From 23225e7f2859663e89fe6fa6d5c7414a15dc171d Mon Sep 17 00:00:00 2001
From: Christopher Krah <48338417+chrikrah@users.noreply.github.com>
Date: Fri, 18 Sep 2026 06:50:39 +0000
Subject: [PATCH 1/2] fix(jats): stop rendering the literal text None in figure
 and table captions

Two places write None into text a reader sees.

_add_figure_captions sets caption = None when a <fig> has a <label> and no
<caption>, which is legal and common for a numbered but uncaptioned figure. The
f-string at :917 then renders the four characters None straight after the label:

  <fig><label>Figure 1.</label><graphic/></fig>  ->  'Figure 1.None'

_add_tables reads table["label"] = node.xpath("label")[0].text without a
guard. An empty <label/> has .text of None in lxml, so a table with neither a
real label nor a caption still produces a caption item reading None:

  <table-wrap><label/><table>...</table></table-wrap>  ->  'None'

Both now keep the empty-string default the surrounding code already uses, the
same way :634 reads etal_node[0].text or DEFAULT_TEXT_ETAL.

Tests: 59 passed in tests/test_backend_jats.py. Two new cases; reverting
jats_backend.py alone fails both. No groundtruth file changes, and no committed
JATS groundtruth contains the string None today, so no current fixture reaches
either path.

Signed-off-by: Christopher Krah <48338417+chrikrah@users.noreply.github.com>
---
 docling/backend/xml/jats_backend.py | 12 +++++++-----
 tests/test_backend_jats.py          | 30 +++++++++++++++++++++++++++++
 2 files changed, 37 insertions(+), 5 deletions(-)

diff --git a/docling/backend/xml/jats_backend.py b/docling/backend/xml/jats_backend.py
index 3a2db8764d..d3497216ea 100755
--- a/docling/backend/xml/jats_backend.py
+++ b/docling/backend/xml/jats_backend.py
@@ -908,16 +908,16 @@ def _add_figure_captions(
         )
 
         caption_node = node.xpath("caption")
-        caption: str | None
+        # A <fig> may carry a <label> and no <caption>, which is common for a
+        # numbered but uncaptioned figure. Default to "" rather than None, or
+        # the f-string below renders the four characters None after the label.
+        caption: str = ""
         if len(caption_node) > 0:
-            caption = ""
             for caption_par in list(caption_node[0]):
                 if caption_par.xpath(".//supplementary-material"):
                     continue
                 caption += JatsDocumentBackend._get_text(caption_par).strip() + " "
             caption = caption.strip()
-        else:
-            caption = None
 
         # TODO: format label vs caption once styling is supported
         fig_text: str = f"{label}{' ' if label and caption else ''}{caption}"
@@ -1178,7 +1178,9 @@ def _add_tables(
 
         # Label
         if len(node.xpath("label")) > 0:
-            table["label"] = node.xpath("label")[0].text
+            # An empty <label/> has .text of None in lxml. Keep the "" default
+            # so it does not reach the rendered caption as the text None.
+            table["label"] = node.xpath("label")[0].text or ""
 
         try:
             self._add_table(doc, parent, table)
diff --git a/tests/test_backend_jats.py b/tests/test_backend_jats.py
index 0c6068b39c..205d8f517c 100644
--- a/tests/test_backend_jats.py
+++ b/tests/test_backend_jats.py
@@ -1104,3 +1104,33 @@ def test_jats_empty_article_title_does_not_crash():
     exported = doc.export_to_markdown()
     # Empty title is serialized as an H1 with no text.
     assert exported == "# "
+
+
+def _caption_texts(doc: DoclingDocument) -> list[str]:
+    return [
+        t.text for t in doc.texts if str(getattr(t, "label", "")).endswith("caption")
+    ]
+
+
+def test_jats_labelled_figure_without_a_caption_omits_the_word_none():
+    """A <fig> may carry a <label> and no <caption>, which is legal and common."""
+    doc = convert_jats_body(
+        """<sec><p>Text.</p>
+        <fig id="f1"><label>Figure 1.</label><graphic/></fig>
+        </sec>"""
+    )
+
+    assert _caption_texts(doc) == ["Figure 1."]
+
+
+def test_jats_table_with_an_empty_label_produces_no_caption():
+    """An empty <label/> has .text of None in lxml."""
+    doc = convert_jats_body(
+        """<sec><p>Text.</p>
+        <table-wrap id="t1"><label/>
+          <table><tbody><tr><td>A</td><td>B</td></tr></tbody></table>
+        </table-wrap>
+        </sec>"""
+    )
+
+    assert _caption_texts(doc) == []

From dab5b1494061afc1f5b05f74eb3b068450555d49 Mon Sep 17 00:00:00 2001
From: Christopher Krah <48338417+chrikrah@users.noreply.github.com>
Date: Sat, 26 Sep 2026 09:05:16 +0000
Subject: [PATCH 2/2] style(jats): apply the review comments from #4291

Three points from ceberam, all about how the change reads:

- The comment before `caption: str = ""` in `_add_figure_captions` said what
  the annotation already says, so it is gone.
- The comment in `_add_tables` is down to the one line that carries the fact
  a reader cannot see: lxml sets `.text` to None for an empty element.
- `_caption_texts` in the test compares `t.label` to `DocItemLabel.CAPTION`
  directly. The `getattr` and the string suffix match were defensive for no
  reason; `DocItemLabel` was already imported.

No behaviour change. 59 passed in tests/test_backend_jats.py, and reverting
jats_backend.py alone to 815f34ab still fails both new cases (2 failed,
57 passed), so the direct comparison pins the fix as well as the old helper.

Signed-off-by: Christopher Krah <48338417+chrikrah@users.noreply.github.com>
---
 docling/backend/xml/jats_backend.py | 6 +-----
 tests/test_backend_jats.py          | 4 +---
 2 files changed, 2 insertions(+), 8 deletions(-)

diff --git a/docling/backend/xml/jats_backend.py b/docling/backend/xml/jats_backend.py
index d3497216ea..99aae88fb2 100755
--- a/docling/backend/xml/jats_backend.py
+++ b/docling/backend/xml/jats_backend.py
@@ -908,9 +908,6 @@ def _add_figure_captions(
         )
 
         caption_node = node.xpath("caption")
-        # A <fig> may carry a <label> and no <caption>, which is common for a
-        # numbered but uncaptioned figure. Default to "" rather than None, or
-        # the f-string below renders the four characters None after the label.
         caption: str = ""
         if len(caption_node) > 0:
             for caption_par in list(caption_node[0]):
@@ -1178,8 +1175,7 @@ def _add_tables(
 
         # Label
         if len(node.xpath("label")) > 0:
-            # An empty <label/> has .text of None in lxml. Keep the "" default
-            # so it does not reach the rendered caption as the text None.
+            # lxml sets .text to None for an empty element; fall back to "".
             table["label"] = node.xpath("label")[0].text or ""
 
         try:
diff --git a/tests/test_backend_jats.py b/tests/test_backend_jats.py
index 205d8f517c..dfa89e9920 100644
--- a/tests/test_backend_jats.py
+++ b/tests/test_backend_jats.py
@@ -1107,9 +1107,7 @@ def test_jats_empty_article_title_does_not_crash():
 
 
 def _caption_texts(doc: DoclingDocument) -> list[str]:
-    return [
-        t.text for t in doc.texts if str(getattr(t, "label", "")).endswith("caption")
-    ]
+    return [t.text for t in doc.texts if t.label == DocItemLabel.CAPTION]
 
 
 def test_jats_labelled_figure_without_a_caption_omits_the_word_none():