fix(docling)!: change default export type to MARKDOWN and add page_number to chunk metadata

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

@@ -82,7 +82,12 @@ class MetaExtractor(BaseMetaExtractor):
 
     def extract_chunk_meta(self, chunk: BaseChunk) -> dict[str, Any]:
         """Extract chunk meta."""
-        return {"dl_meta": chunk.export_json_dict()}
+        meta: dict[str, Any] = {"dl_meta": chunk.export_json_dict()}
+        doc_items = getattr(chunk.meta, "doc_items", [])
+        page_nos = {prov.page_no for item in doc_items for prov in getattr(item, "prov", [])}
+        if page_nos:
+            meta["page_number"] = min(page_nos)
+        return meta
 
     def extract_dl_doc_meta(self, dl_doc: DoclingDocument) -> dict[str, Any]:
         """Extract Docling document meta."""
@@ -97,7 +102,7 @@ def __init__(
         self,
         converter: DocumentConverter | None = None,
         convert_kwargs: dict[str, Any] | None = None,
-        export_type: ExportType = ExportType.DOC_CHUNKS,
+        export_type: ExportType = ExportType.MARKDOWN,
         md_export_kwargs: dict[str, Any] | None = None,
         chunker: BaseChunker | None = None,
         meta_extractor: BaseMetaExtractor | None = None,
@@ -110,10 +115,10 @@ def __init__(
         :param convert_kwargs: Any parameters to pass to Docling conversion; if not set, a
             system default is used.
         :param export_type: The export mode to use:
-            * `ExportType.MARKDOWN` captures each input document as a single
+            * `ExportType.MARKDOWN` (default) captures each input document as a single
               markdown `Document`.
-            * `ExportType.DOC_CHUNKS` (default) first chunks each input document
-              and then returns one `Document` per chunk.
+            * `ExportType.DOC_CHUNKS` first chunks each input document and then returns
+              one `Document` per chunk.
             * `ExportType.JSON` serializes the full Docling document to a JSON string.
         :param md_export_kwargs: Any parameters to pass to Markdown export (applicable in
             case of `ExportType.MARKDOWN`).
@@ -234,15 +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)
@@ -152,7 +152,7 @@ def test_component_to_dict_defaults() -> None:
         "init_parameters": {
             "converter": None,
             "convert_kwargs": {},
-            "export_type": "doc_chunks",
+            "export_type": "markdown",
             "md_export_kwargs": {"image_placeholder": ""},
             "chunker": None,
             "meta_extractor": None,
@@ -233,7 +233,7 @@ def test_component_from_dict_custom_params() -> None:
 
 
 def test_component_to_dict_chunker_warns_and_is_dropped() -> None:
-    converter = DoclingConverter(chunker=HybridChunker(merge_peers=False))
+    converter = DoclingConverter(export_type=ExportType.DOC_CHUNKS, chunker=HybridChunker(merge_peers=False))
 
     assert converter.to_dict() == {
         "type": "haystack_integrations.components.converters.docling.converter.DoclingConverter",
@@ -484,12 +484,43 @@ class TestMetaExtractor:
     def test_extract_chunk_meta_wraps_export_json_dict(self) -> None:
         chunk = MagicMock()
         chunk.export_json_dict.return_value = {"some": "dict"}
+        chunk.meta.doc_items = []
 
         result = MetaExtractor().extract_chunk_meta(chunk=chunk)
 
         assert result == {"dl_meta": {"some": "dict"}}
         chunk.export_json_dict.assert_called_once_with()
 
+    def test_extract_chunk_meta_includes_page_number(self) -> None:
+        prov = MagicMock()
+        prov.page_no = 3
+        doc_item = MagicMock()
+        doc_item.prov = [prov]
+
+        chunk = MagicMock()
+        chunk.export_json_dict.return_value = {"some": "dict"}
+        chunk.meta.doc_items = [doc_item]
+
+        result = MetaExtractor().extract_chunk_meta(chunk=chunk)
+
+        assert result == {"dl_meta": {"some": "dict"}, "page_number": 3}
+
+    def test_extract_chunk_meta_page_number_uses_minimum(self) -> None:
+        prov1 = MagicMock()
+        prov1.page_no = 5
+        prov2 = MagicMock()
+        prov2.page_no = 3
+        doc_item = MagicMock()
+        doc_item.prov = [prov1, prov2]
+
+        chunk = MagicMock()
+        chunk.export_json_dict.return_value = {}
+        chunk.meta.doc_items = [doc_item]
+
+        result = MetaExtractor().extract_chunk_meta(chunk=chunk)
+
+        assert result["page_number"] == 3
+
     def test_extract_dl_doc_meta_with_origin(self) -> None:
         dl_doc = MagicMock()
         dl_doc.origin.model_dump.return_value = {"filename": "foo.pdf", "mimetype": "application/pdf"}
@@ -512,3 +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")