Transforms complex documents like PDFs and Office docs into LLM-ready markdown/JSON for your Agentic workflows.
Python ai4sciencedocument-analysisdocxextract-datalayout-analysisocrparserpdfpdf-converterpdf-extractor-llmpdf-extractor-pretrainpdf-extractor-ragpdf-parserpptxpythonxlsx
Preserve local CSS fragment URLs in MHTML previews
myhloli237c01c856d1
diff --git a/mineru/kit/gradio/mhtml_preview.py b/mineru/kit/gradio/mhtml_preview.pyindex f2738432..52e9beed 100644--- a/mineru/kit/gradio/mhtml_preview.py+++ b/mineru/kit/gradio/mhtml_preview.py@@ -160,7 +160,11 @@ class _PreviewResources:for token in tokens:reference = self._css_reference(token)if reference is not None and getattr(token, "type", "") != "string":- pieces.append(self._css_url(self.resource_url(reference, base_uri, charge=False)))+ # 片段地址引用当前 srcdoc 的 SVG 定义,不能按归档来源地址重写。+ if reference.startswith("#"):+ pieces.append(token.serialize())+ else:+ pieces.append(self._css_url(self.resource_url(reference, base_uri, charge=False)))elif getattr(token, "type", "") == "function":pieces.append(f"{token.name}({self._tokens(token.arguments, base_uri, active)})")elif getattr(token, "type", "") in {"() block", "[] block", "{} block"}:diff --git a/tests/unittest/test_kit_gradio_mhtml_preview.py b/tests/unittest/test_kit_gradio_mhtml_preview.pyindex dc2b9a80..08745ba3 100644--- a/tests/unittest/test_kit_gradio_mhtml_preview.py+++ b/tests/unittest/test_kit_gradio_mhtml_preview.py@@ -83,6 +83,24 @@ def test_mhtml_preview_reuses_archived_assets_and_preserves_media(cycle: bool) -assert "@import" not in nested+def test_mhtml_preview_preserves_fragment_only_css_urls() -> None:+ """归档、内联和属性 CSS 的 SVG 局部引用仍指向预览文档中的定义。"""+ document = _document(+ _archive(+ root_html='<html><head><link rel="stylesheet" href="effects.css">'+ "<style>.inline{filter:url(#blur)}</style></head>"+ '<body><svg><defs><filter id="blur"></filter><clipPath id="mask"></clipPath></defs></svg>'+ '<p class="inline external" style=\'clip-path:url("#mask")\'>正文</p></body></html>',+ extra_parts=(_part(".external{filter:url(#blur)}", "text/css", "https://site.test/effects.css"),),+ )+ )+ stylesheet = document.find("link", rel="stylesheet")+ css = base64.b64decode(stylesheet["href"].split(",", 1)[1]).decode("utf-8")+ assert "filter:url(#blur)" in css+ assert "filter:url(#blur)" in document.style.get_text()+ assert 'clip-path:url("#mask")' in document.p["style"]++def test_mhtml_preview_preserves_inactive_stylesheet_state() -> None:"""归档样式内嵌后仍保留备用和禁用标记,不让它们覆盖当前页面样式。"""document = _document(