From 77d5aa6165ff5d08903421afe1d5245d913a82d2 Mon Sep 17 00:00:00 2001
From: ktz03 <2484593937@qq.com>
Date: Fri, 9 Oct 2026 14:51:41 +0800
Subject: [PATCH] fix(markdown): resolve relative HTML base URLs

---
 crawl4ai/html2text/__init__.py       |  7 +---
 tests/unit/test_markdown_base_url.py | 57 ++++++++++++++++++++++++++++
 2 files changed, 59 insertions(+), 5 deletions(-)
 create mode 100644 tests/unit/test_markdown_base_url.py

diff --git a/crawl4ai/html2text/__init__.py b/crawl4ai/html2text/__init__.py
index ff39fb211..b3079db17 100644
--- a/crawl4ai/html2text/__init__.py
+++ b/crawl4ai/html2text/__init__.py
@@ -320,7 +320,7 @@ def handle_tag(
         if tag == "base" and start:
             href = attrs.get("href")
             if href:
-                self.baseurl = href
+                self.baseurl = urlparse.urljoin(self.baseurl, href)
 
         # first thing inside the anchor tag is another tag
         # that produces some output
@@ -1107,10 +1107,7 @@ def handle_tag(self, tag, attrs, start):
         # Handle <base> tag to update base URL for relative links
         # Must be handled before preserved tags since <base> is in <head>
         if tag == "base" and start:
-            href = attrs.get("href") if attrs else None
-            if href:
-                self.baseurl = href
-            # Also let parent class handle it
+            # Let the parent resolve relative hrefs against the document URL once.
             return super().handle_tag(tag, attrs, start)
 
         # Handle preserved tags
diff --git a/tests/unit/test_markdown_base_url.py b/tests/unit/test_markdown_base_url.py
new file mode 100644
index 000000000..b0a6159eb
--- /dev/null
+++ b/tests/unit/test_markdown_base_url.py
@@ -0,0 +1,57 @@
+"""Relative base elements resolve against the original document URL."""
+
+import pytest
+
+from crawl4ai import DefaultMarkdownGenerator
+from crawl4ai.html2text import CustomHTML2Text, HTML2Text
+
+BASE_CASES = [
+    ("../assets/", "https://example.test/assets/"),
+    ("assets/", "https://example.test/pages/assets/"),
+    ("/docs/", "https://example.test/docs/"),
+    ("//cdn.example.test/v1/", "https://cdn.example.test/v1/"),
+    ("https://cdn.example.test/v1/", "https://cdn.example.test/v1/"),
+    (None, "https://example.test/pages/"),
+    ("", "https://example.test/pages/"),
+]
+PAGE_URL = "https://example.test/pages/index.html"
+
+
+def _page_html(base_href):
+    base_tag = f'<base href="{base_href}">' if base_href is not None else ""
+    return (
+        f"<html><head>{base_tag}</head><body><p>"
+        '<a href="guide.html">Guide</a><img src="logo.png" alt="Logo">'
+        "</p></body></html>"
+    )
+
+
+@pytest.mark.parametrize("converter_class", [HTML2Text, CustomHTML2Text])
+@pytest.mark.parametrize(("base_href", "expected_prefix"), BASE_CASES)
+def test_relative_base_element_resolves_links_and_images(
+    converter_class, base_href, expected_prefix
+):
+    converter = converter_class(baseurl=PAGE_URL, bodywidth=0)
+
+    markdown = converter.handle(_page_html(base_href))
+
+    assert f"[Guide]({expected_prefix}guide.html)" in markdown
+    assert f"![Logo]({expected_prefix}logo.png)" in markdown
+
+
+@pytest.mark.parametrize("citations", [False, True])
+@pytest.mark.parametrize(("base_href", "expected_prefix"), BASE_CASES)
+def test_markdown_generator_resolves_relative_base_element(
+    citations, base_href, expected_prefix
+):
+    result = DefaultMarkdownGenerator().generate_markdown(
+        _page_html(base_href), base_url=PAGE_URL, citations=citations
+    )
+
+    assert f"[Guide]({expected_prefix}guide.html)" in result.raw_markdown
+    assert f"![Logo]({expected_prefix}logo.png)" in result.raw_markdown
+    if citations:
+        assert f"{expected_prefix}guide.html" in result.references_markdown
+        assert f"{expected_prefix}logo.png" in result.references_markdown
+    else:
+        assert result.references_markdown == ""