fix(docling): add split_id and split_idx_start to chunk metadata; revert CHANGELOG

Author: SyedShahmeerAli12

integrations/docling/CHANGELOG.md

@@ -1,15 +1,5 @@
 # Changelog
 
-## [Unreleased]
-
-### 🚀 Features
-
-- Make `ExportType.MARKDOWN` the default export type; add `page_number` to chunk metadata (#3256)
-
-### ⚠ Breaking Changes
-
-- `DoclingConverter` now defaults to `ExportType.MARKDOWN` instead of `ExportType.DOC_CHUNKS`
-
 ## [integrations/docling-v0.4.0] - 2026-05-04
 
 ### 🚀 Features

integrations/docling/src/haystack_integrations/components/converters/docling/converter.py

@@ -239,15 +239,17 @@ def run(
                 merged_meta = source_meta
 
             if self.export_type == ExportType.DOC_CHUNKS:
-                chunk_iter = self._chunker_instance.chunk(dl_doc=dl_doc)
-                hs_docs = [
-                    Document(
-                        content=self._chunker_instance.contextualize(chunk=chunk),
-                        meta={**self._meta_extractor_instance.extract_chunk_meta(chunk=chunk), **merged_meta},
-                    )
-                    for chunk in chunk_iter
-                ]
-                documents.extend(hs_docs)
+                split_idx_start = 0
+                for split_id, chunk in enumerate(self._chunker_instance.chunk(dl_doc=dl_doc)):
+                    content = self._chunker_instance.contextualize(chunk=chunk)
+                    meta = {
+                        **self._meta_extractor_instance.extract_chunk_meta(chunk=chunk),
+                        "split_id": split_id,
+                        "split_idx_start": split_idx_start,
+                        **merged_meta,
+                    }
+                    documents.append(Document(content=content, meta=meta))
+                    split_idx_start += len(chunk.text)
             elif self.export_type == ExportType.MARKDOWN:
                 hs_doc = Document(
                     content=dl_doc.export_to_markdown(**self.md_export_kwargs),

integrations/docling/tests/test_converter.py

@@ -59,7 +59,7 @@ def chunk_side_effect(dl_doc: Any) -> list[SimpleNamespace]:
 
     assert "contextualized-chunk-1-of-dl-doc-for-file-a.pdf" in contents
     assert "contextualized-chunk-2-of-dl-doc-for-file-a.pdf" in contents
-    assert {"chunk_id": "chunk-1-of-dl-doc-for-file-a.pdf"} in metas
+    assert any(m.get("chunk_id") == "chunk-1-of-dl-doc-for-file-a.pdf" for m in metas)
 
     # Ensure our collaborators were actually exercised.
     assert converter_mock.convert.call_count == len(paths)
@@ -543,3 +543,76 @@ def test_run_without_sources_or_paths_raises_value_error() -> None:
     converter = DoclingConverter(converter=MagicMock(), meta_extractor=MagicMock())
     with pytest.raises(ValueError, match=r"Either 'sources' or the deprecated 'paths' parameter must be provided."):
         converter.run()
+
+
+def test_run_doc_chunks_split_id_and_split_idx_start() -> None:
+    converter_mock = MagicMock()
+    chunker_mock = MagicMock()
+    meta_extractor_mock = MagicMock()
+
+    converter_mock.convert.return_value = SimpleNamespace(document="dl-doc")
+
+    chunks = [
+        SimpleNamespace(text="hello world"),
+        SimpleNamespace(text="foo bar baz"),
+    ]
+    chunker_mock.chunk.return_value = chunks
+    chunker_mock.contextualize.side_effect = lambda chunk: f"ctx:{chunk.text}"
+    meta_extractor_mock.extract_chunk_meta.return_value = {}
+
+    converter = DoclingConverter(
+        converter=converter_mock,
+        export_type=ExportType.DOC_CHUNKS,
+        chunker=chunker_mock,
+        meta_extractor=meta_extractor_mock,
+    )
+
+    result = converter.run(sources=["doc.pdf"])
+    documents = result["documents"]
+
+    assert len(documents) == 2
+    assert documents[0].meta["split_id"] == 0
+    assert documents[0].meta["split_idx_start"] == 0
+    assert documents[1].meta["split_id"] == 1
+    assert documents[1].meta["split_idx_start"] == len("hello world")
+
+
+def test_run_doc_chunks_split_id_resets_per_document() -> None:
+    converter_mock = MagicMock()
+    chunker_mock = MagicMock()
+    meta_extractor_mock = MagicMock()
+
+    converter_mock.convert.side_effect = [
+        SimpleNamespace(document="dl-doc-a"),
+        SimpleNamespace(document="dl-doc-b"),
+    ]
+    chunker_mock.chunk.side_effect = lambda dl_doc: [
+        SimpleNamespace(text=f"chunk-1-of-{dl_doc}"),
+        SimpleNamespace(text=f"chunk-2-of-{dl_doc}"),
+    ]
+    chunker_mock.contextualize.side_effect = lambda chunk: chunk.text
+    meta_extractor_mock.extract_chunk_meta.return_value = {}
+
+    converter = DoclingConverter(
+        converter=converter_mock,
+        export_type=ExportType.DOC_CHUNKS,
+        chunker=chunker_mock,
+        meta_extractor=meta_extractor_mock,
+    )
+
+    result = converter.run(sources=["a.pdf", "b.pdf"])
+    documents = result["documents"]
+
+    # split_id and split_idx_start reset for each source document
+    doc_a_chunks = documents[:2]
+    doc_b_chunks = documents[2:]
+
+    assert doc_a_chunks[0].meta["split_id"] == 0
+    assert doc_a_chunks[0].meta["split_idx_start"] == 0
+    assert doc_a_chunks[1].meta["split_id"] == 1
+    assert doc_a_chunks[1].meta["split_idx_start"] == len("chunk-1-of-dl-doc-a")
+
+    assert doc_b_chunks[0].meta["split_id"] == 0
+    assert doc_b_chunks[0].meta["split_idx_start"] == 0
+    assert doc_b_chunks[1].meta["split_id"] == 1
+    assert doc_b_chunks[1].meta["split_idx_start"] == len("chunk-1-of-dl-doc-b")