From 4f6b2ffdea5817f5ccc3729444bc1ee5a265b8b0 Mon Sep 17 00:00:00 2001
From: chuenchen309 <48723787+chuenchen309@users.noreply.github.com>
Date: Sun, 19 Jul 2026 03:16:56 +0800
Subject: [PATCH] fix(html): stop a header-only rowspan table from crashing the
 backend

An HTML table whose only row is a `th` with `rowspan` (no body rows for it to
span into) raised `IndexError`. `get_html_table_row_col` counts no rows for an
all-header-rowspan row, so `num_rows` is 0 and the grid is empty; the
cell-placement `while` loop then reads `grid[row_idx + start_row_span][col_idx]`
without a row bound and goes out of range. The write just below it already
guards `row_idx + r < num_rows`.

Add the same row bound to the read. The cell is kept and the whole document no
longer aborts on this table.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
---
 docling/backend/html_backend.py |  1 +
 tests/test_backend_html.py      | 18 ++++++++++++++++++
 2 files changed, 19 insertions(+)

diff --git a/docling/backend/html_backend.py b/docling/backend/html_backend.py
index 42162b429d..35177511bc 100644
--- a/docling/backend/html_backend.py
+++ b/docling/backend/html_backend.py
@@ -1522,6 +1522,7 @@ def parse_table_data(
                     row_span -= 1
                 while (
                     col_idx < num_cols
+                    and row_idx + start_row_span < num_rows
                     and grid[row_idx + start_row_span][col_idx] is not None
                 ):
                     col_idx += 1
diff --git a/tests/test_backend_html.py b/tests/test_backend_html.py
index d51ba88d29..fc0141cd15 100644
--- a/tests/test_backend_html.py
+++ b/tests/test_backend_html.py
@@ -168,6 +168,24 @@ def test_heading_levels():
     assert found_lvl_1 and found_lvl_2
 
 
+def test_table_header_rowspan_without_body_does_not_crash():
+    # A table whose only row is a `th` with rowspan (no body rows to span into)
+    # used to raise IndexError: get_html_table_row_col counts no rows for an
+    # all-header-rowspan row, so the grid was empty and the cell-placement read
+    # went out of bounds. It should not crash and should keep the cell.
+    src = b"<table><tr><th rowspan='2'>h</th></tr></table>"
+    in_doc = InputDocument(
+        path_or_stream=BytesIO(src),
+        format=InputFormat.HTML,
+        backend=HTMLDocumentBackend,
+        filename="t.html",
+    )
+    doc = HTMLDocumentBackend(in_doc=in_doc, path_or_stream=BytesIO(src)).convert()
+
+    assert len(doc.tables) == 1
+    assert [cell.text for cell in doc.tables[0].data.table_cells] == ["h"]
+
+
 def test_ordered_lists():
     test_set: list[tuple[bytes, str]] = []