docvortex 0.2.2__tar.gz → 0.2.4__tar.gz

{docvortex-0.2.2/src/docvortex.egg-info → docvortex-0.2.4}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: docvortex
-Version: 0.2.2
+Version: 0.2.4
 Summary: A fast, multi-format document parsing and conversion engine
 License-Expression: MIT
 Project-URL: Repository, https://github.com/myhloli/DocVortex
@@ -35,6 +35,7 @@ Requires-Dist: olefile<1,>=0.47
 Requires-Dist: pylatexenc<3,>=2.10
 Requires-Dist: latex2mathml<4,>=3.81.0
 Requires-Dist: mathml2omml==0.0.2
+Requires-Dist: resvg_py<0.6,>=0.5.0
 Requires-Dist: reportlab
 Requires-Dist: ziamath<0.14,>=0.13
 Requires-Dist: fast-langdetect<0.3.0,>=0.2.3
@@ -52,7 +53,7 @@ Requires-Dist: twine; extra == "dev"
 Dynamic: license-file
 
 
-![DocVortex overview: native document inputs flow through a unified document model to Markdown, HTML, LaTeX, DOCX, EPUB, PDF and structured content.](https://raw.githubusercontent.com/myhloli/DocVortex/main/docs/images/docvortex-overview.jpg)
+![DocVortex overview: native document inputs flow through a unified document model to Markdown, HTML, LaTeX, DOCX, EPUB, PDF and structured content.](https://gcore.jsdelivr.net/gh/myhloli/DocVortex@main/docs/images/docvortex-overview.jpg)
 
 # DocVortex
 
@@ -61,12 +62,12 @@ A fast, multi-format document parsing and conversion engine.
 DocVortex provides a complete, standalone document pipeline:
 
 ```text
-Document -> ModelJson + Assets -> MiddleJson + Assets -> Render / Export
+Document -> Unified Intermediate Representation -> Render / Export
 ```
 
 Native inputs include text PDFs, DOC/DOCX, PPT/PPTX, XLS/XLSX, RTF, ODT/ODS/ODP,
 EPUB, HTML, OFD and CSV. Output formats include Markdown, HTML, LaTeX, DOCX,
-EPUB, PDF and structured content. Content List V1/V2 are provided by MinerU.
+EPUB, PDF and structured content. 
 
 Native parsing runs without OCR or VLM inference services.
 PDF classification is an explicit document operation; native analysis does not
@@ -81,8 +82,7 @@ docvortex classify report.pdf
 ```
 
 Python 3.10–3.14 is supported. Native parsing does not require OCR/VLM inference
-services. PDF access uses `pypdfium2>=5.10.1,<6`; the
-compatibility matrix also exercises 5.13.0.
+services. 
 
 ## Parse once, export many times
 
@@ -166,28 +166,5 @@ uv build
 
 DocVortex project code is licensed under the [MIT License](LICENSE.md).
 
-Dependencies include `pydantic>=2.12.5,<3` and `numpy>=1.21.6`; the
-installer selects versions compatible with the active Python interpreter.
-On Apple Silicon, Python 3.14 installation requires macOS 14 or newer because
-of the ONNX Runtime dependency used by file-type detection.
-
-See [the standalone example](demo/README.md) for native parsing and portable
-result bundles, and [the validation record](docs/validation.md) for test coverage.
-
-PDFium uses a bundled, pinned CJK fallback font for non-embedded CJK fonts;
-no system font installation is required. See [PDF font policy](docs/PDF_FONTS.md)
-for initialization, diagnostics and replacement boundaries.
-
-PDF output normalizes fullwidth Latin letters, digits and selected technical
-symbols in natural-language text and table cells, preserving Chinese punctuation,
-formulas, code and link targets. See
-[PDF text normalization](docs/PDF_TEXT_NORMALIZATION.md) for scope and API usage.
-
-See [the DocVortex upgrade guide](docs/DOCVORTEX_UPGRADE.md) for package, protocol,
-PDF text rules and publishing configuration changes.
-
 See [rendering ownership](docs/RENDER_OWNERSHIP.md) for the seven engine targets,
 MinerU Content List integration and public fragment helpers.
-
-See [refactor validation](docs/REFACTOR_PROGRESS.md) for the staged internal
-refactoring, compatibility checks, corpus comparisons and measured performance.

{docvortex-0.2.2 → docvortex-0.2.4}/README.md

@@ -1,5 +1,5 @@
 
-![DocVortex overview: native document inputs flow through a unified document model to Markdown, HTML, LaTeX, DOCX, EPUB, PDF and structured content.](https://raw.githubusercontent.com/myhloli/DocVortex/main/docs/images/docvortex-overview.jpg)
+![DocVortex overview: native document inputs flow through a unified document model to Markdown, HTML, LaTeX, DOCX, EPUB, PDF and structured content.](https://gcore.jsdelivr.net/gh/myhloli/DocVortex@main/docs/images/docvortex-overview.jpg)
 
 # DocVortex
 
@@ -8,12 +8,12 @@ A fast, multi-format document parsing and conversion engine.
 DocVortex provides a complete, standalone document pipeline:
 
 ```text
-Document -> ModelJson + Assets -> MiddleJson + Assets -> Render / Export
+Document -> Unified Intermediate Representation -> Render / Export
 ```
 
 Native inputs include text PDFs, DOC/DOCX, PPT/PPTX, XLS/XLSX, RTF, ODT/ODS/ODP,
 EPUB, HTML, OFD and CSV. Output formats include Markdown, HTML, LaTeX, DOCX,
-EPUB, PDF and structured content. Content List V1/V2 are provided by MinerU.
+EPUB, PDF and structured content. 
 
 Native parsing runs without OCR or VLM inference services.
 PDF classification is an explicit document operation; native analysis does not
@@ -28,8 +28,7 @@ docvortex classify report.pdf
 ```
 
 Python 3.10–3.14 is supported. Native parsing does not require OCR/VLM inference
-services. PDF access uses `pypdfium2>=5.10.1,<6`; the
-compatibility matrix also exercises 5.13.0.
+services. 
 
 ## Parse once, export many times
 
@@ -113,28 +112,5 @@ uv build
 
 DocVortex project code is licensed under the [MIT License](LICENSE.md).
 
-Dependencies include `pydantic>=2.12.5,<3` and `numpy>=1.21.6`; the
-installer selects versions compatible with the active Python interpreter.
-On Apple Silicon, Python 3.14 installation requires macOS 14 or newer because
-of the ONNX Runtime dependency used by file-type detection.
-
-See [the standalone example](demo/README.md) for native parsing and portable
-result bundles, and [the validation record](docs/validation.md) for test coverage.
-
-PDFium uses a bundled, pinned CJK fallback font for non-embedded CJK fonts;
-no system font installation is required. See [PDF font policy](docs/PDF_FONTS.md)
-for initialization, diagnostics and replacement boundaries.
-
-PDF output normalizes fullwidth Latin letters, digits and selected technical
-symbols in natural-language text and table cells, preserving Chinese punctuation,
-formulas, code and link targets. See
-[PDF text normalization](docs/PDF_TEXT_NORMALIZATION.md) for scope and API usage.
-
-See [the DocVortex upgrade guide](docs/DOCVORTEX_UPGRADE.md) for package, protocol,
-PDF text rules and publishing configuration changes.
-
 See [rendering ownership](docs/RENDER_OWNERSHIP.md) for the seven engine targets,
 MinerU Content List integration and public fragment helpers.
-
-See [refactor validation](docs/REFACTOR_PROGRESS.md) for the staged internal
-refactoring, compatibility checks, corpus comparisons and measured performance.

{docvortex-0.2.2 → docvortex-0.2.4}/pyproject.toml

@@ -26,6 +26,7 @@ dependencies = [
     "python-docx>=1.2.0,<2", "pypptx-with-oxml>=1.0.3,<2",
     "mammoth>=1.11.0,<2", "openpyxl>=3.1.5,<4", "olefile>=0.47,<1",
     "pylatexenc>=2.10,<3", "latex2mathml>=3.81.0,<4", "mathml2omml==0.0.2",
+    "resvg_py>=0.5.0,<0.6",
     "reportlab", "ziamath>=0.13,<0.14", "fast-langdetect>=0.2.3,<0.3.0",
     "magika>=0.6.2,<1.1.0", "opencv-python>=4.11.0.86",
 ]

{docvortex-0.2.2 → docvortex-0.2.4}/src/docvortex/analyzers/native/epub/converter.py

@@ -72,6 +72,7 @@ class EpubConverter:
                     logger.warning("Skipping unusable EPUB spine item index={} path={!r}: {}", index, path, exc)
                     blocks = []
                 pages.append(blocks)
+            anchors.finalize_links(pages)
             self.pages = pages
         finally:
             package.close()

{docvortex-0.2.2 → docvortex-0.2.4}/src/docvortex/analyzers/native/epub/xhtml.py

@@ -5,8 +5,10 @@ from __future__ import annotations
 import base64
 from dataclasses import dataclass
 
+from lxml import html as lxml_html
 from lxml import etree  # type: ignore[reportMissingImports]
 
+from ....content.spans import normalize_span_dicts
 from ....foundation.image_payload import parse_image_data_uri_strict
 from .._shared.hyperlink import sanitize_hyperlink_target
 from docvortex.content.markup import (
@@ -52,13 +54,25 @@ _NOTE_NON_TEXT_SUBTREES = frozenset(
 )
 
 
+def _unwrap_epub_type_quotes(value: str) -> str:
+    """移除单层成对外围引号，避免把多个各自带引号的 token 当作一个整体。"""
+    normalized = value.strip()
+    quote_pairs = {'"': '"', "'": "'", "“": "”", "‘": "’"}
+    if len(normalized) < 2 or quote_pairs.get(normalized[0]) != normalized[-1]:
+        return normalized
+    inner = normalized[1:-1]
+    if normalized[0] in inner or normalized[-1] in inner:
+        return normalized
+    return inner.strip()
+
+
 def _epub_types(element: etree._Element) -> frozenset[str]:
     """读取 EPUB 命名空间或未命名 type 属性中的结构语义 token。"""
     values: list[str] = []
     for name, value in element.attrib.items():
         local_name = etree.QName(name).localname if name.startswith("{") else name.split(":", 1)[-1]
         if local_name == "type":
-            values.extend(value.casefold().split())
+            values.extend(_unwrap_epub_type_quotes(token) for token in _unwrap_epub_type_quotes(value).casefold().split())
     return frozenset(values)
 
 
@@ -149,11 +163,12 @@ class _EpubAnchorPolicy:
 
 
 class EpubAnchorRegistry:
-    """建立章节路径、标题与 note fragment 到实际 canonical anchor 的别名表。"""
+    """建立章节路径、正文、标题与 note fragment 到实际 canonical anchor 的别名表。"""
 
     def __init__(self, chapters: list[tuple[str, etree._Element]], package: EpubPackage) -> None:
         """预扫描全部选中 XHTML 章节，建立标题、note 与章节起点映射。"""
         self._package = package
+        self._pending_links: dict[str, tuple[str, str | None]] = {}
         documents = [
             MarkupAnchorDocument(
                 key=chapter_path,
@@ -166,6 +181,35 @@ class EpubAnchorRegistry:
         ]
         self._registry = MarkupAnchorRegistry(documents, _EpubAnchorPolicy())
 
+    def register_text_targets(
+        self,
+        chapter_path: str,
+        root: etree._Element,
+        sources: list[tuple[etree._Element, dict[str, object]]],
+    ) -> None:
+        """在章节投影完成后登记真实正文目标，不修改已有标题和脚注身份。"""
+        self._registry.register_text_targets(chapter_path, root, sources)
+
+    def defer_link(self, href: str, *, base_part: str) -> str | None:
+        """暂存尚未物化的包内目标，跨章节投影完成后统一兑现或降级。"""
+        normalized = sanitize_hyperlink_target(href, allowed_schemes=(), allow_relative=True, allow_fragment=True)
+        target = self._package.resolve_reference(normalized, base_part=base_part) if normalized is not None else None
+        if target is None:
+            return None
+        placeholder = f"#epub-pending-{len(self._pending_links)}"
+        self._pending_links[placeholder] = (target.path, target.fragment)
+        return placeholder
+
+    def finalize_links(self, pages: list[list[dict[str, object]]]) -> None:
+        """消除所有解析期链接占位，保留无效链接的文字与样式。"""
+        targets = {
+            placeholder: (f"#{anchor}" if (anchor := self._registry.resolve_target(path, fragment)) else None)
+            for placeholder, (path, fragment) in self._pending_links.items()
+        }
+        for page in pages:
+            _finalize_pending_content(page, targets)
+        self._pending_links.clear()
+
     def heading_anchor(self, heading: etree._Element) -> str | None:
         """返回一个已预扫描 EPUB 标题的规范 anchor。"""
         return self._registry.heading_anchor(heading)
@@ -179,7 +223,7 @@ class EpubAnchorRegistry:
         return self._registry.note_anchor(note)
 
     def resolve_anchor(self, href: str, *, base_part: str) -> str | None:
-        """解析指向正文标题或 note 的 EPUB 包内链接，并返回不带井号的 anchor。"""
+        """解析指向已登记正文、标题或 note 的包内链接，返回不带井号的 anchor。"""
         normalized = sanitize_hyperlink_target(
             href,
             allowed_schemes=(),
@@ -194,7 +238,7 @@ class EpubAnchorRegistry:
         return self._registry.resolve_target(target.path, target.fragment)
 
     def resolve_link(self, href: str, *, base_part: str) -> str | None:
-        """解析安全外部链接或指向已输出标题/note 的 EPUB 内部链接。"""
+        """解析安全外部链接或指向已登记输出块的 EPUB 内部链接。"""
         external = sanitize_hyperlink_target(href)
         if external is not None:
             return external
@@ -217,7 +261,9 @@ class _EpubMarkupContext:
 
     def resolve_link(self, href: str) -> str | None:
         """解析安全外部链接或实际存在的 EPUB 包内 anchor。"""
-        return self.anchors.resolve_link(href, base_part=self.chapter_path)
+        return self.anchors.resolve_link(href, base_part=self.chapter_path) or self.anchors.defer_link(
+            href, base_part=self.chapter_path
+        )
 
     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。"""
@@ -278,12 +324,50 @@ class EpubChapterConverter:
         if body is None:
             return []
         context = _EpubMarkupContext(self.package, self.chapter_path, self.anchors)
-        return MarkupProjector(
+        projector = MarkupProjector(
             body,
             context,
             self.stylesheet,
             single_document_title=False,
-        ).convert()
+            track_text_sources=True,
+        )
+        blocks = projector.convert()
+        self.anchors.register_text_targets(self.chapter_path, self.root, projector.text_sources)
+        return blocks
+
+
+def _finalize_pending_content(items: list[dict[str, object]], targets: dict[str, str | None]) -> None:
+    """递归处理 raw block 和行内 Span，表格 HTML 同步替换或解包失效链接。"""
+    output: list[dict[str, object]] = []
+    flattened = False
+    for item in items:
+        content = item.get("content")
+        if isinstance(content, list):
+            _finalize_pending_content(content, targets)
+        elif item.get("type") == "table" and isinstance(content, str) and "#epub-pending-" in content:
+            root = lxml_html.fragment_fromstring(content, create_parent="div")
+            for link in root.iter("a"):
+                href = link.get("href")
+                if href in targets:
+                    if target := targets[href]:
+                        link.set("href", target)
+                    else:
+                        link.drop_tag()
+            item["content"] = (root.text or "") + "".join(lxml_html.tostring(child, encoding="unicode") for child in root)
+        url = item.get("url")
+        if item.get("type") == "hyperlink" and isinstance(url, str) and url in targets:
+            if target := targets[url]:
+                item["url"] = target
+            else:
+                if isinstance(content, list):
+                    output.extend(content)
+                flattened = True
+                continue
+        output.append(item)
+    if flattened:
+        # 只有失效链接被解包的 Span 列表需要重新合并，避免重复校验整本正文。
+        output = normalize_span_dicts(output)
+    items[:] = output
 
 
 def convert_svg_spine(

{docvortex-0.2.2 → docvortex-0.2.4}/src/docvortex/analyzers/native/models.py

@@ -66,14 +66,20 @@ class HtmlModel:
 
 
 class OfdModel:
-    """将 OFD 固定版式文档包装为无状态 Flash 模型。"""
+    """包装 OFD 固定版式解析，保留最近一次调用的非致命诊断。"""
+
+    def __init__(self) -> None:
+        """初始化实例级诊断，避免不同文档间串用结果。"""
+        self.diagnostics: list[dict[str, Any]] = []
 
     def predict(self, file_binary: BinaryIO) -> list[list[dict[str, Any]]]:
         """转换调用方持有的整份 OFD 流，并返回逐物理页 model-list。"""
         from .ofd.converter import OfdConverter
 
+        self.diagnostics = []
         converter = OfdConverter()
         converter.convert(file_binary)
+        self.diagnostics = converter.diagnostics
         return converter.pages
 
 

docvortex-0.2.4/src/docvortex/analyzers/native/ofd/assembly.py

@@ -0,0 +1,192 @@
+"""在 OFD 所属区域内恢复视觉行、样式片段和保守段落。"""
+
+from __future__ import annotations
+
+import math
+import re
+import statistics
+from dataclasses import replace
+from typing import Any
+
+from ....schema import BBox, BlockType
+from .geometry import bbox_union
+from .models import TextLine
+from .text import format_line_html, format_line_spans
+
+_LIST_START = re.compile(r"^\s*(?:[•●▪◆]|[-*]\s|\d+[.)、]\s*|[一二三四五六七八九十]+[、）])")
+
+
+def upright_point(x: float, y: float, angle: int) -> tuple[float, float]:
+    """将页面点旋转到文字阅读方向，统一横排与直角旋转的几何计算。"""
+    radians = math.radians(angle)
+    cosine, sine = round(math.cos(radians), 12), round(math.sin(radians), 12)
+    return cosine * x + sine * y, -sine * x + cosine * y
+
+
+def upright_box(bbox: BBox, angle: int) -> BBox:
+    """在文字方向坐标内计算矩形外接框。"""
+    points = [upright_point(x, y, angle) for x in (bbox[0], bbox[2]) for y in (bbox[1], bbox[3])]
+    return min(p[0] for p in points), min(p[1] for p in points), max(p[0] for p in points), max(p[1] for p in points)
+
+
+def line_baseline(line: TextLine) -> float:
+    """优先读取真实字形基线，缺少字形时才退回视觉框中心。"""
+    if line.glyphs:
+        return statistics.median(upright_point(*glyph.origin, line.angle)[1] for glyph in line.glyphs)
+    box = upright_box(line.bbox, line.angle)
+    return (box[1] + box[3]) / 2
+
+
+def line_runs(line: TextLine) -> tuple[tuple[str, tuple[str, ...]], ...]:
+    """读取已组装行的样式片段，原始行按其自身样式返回单个片段。"""
+    return line.runs or ((line.text, line.styles),)
+
+
+def line_spans(line: TextLine) -> list[dict[str, object]]:
+    """将视觉行的各样式片段投影为既有结构化 Span。"""
+    return [span for text, styles in line_runs(line) for span in format_line_spans(text, styles)]
+
+
+def line_html(line: TextLine) -> str:
+    """按片段样式序列化单元格文字，不把首片段样式扩散到整行。"""
+    return "".join(format_line_html(text, styles) for text, styles in line_runs(line))
+
+
+def _fragment_separator(first: TextLine, second: TextLine) -> str:
+    """仅在英文单词片段之间存在可见词间距时补空格。"""
+    if not first.text or not second.text:
+        return ""
+    if not (first.text[-1].isascii() and second.text[0].isascii() and first.text[-1].isalnum() and second.text[0].isalnum()):
+        return ""
+    first_box = upright_box(first.glyphs[-1].bbox if first.glyphs else first.bbox, first.angle)
+    second_box = upright_box(second.glyphs[0].bbox if second.glyphs else second.bbox, second.angle)
+    widths = [box[2] - box[0] for box in (first_box, second_box)]
+    threshold = 0.2 * max(min(first.font_size, second.font_size), statistics.median(widths), 0.5)
+    return " " if second_box[0] - first_box[2] >= threshold else ""
+
+
+def merge_same_baseline_lines(lines: list[TextLine]) -> list[TextLine]:
+    """按真实基线聚行并沿阅读方向拼接，保留样式及层和模板隔离。"""
+    metrics = {id(line): (line_baseline(line), upright_box(line.bbox, line.angle)) for line in lines}
+    ordered = sorted(lines, key=lambda line: (line.angle, line.layer_type, line.template_id or -1, metrics[id(line)][0]))
+    rows: list[list[TextLine]] = []
+    for line in ordered:
+        previous = rows[-1][0] if rows else None
+        if (
+            previous is not None
+            and (line.angle, line.layer_type, line.template_id) == (previous.angle, previous.layer_type, previous.template_id)
+            and abs(metrics[id(line)][0] - metrics[id(previous)][0]) <= 0.2 * min(line.font_size, previous.font_size)
+        ):
+            rows[-1].append(line)
+        else:
+            rows.append([line])
+    output: list[TextLine] = []
+    for row in rows:
+        fragments = sorted(row, key=lambda line: (metrics[id(line)][1][0], line.paint_order))
+        assembled: list[TextLine] = []
+        for line in fragments:
+            previous = assembled[-1] if assembled else None
+            if previous is not None:
+                first_box, second_box = upright_box(previous.bbox, line.angle), metrics[id(line)][1]
+                size = min(previous.font_size, line.font_size)
+                gap = second_box[0] - first_box[2]
+                if size > 0 and max(previous.font_size, line.font_size) <= 1.35 * size and -0.25 * size <= gap <= 0.8 * size:
+                    separator = _fragment_separator(previous, line)
+                    runs = line_runs(previous) + (((separator, ()),) if separator else ()) + line_runs(line)
+                    assembled[-1] = replace(
+                        previous,
+                        text=previous.text + separator + line.text,
+                        bbox=bbox_union((previous.bbox, line.bbox)) or previous.bbox,
+                        glyphs=previous.glyphs + line.glyphs,
+                        paint_order=min(previous.paint_order, line.paint_order),
+                        runs=runs,
+                    )
+                    continue
+            assembled.append(line)
+        output.extend(assembled)
+    return output
+
+
+def _block_text(block: dict[str, Any]) -> str:
+    """读取 OFD 私有视觉行文本，避免依赖公共 Span 的序列化细节。"""
+    return block.get("text_line", "")
+
+
+def _paragraph_separator(previous: str, current: str) -> str:
+    """中文换行直接续接，英文词和句末标点后的换行补空格，保留已有空白与连字符。"""
+    last, first = previous[-1:], current[:1]
+    if not last or not first or not (last.isascii() and first.isascii()):
+        return ""
+    return " " if (last.isalnum() or last in '.,;:!?)]}"') and (first.isalnum() or first in '(["') else ""
+
+
+def merge_paragraph_blocks(blocks: list[dict[str, Any]]) -> list[dict[str, Any]]:
+    """在阅读顺序内按局部栏宽、行距及缩进合段，视觉块和辅助块作为屏障。"""
+    output: list[dict[str, Any]] = []
+    for index, block in enumerate(blocks):
+        if block["type"] != BlockType.TEXT:
+            output.append(block)
+            continue
+        angle = block["angle"]
+        current_box = upright_box(block["bbox_mm"], angle)
+        size = block["font_size_mm"]
+        # 只观察邻近且横向重叠的正文，防止另一栏或远处区域影响本栏尺度。
+        neighbors: list[dict[str, Any]] = []
+        for direction in (-1, 1):
+            for offset in range(1, 9):
+                position = index + direction * offset
+                if not 0 <= position < len(blocks):
+                    break
+                candidate = blocks[position]
+                if (
+                    candidate["type"] != BlockType.TEXT
+                    or candidate["angle"] != angle
+                    or candidate["text_scope"] != block["text_scope"]
+                ):
+                    break
+                box = upright_box(candidate["bbox_mm"], angle)
+                if abs(box[1] - current_box[1]) > 20 * size or min(box[2], current_box[2]) <= max(box[0], current_box[0]):
+                    break
+                if abs(box[0] - current_box[0]) <= 3 * size:
+                    neighbors.append(candidate)
+        context = neighbors + [block]
+        boxes = [upright_box(item["bbox_mm"], angle) for item in context]
+        right = max(box[2] for box in boxes)
+        baselines = sorted({item["baseline_mm"] for item in context})
+        gaps = [b - a for a, b in zip(baselines[:-1], baselines[1:], strict=True) if b - a >= 0.5 * size]
+        pitch = statistics.median(gaps) if gaps else 1.5 * size
+        previous = output[-1] if output else None
+        can_merge = (
+            previous is not None
+            and previous["type"] == BlockType.TEXT
+            and previous["angle"] == angle
+            and previous["text_scope"] == block["text_scope"]
+        )
+        if can_merge:
+            last_box = upright_box(previous["line_bboxes_mm"][-1], angle)
+            previous_size = previous["font_size_mm"]
+            delta = block["baseline_mm"] - previous["baseline_mm"]
+            previous_text, current_text = _block_text(previous).strip(), _block_text(block).strip()
+            can_merge = (
+                min(size, previous_size) > 0
+                and max(size, previous_size) <= 1.25 * min(size, previous_size)
+                and 0.5 * size <= delta <= min(1.4 * pitch, 3 * size)
+                and -2.5 * size <= current_box[0] - last_box[0] <= 0.8 * size
+                and right - last_box[2] <= 1.5 * size
+                and min(last_box[2], current_box[2]) > max(last_box[0], current_box[0])
+                and not _LIST_START.match(current_text)
+                and not (len(previous_text) <= 4 or len(current_text) <= 4)
+            )
+        if can_merge:
+            previous_text, current_text = _block_text(previous), _block_text(block)
+            separator = _paragraph_separator(previous_text, current_text)
+            previous["content"].extend(format_line_spans(separator, ()))
+            previous["content"].extend(block["content"])
+            previous["bbox_mm"] = bbox_union((previous["bbox_mm"], block["bbox_mm"]))
+            previous["line_bboxes_mm"].extend(block["line_bboxes_mm"])
+            previous["baseline_mm"] = block["baseline_mm"]
+            previous["text_line"] = current_text
+        else:
+            # 复制将被追加的列表，确保组段不会修改调用方的原始视觉行。
+            output.append({**block, "content": list(block["content"]), "line_bboxes_mm": list(block["line_bboxes_mm"])})
+    return output

docvortex-0.2.4/src/docvortex/analyzers/native/ofd/converter.py

@@ -0,0 +1,69 @@
+"""OFD 二进制流到分页 raw model-list 的转换入口。"""
+
+from __future__ import annotations
+
+from typing import BinaryIO
+import base64
+
+from .constants import MAX_TOTAL_BYTES
+from .errors import OfdResourceLimitError
+from .package import OfdPackage
+from .reading_order import OfdReadingOrderProjector
+from .scene import OfdSceneBuilder
+from .models import ImageItem
+from .vector import render_vector_page, UnsupportedVector
+
+
+class OfdConverter:
+    """编排 OFD 包读取、场景构建和阅读顺序投影。"""
+
+    def __init__(self) -> None:
+        """初始化空的分页输出。"""
+        self.pages: list[list[dict[str, object]]] = []
+        self.diagnostics: list[dict[str, object]] = []
+
+    def convert(self, file_binary: BinaryIO) -> None:
+        """读取整份 OFD 并更新分页 model-list。"""
+        self.pages = []
+        self.diagnostics = []
+        file_bytes = file_binary.read(MAX_TOTAL_BYTES + 1)
+        if len(file_bytes) > MAX_TOTAL_BYTES:
+            raise OfdResourceLimitError(f"OFD resource limit exceeded: max_total_bytes={MAX_TOTAL_BYTES}")
+        with OfdPackage(file_bytes) as package:
+            scenes = OfdSceneBuilder(package).build()
+            for scene in scenes:
+                if (
+                    not scene.text_object_count
+                    and not scene.image_object_count
+                    and scene.vector_paths
+                    and not scene.vector_unsupported
+                ):
+                    try:
+                        payload = render_vector_page(scene, package)
+                    except OfdResourceLimitError:
+                        raise
+                    except (UnsupportedVector, ValueError, RuntimeError, OSError) as exc:
+                        scene.diagnostics.append({"code": "ofd_vector_render_failed", "message": str(exc)})
+                    else:
+                        if payload is not None:
+                            scene.images.append(
+                                ImageItem(
+                                    scene.physical_box,
+                                    "data:image/png;base64," + base64.b64encode(payload).decode("ascii"),
+                                    0,
+                                    None,
+                                    "Body",
+                                    None,
+                                )
+                            )
+                            scene.diagnostics.append(
+                                {
+                                    "code": "ofd_vector_rasterized",
+                                    "message": "Vector-only page preserved as an image; no text recognition was performed",
+                                }
+                            )
+                self.diagnostics.extend({**item, "page_index": scene.page_idx} for item in scene.diagnostics)
+            self.pages = OfdReadingOrderProjector(scenes).project()
+
+
+__all__ = ["OfdConverter"]

{docvortex-0.2.2 → docvortex-0.2.4}/src/docvortex/analyzers/native/ofd/models.py

@@ -3,13 +3,16 @@
 from __future__ import annotations
 
 from dataclasses import dataclass, field
-from typing import Any
+from typing import Any, TYPE_CHECKING
 
 from lxml import etree  # type: ignore[reportMissingImports]
 
 from ....schema import BBox
 from .geometry import Affine, Point, Quad
 
+if TYPE_CHECKING:
+    from .vector import VectorPath, ClipPath
+
 
 @dataclass(frozen=True, slots=True)
 class FontResource:
@@ -78,6 +81,7 @@ class TextLine:
     layer_type: str
     template_id: int | None
     styles: tuple[str, ...] = ()
+    runs: tuple[tuple[str, tuple[str, ...]], ...] = ()
 
 
 @dataclass(frozen=True, slots=True)
@@ -115,6 +119,10 @@ class OfdPageScene:
     axis_lines: list[AxisLine] = field(default_factory=list)
     images: list[ImageItem] = field(default_factory=list)
     diagnostics: list[dict[str, Any]] = field(default_factory=list)
+    vector_paths: list[VectorPath] = field(default_factory=list)
+    text_object_count: int = 0
+    image_object_count: int = 0
+    vector_unsupported: bool = False
 
 
 @dataclass(frozen=True, slots=True)
@@ -151,6 +159,7 @@ class PageBuildContext:
     layer_type: str
     template_id: int | None
     draw_style: dict[str, str] = field(default_factory=dict)
+    vector_clips: tuple[tuple[ClipPath, ...], ...] = ()
 
 
 __all__ = [

{docvortex-0.2.2 → docvortex-0.2.4}/src/docvortex/analyzers/native/ofd/package.py

@@ -101,6 +101,12 @@ class OfdPackage:
         self._asset_bytes = 0
         self._root: etree._Element | None = None
 
+    def charge_generated_asset(self, size: int) -> None:
+        """将生成的 PNG 与包内读取素材累计到同一文档资产预算。"""
+        if self._asset_bytes + size > MAX_ASSET_TOTAL_BYTES:
+            raise OfdResourceLimitError(f"OFD resource limit exceeded: max_asset_total_bytes={MAX_ASSET_TOTAL_BYTES}")
+        self._asset_bytes += size
+
     @staticmethod
     def _is_safe_member_name(name: str) -> bool:
         """判断 ZIP 成员是否为包内安全 POSIX 路径。"""