fix(table): row-tokenizer chunks must start at index 0 (header was left uncovered)

src/chonkie/chunker/table.py

@@ -144,24 +144,34 @@ def chunk(self, text: str) -> list[Chunk]:
                     ),
                 ]
             else:
-                # Track character position for data rows (after header)
-                header_len = len(header)
-                current_char_index = header_len
+                # Track character position over the original, unstripped input.
+                # The first span absorbs any prefix before the parsed header, and
+                # the final span absorbs the footer and trailing whitespace so the
+                # spans remain contiguous over the complete table text.
+                header_start = text.find(header)
+                header_span_len = len(header) + (header_start if header_start != -1 else 0)
+                current_char_index = 0
 
                 for i in range(0, len(data_rows), self.chunk_size):
                     chunk_rows = data_rows[i : i + self.chunk_size]
-                    chunk_text = header + "".join(chunk_rows) + footer
-                    data_rows_len = len("".join(chunk_rows))
+                    chunk_rows_text = "".join(chunk_rows)
+                    chunk_text = header + chunk_rows_text + footer
+                    is_last_chunk = i + self.chunk_size >= len(data_rows)
+                    span_len = (
+                        len(text) - current_char_index
+                        if is_last_chunk
+                        else len(chunk_rows_text) + (header_span_len if i == 0 else 0)
+                    )
 
                     chunks.append(
                         Chunk(
                             text=chunk_text,
                             token_count=len(chunk_rows),
                             start_index=current_char_index,
-                            end_index=current_char_index + data_rows_len,
+                            end_index=current_char_index + span_len,
                         ),
                     )
-                    current_char_index += data_rows_len
+                    current_char_index += span_len
 
             return chunks
 

tests/chunkers/test_table_chunker.py

@@ -861,3 +861,35 @@ def test_table_chunker_markdown_document_no_tables_preserves_chunks() -> None:
 
     assert len(result.chunks) == 1
     assert result.chunks[0].text == "Some prose text here."
+
+
+def test_table_chunker_row_tokenizer_indices():
+    table = "| A | B |\n|---|---|\n| 1 | 2 |\n| 3 | 4 |\n| 5 | 6 |\n| 7 | 8 |"
+    chunker = TableChunker(tokenizer="row", chunk_size=2)
+    chunks = chunker.chunk(table)
+    assert len(chunks) > 1
+    # First chunk must start at 0 (matches the character path and every other chunker).
+    assert chunks[0].start_index == 0
+    # Spans are contiguous and reconstruct the full table (no lost/duplicated regions).
+    for i in range(len(chunks) - 1):
+        assert chunks[i].end_index == chunks[i + 1].start_index
+    assert chunks[-1].end_index == len(table)
+    assert "".join(table[c.start_index : c.end_index] for c in chunks) == table
+
+
+def test_table_chunker_row_tokenizer_html_indices_cover_full_input() -> None:
+    table = (
+        "  <table><thead><tr><th>A</th></tr></thead><tbody>"
+        "<tr><td>1</td></tr><tr><td>2</td></tr><tr><td>3</td></tr>"
+        "</tbody></table>\n"
+    )
+    chunker = TableChunker(tokenizer="row", chunk_size=1)
+
+    chunks = chunker.chunk(table)
+
+    assert len(chunks) == 3
+    assert chunks[0].start_index == 0
+    for current, following in zip(chunks, chunks[1:]):
+        assert current.end_index == following.start_index
+    assert chunks[-1].end_index == len(table)
+    assert "".join(table[chunk.start_index : chunk.end_index] for chunk in chunks) == table