docvortex 0.4.9__tar.gz → 0.4.10__tar.gz

{docvortex-0.4.9/src/docvortex.egg-info → docvortex-0.4.10}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: docvortex
-Version: 0.4.9
+Version: 0.4.10
 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-0.4.10}/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-0.4.10}/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-0.4.10}/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]
 
@@ -19,6 +20,7 @@ from .constants import (
     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 +67,7 @@ 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 +160,14 @@ 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
@@ -243,6 +246,41 @@ 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-0.4.10}/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-0.4.10}/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-0.4.10}/src/docvortex/analyzers/native/office/odf/text.py

@@ -453,7 +453,7 @@ 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 +466,20 @@ 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)
@@ -875,6 +878,20 @@ 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 +899,7 @@ 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 +920,8 @@ 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 +947,30 @@ 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-0.4.10}/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"]

{docvortex-0.4.9 → docvortex-0.4.10}/src/docvortex/version.py

@@ -1,3 +1,3 @@
-__version__ = "0.4.9"
+__version__ = "0.4.10"
 
 __all__ = ["__version__"]

{docvortex-0.4.9 → docvortex-0.4.10/src/docvortex.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: docvortex
-Version: 0.4.9
+Version: 0.4.10
 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-0.4.10}/src/docvortex.egg-info/SOURCES.txt

@@ -430,7 +430,6 @@ src/docvortex/render/_internal/pdf/title_layout.py
 src/docvortex/render/_internal/structured_content/__init__.py
 src/docvortex/render/_internal/structured_content/renderer.py
 src/docvortex/resources/epub/docvortex.css
-src/docvortex/resources/fasttext-langdetect/lid.176.ftz
 src/docvortex/resources/fonts/DroidSansFallbackFull.ttf
 src/docvortex/resources/fonts/NOTICE
 src/docvortex/resources/fonts/manifest.json

{docvortex-0.4.9 → docvortex-0.4.10}/tests/test_cli_logging.py

@@ -20,7 +20,7 @@ def isolated_logger() -> Iterator[None]:
     logger.remove()
     yield
     logger.remove()
-    logger.add(sys.stderr)
+    logger.add(sys.stderr, level="INFO")
 
 
 class _DummyPDFDocument:

{docvortex-0.4.9 → docvortex-0.4.10}/tests/unittest/test_flash_html.py

@@ -1,7 +1,11 @@
 from __future__ import annotations
 
 import base64
+import threading
+from collections.abc import Iterator
+from contextlib import contextmanager
 from copy import copy
+from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
 from io import BytesIO
 from pathlib import Path
 
@@ -19,6 +23,7 @@ from docvortex.analyzers.native.html import HtmlResourceLimitError
 from docvortex.analyzers.native.html import document as html_document_module
 from docvortex.analyzers.native.html import resources as html_resources_module
 from docvortex.analyzers.native.html import selector as html_selector_module
+from docvortex.analyzers.native.html.constants import MAX_HTML_IMAGE_BYTES
 from docvortex.analyzers.native.html.resources import HtmlResourceContext
 from docvortex.api import parse
 from docvortex.codecs.html import decode_docvortex_html_wire
@@ -1982,3 +1987,134 @@ def test_html_invalid_mathml_and_asciimath_remain_visible_text() -> None:
 
     assert "not-latex" in markdown and "sqrt(2)" in markdown
     assert not any(block.type == BlockType.EQUATION for block in middle.pages[0].blocks)
+
+
+def _tiny_png_bytes() -> bytes:
+    """构造 1x1 红色像素的最小合法 PNG。"""
+    buffer = BytesIO()
+    Image.new("RGB", (1, 1), (255, 0, 0)).save(buffer, format="PNG")
+    return buffer.getvalue()
+
+
+@contextmanager
+def _local_image_server(routes: dict[str, bytes]) -> Iterator[tuple[str, list[str]]]:
+    """在回环地址启动按路径返回固定字节的计数服务，退出时关闭。"""
+    hits: list[str] = []
+
+    class Handler(BaseHTTPRequestHandler):
+        def do_GET(self) -> None:
+            hits.append(self.path)
+            body = routes[self.path]
+            self.send_response(200)
+            self.send_header("Content-Length", str(len(body)))
+            self.end_headers()
+            self.wfile.write(body)
+
+        def log_message(self, *args: object) -> None:
+            return
+
+    server = ThreadingHTTPServer(("127.0.0.1", 0), Handler)
+    threading.Thread(target=server.serve_forever, daemon=True).start()
+    try:
+        yield f"http://127.0.0.1:{server.server_address[1]}", hits
+    finally:
+        server.shutdown()
+        server.server_close()
+
+
+def _remote_chart_middle(url: str) -> MiddleJson:
+    """构造主图片指向远程 URL 的 chart wire 文档，复现导出 HTML 回灌场景。"""
+    return MiddleJson.model_validate(
+        {
+            "pages": [
+                {
+                    "page_idx": 0,
+                    "blocks": [
+                        {
+                            "type": "chart",
+                            "index": 0,
+                            "content": [{"type": "chart_body", "index": 0, "content": "", "image_url": url}],
+                        }
+                    ],
+                }
+            ],
+            "is_full_document": True,
+            "metadata": {"file_suffix": "html", "producer": {"name": "docvortex", "version": "0.2.0"}},
+            "schema": "docvortex.middle",
+            "schema_version": "2.0",
+        }
+    )
+
+
+def test_html_remote_images_stay_external_and_offline_by_default(tmp_path: Path) -> None:
+    """默认不下载远程图片：保持外链、零网络请求，save_bundle 仍拒绝未物化图片。"""
+    with _local_image_server({"/chart.png": _tiny_png_bytes()}) as (base, hits):
+        url = f"{base}/chart.png"
+        default = HtmlResourceContext(HtmlSourceContext())
+        resolved = default.resolve_image(url, alt="chart")
+        assert resolved is not None and resolved.image_url == url and not resolved.image_base64
+
+        rendered = render_html(_remote_chart_middle(url), standalone=False)
+        result = parse(rendered.encode(), file_suffix="html")
+        body = result.middle_json.pages[0].blocks[0].content[0]
+        assert body.image_url == url and not body.image_path
+
+        with pytest.raises(ValueError, match="must be materialized"):
+            result.save_bundle(tmp_path / "bundle")
+        assert hits == []
+
+
+def test_html_remote_image_fetch_embeds_and_caches_by_url() -> None:
+    """开启下载后有效图片内嵌为 data URI，同一 URL 复用缓存不重复请求。"""
+    png = _tiny_png_bytes()
+    with _local_image_server({"/chart.png": png}) as (base, hits):
+        context = HtmlResourceContext(HtmlSourceContext(fetch_remote_images=True))
+        first = context.resolve_image(f"{base}/chart.png", alt="chart")
+        assert first is not None and first.image_url is None
+        assert first.image_base64 is not None and first.image_base64.startswith("data:image/png;base64,")
+        assert base64.b64decode(first.image_base64.split(",", 1)[1]) == png
+        assert hits == ["/chart.png"]
+
+        second = context.resolve_image(f"{base}/chart.png", alt="chart")
+        assert second is not None and second.image_base64 == first.image_base64
+        assert hits == ["/chart.png"]
+
+
+def test_html_remote_image_fetch_falls_back_to_url_when_unavailable() -> None:
+    """非图片字节或网络不可达时下载退回外链，不阻断解析。"""
+    with _local_image_server({"/text.png": b"not-an-image"}) as (base, hits):
+        context = HtmlResourceContext(HtmlSourceContext(fetch_remote_images=True))
+        resolved = context.resolve_image(f"{base}/text.png", alt="chart")
+        assert resolved is not None and resolved.image_url == f"{base}/text.png" and not resolved.image_base64
+
+        dead = "http://127.0.0.1:9/chart.png"
+        resolved = context.resolve_image(dead)
+        assert resolved is not None and resolved.image_url == dead and not resolved.image_base64
+
+
+def test_html_remote_image_fetch_rejects_oversized_download() -> None:
+    """下载超过单图上限的远程图片时与本地文件一致地终止整份文档。"""
+    oversized = b"\x89PNG\r\n\x1a\n" + b"\x00" * (MAX_HTML_IMAGE_BYTES + 1)
+    with _local_image_server({"/big.png": oversized}) as (base, hits):
+        context = HtmlResourceContext(HtmlSourceContext(fetch_remote_images=True))
+        with pytest.raises(HtmlResourceLimitError, match="max_html_image_bytes"):
+            context.resolve_image(f"{base}/big.png")
+
+
+def test_html_remote_image_fetch_roundtrips_into_saved_bundle(tmp_path: Path) -> None:
+    """开启下载后：远程图片在解析阶段内嵌，save_bundle 不再依赖网络且图片入包。"""
+    png = _tiny_png_bytes()
+    with _local_image_server({"/chart.png": png}) as (base, hits):
+        url = f"{base}/chart.png"
+        rendered = render_html(_remote_chart_middle(url), standalone=False)
+        result = parse(
+            rendered.encode(),
+            file_suffix="html",
+            source_context=HtmlSourceContext(fetch_remote_images=True),
+        )
+        assert hits == ["/chart.png"]
+        body = result.middle_json.pages[0].blocks[0].content[0]
+        assert body.image_path and not body.image_url and not body.image_base64
+
+        result.save_bundle(tmp_path / "bundle")
+        assert [path.read_bytes() for path in sorted((tmp_path / "bundle" / "images").glob("*"))] == [png]