fix: fix parse mailto / http link in table cell (#33224)

Co-authored-by: Oz <oz-agent@warp.dev>

api/core/rag/extractor/word_extractor.py

@@ -204,26 +204,61 @@ class WordExtractor(BaseExtractor):
         return " ".join(unique_content)
 
     def _parse_cell_paragraph(self, paragraph, image_map):
-        paragraph_content = []
-        for run in paragraph.runs:
-            if run.element.xpath(".//a:blip"):
-                for blip in run.element.xpath(".//a:blip"):
-                    image_id = blip.get("{http://schemas.openxmlformats.org/officeDocument/2006/relationships}embed")
-                    if not image_id:
-                        continue
-                    rel = paragraph.part.rels.get(image_id)
-                    if rel is None:
-                        continue
-                    # For external images, use image_id as key; for internal, use target_part
-                    if rel.is_external:
-                        if image_id in image_map:
-                            paragraph_content.append(image_map[image_id])
-                    else:
-                        image_part = rel.target_part
-                        if image_part in image_map:
-                            paragraph_content.append(image_map[image_part])
-            else:
-                paragraph_content.append(run.text)
+        paragraph_content: list[str] = []
+
+        for child in paragraph._element:
+            tag = child.tag
+            if tag == qn("w:hyperlink"):
+                # Note: w:hyperlink elements may also use w:anchor for internal bookmarks.
+                # This extractor intentionally only converts external links (HTTP/mailto, etc.)
+                # that are backed by a relationship id (r:id) with rel.is_external == True.
+                # Hyperlinks without such an external rel (including anchor-only bookmarks)
+                # are left as plain text link_text.
+                r_id = child.get(qn("r:id"))
+                link_text_parts: list[str] = []
+                for run_elem in child.findall(qn("w:r")):
+                    run = Run(run_elem, paragraph)
+                    if run.text:
+                        link_text_parts.append(run.text)
+                link_text = "".join(link_text_parts).strip()
+                if r_id:
+                    try:
+                        rel = paragraph.part.rels.get(r_id)
+                        if rel:
+                            target_ref = getattr(rel, "target_ref", None)
+                            if target_ref:
+                                parsed_target = urlparse(str(target_ref))
+                                if rel.is_external or parsed_target.scheme in ("http", "https", "mailto"):
+                                    display_text = link_text or str(target_ref)
+                                    link_text = f"[{display_text}]({target_ref})"
+                    except Exception:
+                        logger.exception("Failed to resolve URL for hyperlink with r:id: %s", r_id)
+                if link_text:
+                    paragraph_content.append(link_text)
+
+            elif tag == qn("w:r"):
+                run = Run(child, paragraph)
+                if run.element.xpath(".//a:blip"):
+                    for blip in run.element.xpath(".//a:blip"):
+                        image_id = blip.get(
+                            "{http://schemas.openxmlformats.org/officeDocument/2006/relationships}embed"
+                        )
+                        if not image_id:
+                            continue
+                        rel = paragraph.part.rels.get(image_id)
+                        if rel is None:
+                            continue
+                        if rel.is_external:
+                            if image_id in image_map:
+                                paragraph_content.append(image_map[image_id])
+                        else:
+                            image_part = rel.target_part
+                            if image_part in image_map:
+                                paragraph_content.append(image_map[image_part])
+                else:
+                    if run.text:
+                        paragraph_content.append(run.text)
+
         return "".join(paragraph_content).strip()
 
     def parse_docx(self, docx_path):

api/tests/unit_tests/core/rag/extractor/test_word_extractor.py

@@ -423,15 +423,6 @@ def test_table_to_markdown_and_parse_helpers(monkeypatch):
     markdown = extractor._table_to_markdown(table, {})
     assert markdown == "| H1 | H2 |\n| --- | --- |\n| A | B |"
 
-    class FakeRunElement:
-        def __init__(self, blips):
-            self._blips = blips
-
-        def xpath(self, pattern):
-            if pattern == ".//a:blip":
-                return self._blips
-            return []
-
     class FakeBlip:
         def __init__(self, image_id):
             self.image_id = image_id
@@ -439,11 +430,31 @@ def test_table_to_markdown_and_parse_helpers(monkeypatch):
         def get(self, key):
             return self.image_id
 
+    class FakeRunChild:
+        def __init__(self, blips, text=""):
+            self._blips = blips
+            self.text = text
+            self.tag = qn("w:r")
+
+        def xpath(self, pattern):
+            if pattern == ".//a:blip":
+                return self._blips
+            return []
+
+    class FakeRun:
+        def __init__(self, element, paragraph):
+            # Mirror the subset used by _parse_cell_paragraph
+            self.element = element
+            self.text = getattr(element, "text", "")
+
+    # Patch we.Run so our lightweight child objects work with the extractor
+    monkeypatch.setattr(we, "Run", FakeRun)
+
     image_part = object()
     paragraph = SimpleNamespace(
-        runs=[
-            SimpleNamespace(element=FakeRunElement([FakeBlip(None), FakeBlip("ext"), FakeBlip("int")]), text=""),
-            SimpleNamespace(element=FakeRunElement([]), text="plain"),
+        _element=[
+            FakeRunChild([FakeBlip(None), FakeBlip("ext"), FakeBlip("int")], text=""),
+            FakeRunChild([], text="plain"),
         ],
         part=SimpleNamespace(
             rels={
@@ -452,6 +463,7 @@ def test_table_to_markdown_and_parse_helpers(monkeypatch):
             }
         ),
     )
+
     image_map = {"ext": "EXT-IMG", image_part: "INT-IMG"}
     assert extractor._parse_cell_paragraph(paragraph, image_map) == "EXT-IMGINT-IMGplain"
 
@@ -625,3 +637,83 @@ def test_parse_docx_covers_drawing_shapes_hyperlink_error_and_table_branch(monke
     assert "BrokenLink" in content
     assert "TABLE-MARKDOWN" in content
     logger_exception.assert_called_once()
+
+
+def test_parse_cell_paragraph_hyperlink_in_table_cell_http():
+    doc = Document()
+    table = doc.add_table(rows=1, cols=1)
+    cell = table.cell(0, 0)
+    p = cell.paragraphs[0]
+
+    # Build modern hyperlink inside table cell
+    r_id = "rIdHttp1"
+    hyperlink = OxmlElement("w:hyperlink")
+    hyperlink.set(qn("r:id"), r_id)
+
+    run_elem = OxmlElement("w:r")
+    t = OxmlElement("w:t")
+    t.text = "Dify"
+    run_elem.append(t)
+    hyperlink.append(run_elem)
+    p._p.append(hyperlink)
+
+    # Relationship for external http link
+    doc.part.rels.add_relationship(
+        "http://schemas.openxmlformats.org/officeDocument/2006/relationships/hyperlink",
+        "https://dify.ai",
+        r_id,
+        is_external=True,
+    )
+
+    with tempfile.NamedTemporaryFile(suffix=".docx", delete=False) as tmp:
+        doc.save(tmp.name)
+        tmp_path = tmp.name
+
+    try:
+        reopened = Document(tmp_path)
+        para = reopened.tables[0].cell(0, 0).paragraphs[0]
+        extractor = object.__new__(WordExtractor)
+        out = extractor._parse_cell_paragraph(para, {})
+        assert out == "[Dify](https://dify.ai)"
+    finally:
+        if os.path.exists(tmp_path):
+            os.remove(tmp_path)
+
+
+def test_parse_cell_paragraph_hyperlink_in_table_cell_mailto():
+    doc = Document()
+    table = doc.add_table(rows=1, cols=1)
+    cell = table.cell(0, 0)
+    p = cell.paragraphs[0]
+
+    r_id = "rIdMail1"
+    hyperlink = OxmlElement("w:hyperlink")
+    hyperlink.set(qn("r:id"), r_id)
+
+    run_elem = OxmlElement("w:r")
+    t = OxmlElement("w:t")
+    t.text = "john@test.com"
+    run_elem.append(t)
+    hyperlink.append(run_elem)
+    p._p.append(hyperlink)
+
+    doc.part.rels.add_relationship(
+        "http://schemas.openxmlformats.org/officeDocument/2006/relationships/hyperlink",
+        "mailto:john@test.com",
+        r_id,
+        is_external=True,
+    )
+
+    with tempfile.NamedTemporaryFile(suffix=".docx", delete=False) as tmp:
+        doc.save(tmp.name)
+        tmp_path = tmp.name
+
+    try:
+        reopened = Document(tmp_path)
+        para = reopened.tables[0].cell(0, 0).paragraphs[0]
+        extractor = object.__new__(WordExtractor)
+        out = extractor._parse_cell_paragraph(para, {})
+        assert out == "[john@test.com](mailto:john@test.com)"
+    finally:
+        if os.path.exists(tmp_path):
+            os.remove(tmp_path)