated")))
-        _validate_grid_extent(row_index + row_repeat, grid.width)
         cell_templates: list[tuple[bool, int, GridCell | None]] = []
         for cell in row_element:
             if not isinstance(cell.tag, str):
@@ -219,18 +217,14 @@ def parse_table_grid(
         )
         if not row_has_content:
             pending_empty_rows += row_repeat
-            pending_empty_width = max(
-                pending_empty_width,
-                sum(repeat * (template.col_span if template is not None else 1) for _, repeat, template in cell_templates),
-            )
             continue
         if pending_empty_rows:
-            _validate_grid_extent(row_index + pending_empty_rows, max(grid.width, pending_empty_width))
+            _validate_grid_extent(row_index + pending_empty_rows, grid.width)
             for _ in range(pending_empty_rows):
-                ensure_row(row_index, pending_empty_width)
+                ensure_row(row_index, grid.width)
                 row_index += 1
             pending_empty_rows = 0
-            pending_empty_width = 0
+        _validate_grid_extent(row_index + row_repeat, grid.width)
         for _ in range(row_repeat):
             row = ensure_row(row_index)
             col_index = 0

{docvortex-0.4.9 → docvortex-0.4.11}/src/docvortex/analyzers/native/office/odf/text.py

@@ -20,6 +20,7 @@ from .....content.spans import (
     inline_span_plain_text,
     strip_span_dicts,
 )
+from .....foundation._svg_raster import serialize_svg_image
 from .....foundation.image_encoding import image_to_b64str
 from .....schema import BlockType
 from ..._shared.hyperlink import sanitize_hyperlink_target
@@ -108,16 +109,36 @@ def _style_html(text: str, style: TextStyle) -> str:
     return rendered
 
 
+_SVG_RENDER_DPI = 200
+
+
+def _frame_render_size(frame: etree._Element | None) -> tuple[int, int] | None:
+    """按 draw:frame 显示尺寸换算渲染像素，作为 SVG 光栅化的目标尺寸。"""
+    from .converters import _length_to_points
+
+    if frame is None:
+        return None
+    width = _length_to_points(frame.get(qname("svg", "width")))
+    height = _length_to_points(frame.get(qname("svg", "height")))
+    if width <= 0 or height <= 0:
+        return None
+    return (max(1, round(width / 72 * _SVG_RENDER_DPI)), max(1, round(height / 72 * _SVG_RENDER_DPI)))
+
+
 def _serialize_odf_image(
     image_bytes: bytes,
     *,
     part_name: str | None,
     content_type: str | None,
+    size_hint: tuple[int, int] | None = None,
 ) -> str | None:
-    """序列化 ODF 图片；SVG、SVM 和 GDIMeta 使用安全占位图保留对象位置。"""
+    """序列化 ODF 图片；SVG 光栅化为 PNG，SVM 和 GDIMeta 使用安全占位图保留对象位置。"""
     normalized_type = (content_type or "").split(";", 1)[0].strip().casefold()
     suffix = (part_name or "").rsplit(".", 1)[-1].casefold() if "." in (part_name or "") else ""
     if normalized_type == "image/svg+xml" or suffix == "svg":
+        rendered = serialize_svg_image(image_bytes, size_hint=size_hint)
+        if rendered is not None:
+            return rendered
         placeholder = create_text_placeholder((320, 180), ["SVG image", "Preview unavailable"])
         return image_to_b64str(placeholder, image_format="JPEG")
     if suffix == "svm" or "gdimetafile" in normalized_type or image_bytes.startswith(b"VCLMTF"):
@@ -453,7 +474,7 @@ class OdfBlockParser:
         style_name = element.get(qname("text", "style-name")) or inherited_style
         level = self.styles.list_level(style_name, depth)
         key = (style_name or "", depth)
-        start = level.start
+        start = level.start if level is not None else 1
         continue_list = element.get(qname("text", "continue-list"))
         if continue_list and continue_list in self._list_ids:
             start = self._list_ids[continue_list]
@@ -466,17 +487,20 @@ class OdfBlockParser:
         active_master: str | None = None
 
         def flush_content(fragment_start: int) -> None:
-            """把当前合法子块冻结为一个 LIST 分片。"""
+            """把当前合法子块冻结为一个 LIST 分片；无标记列表按普通段落序列输出。"""
             if content:
-                block: dict[str, Any] = {
-                    "type": BlockType.LIST,
-                    "attribute": "ordered" if level.ordered else "unordered",
-                    "ilevel": depth,
-                    "content": list(content),
-                }
-                if level.ordered:
-                    block["start"] = fragment_start
-                results.append(block)
+                if level is None:
+                    results.extend(content)
+                else:
+                    block: dict[str, Any] = {
+                        "type": BlockType.LIST,
+                        "attribute": "ordered" if level.ordered else "unordered",
+                        "ilevel": depth,
+                        "content": list(content),
+                    }
+                    if level.ordered:
+                        block["start"] = fragment_start
+                    results.append(block)
                 content.clear()
             if fragment_notes:
                 results.extend(fragment_notes)
@@ -709,7 +733,15 @@ class OdfBlockParser:
             )
         if not image_bytes:
             return None, alt
-        return _serialize_odf_image(image_bytes, part_name=part_name, content_type=content_type), alt
+        return (
+            _serialize_odf_image(
+                image_bytes,
+                part_name=part_name,
+                content_type=content_type,
+                size_hint=_frame_render_size(parent),
+            ),
+            alt,
+        )
 
     def _object_root(self, object_element: etree._Element) -> tuple[etree._Element | None, str | None]:
         """读取 draw:object 指向的子文档内容树和成员路径。"""
@@ -875,6 +907,20 @@ class OdfBlockParser:
         """把单元格行内流中的 note marker 排入当前逻辑页队列。"""
         self.notes.extend(atom.content for atom in atoms if isinstance(atom, InlineNote))
 
+    def _render_cell_paragraph(self, child: etree._Element, parts: list[str]) -> None:
+        """把单元格里的一个裸段落渲染为 <p>，并回收行内块组的单元格子块。"""
+        atoms = self.parse_inline_atoms(child)
+        self._queue_inline_notes(atoms)
+        rendered_atoms = [atom for atom in atoms if not isinstance(atom, InlineImage)] if self._collect_cell_visuals else atoms
+        parts.append(f"<p>{render_atoms_to_html(rendered_atoms)}</p>")
+        for atom in atoms:
+            if isinstance(atom, InlineBlockGroup):
+                self._append_cell_blocks(
+                    parts,
+                    atom.blocks,
+                    inline_image_rendered=atom.inline_image_rendered and not self._collect_cell_visuals,
+                )
+
     def render_cell_html(self, cell: etree._Element) -> str:
         """把表格单元格中的段落、列表、嵌套表和 frame 转为 HTML。"""
         parts: list[str] = []
@@ -882,19 +928,7 @@ class OdfBlockParser:
             if not isinstance(child.tag, str):
                 continue
             if child.tag in {qname("text", "p"), qname("text", "h")}:
-                atoms = self.parse_inline_atoms(child)
-                self._queue_inline_notes(atoms)
-                rendered_atoms = (
-                    [atom for atom in atoms if not isinstance(atom, InlineImage)] if self._collect_cell_visuals else atoms
-                )
-                parts.append(f"<p>{render_atoms_to_html(rendered_atoms)}</p>")
-                for atom in atoms:
-                    if isinstance(atom, InlineBlockGroup):
-                        self._append_cell_blocks(
-                            parts,
-                            atom.blocks,
-                            inline_image_rendered=atom.inline_image_rendered and not self._collect_cell_visuals,
-                        )
+                self._render_cell_paragraph(child, parts)
             elif child.tag == qname("text", "list"):
                 parts.append(self._render_list_html(child))
             elif child.tag == qname("table", "table"):
@@ -915,6 +949,8 @@ class OdfBlockParser:
         """把单元格内 ODF 列表递归渲染为 ol/ul HTML。"""
         style_name = element.get(qname("text", "style-name")) or inherited_style
         level = self.styles.list_level(style_name, depth)
+        if level is None:
+            return self._render_unmarked_list_html(element, depth=depth, inherited_style=style_name)
         tag = "ol" if level.ordered else "ul"
         start = f' start="{level.start}"' if level.ordered and level.start != 1 else ""
         parts = [f"<{tag}{start}>"]
@@ -940,6 +976,30 @@ class OdfBlockParser:
         parts.append(f"</{tag}>")
         return "".join(parts)
 
+    def _render_unmarked_list_html(
+        self,
+        element: etree._Element,
+        *,
+        depth: int,
+        inherited_style: str | None,
+    ) -> str:
+        """把引用空列表样式（无可见标记）的单元格列表按裸段落序列渲染。"""
+        parts: list[str] = []
+        for item in element:
+            if not isinstance(item.tag, str) or item.tag not in {
+                qname("text", "list-item"),
+                qname("text", "list-header"),
+            }:
+                continue
+            for child in item:
+                if not isinstance(child.tag, str):
+                    continue
+                if child.tag in {qname("text", "p"), qname("text", "h")}:
+                    self._render_cell_paragraph(child, parts)
+                elif child.tag == qname("text", "list"):
+                    parts.append(self._render_list_html(child, depth=depth + 1, inherited_style=inherited_style))
+        return "".join(part for part in parts if part)
+
     def drain_notes(self) -> list[str]:
         """取出当前累计脚注并清空共享队列。"""
         values = list(self.notes)

{docvortex-0.4.9 → docvortex-0.4.11}/src/docvortex/analyzers/native/office/pptx/resources.py

@@ -1,6 +1,5 @@
 """PPTX 图片、图表与公式资源，复用当前转换器的单文档状态。"""
 
-import base64
 from typing import Any, Optional
 from loguru import logger
 from pptx.enum.shapes import MSO_SHAPE_TYPE
@@ -222,11 +221,19 @@ class _PptxResources:
             return
 
         if content_type == "image/svg+xml":
-            image_block = {
-                "type": BlockType.IMAGE,
-                "image_base64": self._bytes_to_data_uri(image_bytes, content_type),
-            }
-            self.cur_page.append(image_block)
+            img_base64 = serialize_office_image(image_bytes, content_type=content_type)
+            if img_base64 is None:
+                # SVG 光栅化失败时回退到 PowerPoint 为 SVG 原生伴随存储的栅量 blip。
+                fallback_data = self._get_shape_image_data(shape, include_svg=False)
+                if fallback_data is not None and fallback_data[1] != "image/svg+xml":
+                    img_base64 = serialize_office_image(fallback_data[0], content_type=fallback_data[1])
+            if img_base64 is not None:
+                self.cur_page.append(
+                    {
+                        "type": BlockType.IMAGE,
+                        "image_base64": img_base64,
+                    }
+                )
             return
 
         img_base64 = serialize_office_image(
@@ -256,19 +263,14 @@ class _PptxResources:
         )
 
     @staticmethod
-    def _bytes_to_data_uri(image_bytes: bytes, content_type: str) -> str:
-        """按原有图片、图表与公式资源规则执行 _bytes_to_data_uri，保持输入顺序与降级行为。"""
-        encoded = base64.b64encode(image_bytes).decode("utf-8")
-        return f"data:{content_type};base64,{encoded}"
-
-    @staticmethod
-    def _find_first_embedded_image_rid(shape) -> Optional[str]:
+    def _find_first_embedded_image_rid(shape, *, include_svg: bool = True) -> Optional[str]:
         """按原有图片、图表与公式资源规则执行 _find_first_embedded_image_rid，保持输入顺序与降级行为。"""
-        svg_blips = shape._element.findall(f".//{{{SVG_BLIP_NS}}}svgBlip")
-        for svg_blip in svg_blips:
-            relationship_id = svg_blip.get(f"{{{RELATIONSHIP_NS}}}embed")
-            if relationship_id:
-                return relationship_id
+        if include_svg:
+            svg_blips = shape._element.findall(f".//{{{SVG_BLIP_NS}}}svgBlip")
+            for svg_blip in svg_blips:
+                relationship_id = svg_blip.get(f"{{{RELATIONSHIP_NS}}}embed")
+                if relationship_id:
+                    return relationship_id
 
         blips = shape._element.findall(f".//{{{DRAWINGML_NS}}}blip")
         for blip in blips:
@@ -296,11 +298,11 @@ class _PptxResources:
                 return False
         return True
 
-    def _get_shape_image_data(self, shape) -> Optional[tuple[bytes, Optional[str]]]:
+    def _get_shape_image_data(self, shape, *, include_svg: bool = True) -> Optional[tuple[bytes, Optional[str]]]:
         """按原有图片、图表与公式资源规则执行 _get_shape_image_data，保持输入顺序与降级行为。"""
         relationship_id = None
         if hasattr(shape, "_element"):
-            relationship_id = self._find_first_embedded_image_rid(shape)
+            relationship_id = self._find_first_embedded_image_rid(shape, include_svg=include_svg)
 
         if relationship_id:
             try:

{docvortex-0.4.9 → docvortex-0.4.11}/src/docvortex/content/mathml.py

@@ -6,6 +6,7 @@ import re
 
 from lxml import etree  # type: ignore[reportMissingImports]
 
+from ..foundation._text import normalize_formula_tag_content
 from ..foundation.xml_names import local_name
 
 _OPERATOR_MAP = {
@@ -71,6 +72,8 @@ _LATEX_MATH_TOKEN_ESCAPES = {
     "}": r"\}",
     "~": r"\~{}",
 }
+# Word 线性格式把公式编号分隔符写作 ``#(n)``；经 mo 转义后行尾呈现为 ``\#(n)``。
+_EQUATION_TAG_RE = re.compile(r"\\#\s*\(([^()]*)\)\s*$")
 
 
 def _escape_text(value: str) -> str:
@@ -83,6 +86,22 @@ def _escape_math_token(value: str) -> str:
     return "".join(_LATEX_MATH_TOKEN_ESCAPES.get(char, char) for char in value)
 
 
+def _tag_equation_number(latex: str) -> str | None:
+    """把 Word 公式编号分隔符残留的行尾 ``\\#(n)`` 转换为 LaTeX ``\\tag``。
+
+    参考 OMML 转换器 ``do_eqarr`` 的既有语义：单行公式数组末尾的编号
+    ``#(n)`` 输出为 ``公式\\tag{n}``，同时避免 ``\\tag`` 进入 matrix 环境。
+    """
+    match = _EQUATION_TAG_RE.search(latex)
+    if match is None:
+        return None
+    formula = latex[: match.start()].rstrip()
+    tag_content = normalize_formula_tag_content(match.group(1))
+    if not formula or not tag_content:
+        return None
+    return rf"{formula}\tag{{{tag_content}}}"
+
+
 def _children(element: etree._Element) -> list[etree._Element]:
     """返回当前元素的全部普通 XML 子元素。"""
     return [child for child in element if isinstance(child.tag, str)]
@@ -107,7 +126,7 @@ def _convert(element: etree._Element) -> str:
     if name in {"mi", "mn"}:
         return _GREEK_MAP.get(text, _escape_math_token(text))
     if name == "mo":
-        return _OPERATOR_MAP.get(text, text)
+        return _OPERATOR_MAP.get(text, _escape_math_token(text))
     if name == "mtext":
         return rf"\text{{{_escape_text(text)}}}"
     if name == "mspace":
@@ -139,6 +158,8 @@ def _convert(element: etree._Element) -> str:
         return rf"\left{opening}{separators[0].join(values)}\right{closing}"
     if name == "mtable":
         rows = [_convert(child) for child in children if local_name(child) in {"mtr", "mlabeledtr"}]
+        if len(rows) == 1 and (tagged := _tag_equation_number(rows[0])) is not None:
+            return tagged
         return r"\begin{matrix}" + r" \\ ".join(rows) + r"\end{matrix}"
     if name in {"mtr", "mlabeledtr"}:
         return " & ".join(_convert(child) for child in children)
@@ -160,7 +181,7 @@ def mathml_to_latex(math_element: etree._Element) -> str | None:
         if "tex" in encoding and (annotation.text or "").strip():
             return (annotation.text or "").strip()
     latex = _convert(math_element).strip()
-    return latex or None
+    return _tag_equation_number(latex) or latex or None
 
 
 __all__ = ["mathml_to_latex"]

{docvortex-0.4.9 → docvortex-0.4.11}/src/docvortex/document/contracts.py

@@ -8,11 +8,16 @@ from pathlib import Path
 
 @dataclass(frozen=True, slots=True)
 class HtmlSourceContext:
-    """保存相对链接解析及 HTML 解码所需的来源上下文。"""
+    """保存相对链接解析及 HTML 解码所需的来源上下文。
+
+    fetch_remote_images 默认关闭，远程图片仅记录 URL；开启后解析阶段会对受限
+    HTTP(S) 图片发起网络请求并内嵌，不可信输入的 SSRF 风险由调用者承担。
+    """
 
     source_uri: str | None = None
     local_resource_root: Path | None = None
     transport_encoding: str | None = None
+    fetch_remote_images: bool = False
 
 
 __all__ = ["HtmlSourceContext"]