docling-haystack 0.2.0__tar.gz → 0.4.0__tar.gz

docling_haystack-0.4.0/CHANGELOG.md

@@ -0,0 +1,20 @@
+# Changelog
+
+## [integrations/docling-v0.3.0] - 2026-04-10
+
+### 🚀 Features
+
+- (docling) Drop temp files for ByteStream sources (#3130)
+
+
+## [integrations/docling-v0.2.0] - 2026-04-08
+
+### 🚀 Features
+
+- Add Docling document converter (#3066)
+
+### 🚜 Refactor
+
+- *(docling)* Add meta parameter to run(); introduce sources; deprecate paths (#3103)
+
+<!-- generated by git-cliff -->

{docling_haystack-0.2.0 → docling_haystack-0.4.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: docling-haystack
-Version: 0.2.0
+Version: 0.4.0
 Summary: Haystack integration for docling
 Project-URL: Documentation, https://github.com/deepset-ai/haystack-core-integrations/tree/main/integrations/docling#readme
 Project-URL: Issues, https://github.com/deepset-ai/haystack-core-integrations/issues

{docling_haystack-0.2.0 → docling_haystack-0.4.0}/src/haystack_integrations/components/converters/docling/converter.py

@@ -1,22 +1,50 @@
 """Docling Haystack converter module."""
 
 import json
-import os
-import tempfile
+import mimetypes
 import warnings
 from abc import ABC, abstractmethod
 from enum import Enum
+from io import BytesIO
 from pathlib import Path
 from typing import Any
 
-from haystack import Document, component
+from docling_core.types.io import DocumentStream
+from haystack import Document, component, logging
 from haystack.components.converters.utils import normalize_metadata
+from haystack.core.serialization import default_from_dict, default_to_dict
 from haystack.dataclasses import ByteStream
+from haystack.utils.base_serialization import deserialize_class_instance, serialize_class_instance
 
 from docling.chunking import BaseChunk, BaseChunker, HybridChunker
 from docling.datamodel.document import DoclingDocument
 from docling.document_converter import DocumentConverter
 
+logger = logging.getLogger(__name__)
+
+
+def _bytestream_to_document_stream(source: ByteStream) -> DocumentStream:
+    """
+    Build a `DocumentStream` from a Haystack `ByteStream`.
+
+    Resolves the stream name by checking common metadata keys (`file_path`, `file_name`, `name`) and falling back to
+    MIME-type extension guessing so that docling can reliably detect the input format.
+    """
+    meta = source.meta or {}
+    raw_name = meta.get("file_path") or meta.get("file_name") or meta.get("name")
+
+    if raw_name:
+        name = Path(raw_name).name
+    else:
+        name = "document"
+
+    if not Path(name).suffix and source.mime_type:
+        ext = mimetypes.guess_extension(source.mime_type)
+        if ext:
+            name = f"{name}{ext}"
+
+    return DocumentStream(name=name, stream=BytesIO(source.data))
+
 
 class ExportType(str, Enum):
     """Enumeration of available export types."""
@@ -39,6 +67,15 @@ class BaseMetaExtractor(ABC):
         """Extract Docling document meta."""
         raise NotImplementedError()
 
+    def to_dict(self) -> dict[str, Any]:
+        """Serialize to a dictionary."""
+        return {}
+
+    @classmethod
+    def from_dict(cls, data: dict[str, Any]) -> "BaseMetaExtractor":  # noqa: ARG003
+        """Deserialize from a dictionary."""
+        return cls()
+
 
 class MetaExtractor(BaseMetaExtractor):
     """MetaExtractor."""
@@ -99,6 +136,53 @@ class DoclingConverter:
             self._chunker_instance = chunker or HybridChunker()
         self._meta_extractor_instance = meta_extractor or MetaExtractor()
 
+    def to_dict(self) -> dict[str, Any]:
+        """Serialize this component to a dictionary."""
+        if self.converter is not None:
+            logger.warning(
+                "DoclingConverter.to_dict: the 'converter' parameter cannot be serialized and will be dropped. "
+                "The component will use the default DocumentConverter when restored from the serialized form."
+            )
+        if self.chunker is not None:
+            logger.warning(
+                "DoclingConverter.to_dict: the 'chunker' parameter cannot be serialized and will be dropped. "
+                "The component will use the default chunker when restored from the serialized form."
+            )
+
+        meta_extractor_data = None
+        if self.meta_extractor is not None:
+            meta_extractor_data = serialize_class_instance(self.meta_extractor)
+
+        return default_to_dict(
+            self,
+            converter=None,
+            convert_kwargs=self.convert_kwargs,
+            export_type=self.export_type.value,
+            md_export_kwargs=self.md_export_kwargs,
+            chunker=None,
+            meta_extractor=meta_extractor_data,
+        )
+
+    @classmethod
+    def from_dict(cls, data: dict[str, Any]) -> "DoclingConverter":
+        """
+        Deserialize this component from a dictionary.
+
+        The `converter` and `chunker` parameters are not serializable and are always ignored during
+        deserialization; the restored instance will use the default `DocumentConverter` and `HybridChunker`
+        respectively.
+
+        :param data: Dictionary with keys `type` and `init_parameters`, as produced by `to_dict`.
+        :returns: A new `DoclingConverter` instance.
+        """
+        init_params = data.get("init_parameters", {})
+
+        meta_extractor_data = init_params.get("meta_extractor")
+        if meta_extractor_data is not None:
+            init_params["meta_extractor"] = deserialize_class_instance(meta_extractor_data)
+
+        return default_from_dict(cls, data)
+
     @component.output_types(documents=list[Document])
     def run(
         self,
@@ -141,14 +225,8 @@ class DoclingConverter:
         documents: list[Document] = []
         for source, source_meta in zip(sources, meta_list, strict=True):
             if isinstance(source, ByteStream):
-                # docling requires a file path; write ByteStream data to a temp file
-                with tempfile.NamedTemporaryFile(delete=False) as tmp:
-                    tmp.write(source.data)
-                    tmp_path = Path(tmp.name)
-                try:
-                    dl_doc = self._converter_instance.convert(source=tmp_path, **self.convert_kwargs).document
-                finally:
-                    os.unlink(tmp_path)
+                doc_stream = _bytestream_to_document_stream(source)
+                dl_doc = self._converter_instance.convert(source=doc_stream, **self.convert_kwargs).document
                 # merge ByteStream meta (e.g. file_path, mime_type) with user-supplied meta
                 merged_meta = {**(source.meta or {}), **source_meta}
             else:

{docling_haystack-0.2.0 → docling_haystack-0.4.0}/tests/test_converter.py

@@ -1,14 +1,23 @@
 import json
+import mimetypes
 import warnings
+from io import BytesIO
 from types import SimpleNamespace
 from typing import Any
-from unittest.mock import MagicMock, patch
+from unittest.mock import MagicMock
 
 import pytest
-from haystack.core.serialization import component_from_dict, component_to_dict
+from docling.chunking import HybridChunker
+from docling.document_converter import DocumentConverter
+from docling_core.types.io import DocumentStream
 from haystack.dataclasses import ByteStream
 
-from haystack_integrations.components.converters.docling import DoclingConverter, ExportType
+from haystack_integrations.components.converters.docling import (
+    DoclingConverter,
+    ExportType,
+    MetaExtractor,
+)
+from haystack_integrations.components.converters.docling.converter import _bytestream_to_document_stream
 
 
 def test_run_doc_chunks_minimal() -> None:
@@ -126,8 +135,6 @@ def test_run_json_minimal() -> None:
 
 
 def test_legacy_import_path() -> None:
-    import warnings
-
     with warnings.catch_warnings(record=True) as caught:
         warnings.simplefilter("always")
         from docling_haystack.converter import DoclingConverter as LegacyDoclingConverter
@@ -138,63 +145,59 @@ def test_legacy_import_path() -> None:
     )
 
 
-def test_component_from_dict_legacy_nulls() -> None:
-    # Before the public-attribute refactor, default serialization couldn't find
-    # the _-prefixed attributes and fell back to the init defaults, so
-    # convert_kwargs and md_export_kwargs were always serialized as null.
-    # Verify that such a serialized dict still deserializes correctly.
-    legacy_data = {
+def test_component_to_dict_defaults() -> None:
+    converter = DoclingConverter()
+    assert converter.to_dict() == {
         "type": "haystack_integrations.components.converters.docling.converter.DoclingConverter",
         "init_parameters": {
             "converter": None,
-            "convert_kwargs": None,
+            "convert_kwargs": {},
             "export_type": "doc_chunks",
-            "md_export_kwargs": None,
+            "md_export_kwargs": {"image_placeholder": ""},
             "chunker": None,
             "meta_extractor": None,
         },
     }
-    restored = component_from_dict(DoclingConverter, legacy_data, "docling_converter")
-
-    assert restored.convert_kwargs == {}
-    assert restored.md_export_kwargs == {"image_placeholder": ""}
-    assert restored.export_type == ExportType.DOC_CHUNKS
-    assert restored.converter is None
-    assert restored.chunker is None
-    assert restored.meta_extractor is None
-
-
-def test_component_to_dict_defaults() -> None:
-    converter = DoclingConverter()
-    data = component_to_dict(converter, "docling_converter")
-
-    init_params = data["init_parameters"]
-    assert init_params["converter"] is None
-    assert init_params["convert_kwargs"] == {}
-    assert init_params["export_type"] == ExportType.DOC_CHUNKS
-    assert init_params["md_export_kwargs"] == {"image_placeholder": ""}
-    assert init_params["chunker"] is None
-    assert init_params["meta_extractor"] is None
 
 
 def test_component_to_dict_custom_params() -> None:
     converter = DoclingConverter(
+        converter=DocumentConverter(),
         convert_kwargs={"raises_on_error": False},
         export_type=ExportType.MARKDOWN,
         md_export_kwargs={"image_placeholder": "[img]"},
+        meta_extractor=MetaExtractor(),
     )
-    data = component_to_dict(converter, "docling_converter")
-
-    init_params = data["init_parameters"]
-    assert init_params["convert_kwargs"] == {"raises_on_error": False}
-    assert init_params["export_type"] == ExportType.MARKDOWN
-    assert init_params["md_export_kwargs"] == {"image_placeholder": "[img]"}
+    assert converter.to_dict() == {
+        "type": "haystack_integrations.components.converters.docling.converter.DoclingConverter",
+        "init_parameters": {
+            "converter": None,
+            "convert_kwargs": {"raises_on_error": False},
+            "export_type": "markdown",
+            "md_export_kwargs": {"image_placeholder": "[img]"},
+            "chunker": None,
+            "meta_extractor": {
+                "type": "haystack_integrations.components.converters.docling.converter.MetaExtractor",
+                "data": {},
+            },
+        },
+    }
 
 
 def test_component_from_dict_defaults() -> None:
-    converter = DoclingConverter()
-    data = component_to_dict(converter, "docling_converter")
-    restored = component_from_dict(DoclingConverter, data, "docling_converter")
+    # null kwargs mirror the pre-refactor serialization format and must still deserialize correctly
+    data = {
+        "type": "haystack_integrations.components.converters.docling.converter.DoclingConverter",
+        "init_parameters": {
+            "converter": None,
+            "convert_kwargs": None,
+            "export_type": "doc_chunks",
+            "md_export_kwargs": None,
+            "chunker": None,
+            "meta_extractor": None,
+        },
+    }
+    restored = DoclingConverter.from_dict(data)
 
     assert restored.converter is None
     assert restored.convert_kwargs == {}
@@ -205,17 +208,44 @@ def test_component_from_dict_defaults() -> None:
 
 
 def test_component_from_dict_custom_params() -> None:
-    converter = DoclingConverter(
-        convert_kwargs={"raises_on_error": False},
-        export_type=ExportType.JSON,
-        md_export_kwargs={"image_placeholder": "[img]"},
-    )
-    data = component_to_dict(converter, "docling_converter")
-    restored = component_from_dict(DoclingConverter, data, "docling_converter")
+    data = {
+        "type": "haystack_integrations.components.converters.docling.converter.DoclingConverter",
+        "init_parameters": {
+            "converter": None,
+            "convert_kwargs": {"raises_on_error": False},
+            "export_type": "json",
+            "md_export_kwargs": {"image_placeholder": "[img]"},
+            "chunker": None,
+            "meta_extractor": {
+                "type": "haystack_integrations.components.converters.docling.converter.MetaExtractor",
+                "data": {},
+            },
+        },
+    }
+    restored = DoclingConverter.from_dict(data)
 
+    assert restored.converter is None
     assert restored.convert_kwargs == {"raises_on_error": False}
     assert restored.export_type == ExportType.JSON
     assert restored.md_export_kwargs == {"image_placeholder": "[img]"}
+    assert restored.chunker is None
+    assert isinstance(restored.meta_extractor, MetaExtractor)
+
+
+def test_component_to_dict_chunker_warns_and_is_dropped() -> None:
+    converter = DoclingConverter(chunker=HybridChunker(merge_peers=False))
+
+    assert converter.to_dict() == {
+        "type": "haystack_integrations.components.converters.docling.converter.DoclingConverter",
+        "init_parameters": {
+            "converter": None,
+            "convert_kwargs": {},
+            "export_type": "doc_chunks",
+            "md_export_kwargs": {"image_placeholder": ""},
+            "chunker": None,
+            "meta_extractor": None,
+        },
+    }
 
 
 def test_run_with_sources_parameter() -> None:
@@ -356,13 +386,129 @@ def test_run_with_bytestream_source() -> None:
 
     bytestream = ByteStream(data=b"%PDF-1.4 fake pdf content", meta={"file_path": "uploaded.pdf"})
 
-    with patch("os.unlink"):
-        result = converter.run(sources=[bytestream])
+    result = converter.run(sources=[bytestream])
 
     documents = result["documents"]
     assert len(documents) == 1
     # ByteStream meta is merged into the output document
     assert documents[0].meta["file_path"] == "uploaded.pdf"
-    # docling was called with a temp file path, not the ByteStream directly
+    # docling was called with a DocumentStream, not a temp file path
     call_args = converter_mock.convert.call_args
-    assert call_args.kwargs["source"] != bytestream
+    passed_source = call_args.kwargs["source"]
+    assert isinstance(passed_source, DocumentStream)
+    assert passed_source.name == "uploaded.pdf"
+    assert isinstance(passed_source.stream, BytesIO)
+
+
+class TestBytestreamToDocumentStream:
+    def test_uses_file_path(self) -> None:
+        bs = ByteStream(data=b"data", meta={"file_path": "report.pdf"})
+        ds = _bytestream_to_document_stream(bs)
+        assert ds.name == "report.pdf"
+        assert ds.stream.read() == b"data"
+
+    def test_strips_directory_from_file_path(self) -> None:
+        bs = ByteStream(data=b"data", meta={"file_path": "/some/deep/path/report.pdf"})
+        ds = _bytestream_to_document_stream(bs)
+        assert ds.name == "report.pdf"
+
+    def test_uses_file_name_key(self) -> None:
+        bs = ByteStream(data=b"data", meta={"file_name": "slide-deck.pptx"})
+        ds = _bytestream_to_document_stream(bs)
+        assert ds.name == "slide-deck.pptx"
+
+    def test_uses_name_key(self) -> None:
+        bs = ByteStream(data=b"data", meta={"name": "notes.docx"})
+        ds = _bytestream_to_document_stream(bs)
+        assert ds.name == "notes.docx"
+
+    def test_file_path_takes_priority_over_file_name(self) -> None:
+        bs = ByteStream(data=b"data", meta={"file_path": "real.pdf", "file_name": "other.pdf"})
+        ds = _bytestream_to_document_stream(bs)
+        assert ds.name == "real.pdf"
+
+    def test_file_name_takes_priority_over_name(self) -> None:
+        bs = ByteStream(data=b"data", meta={"file_name": "chosen.pdf", "name": "ignored.pdf"})
+        ds = _bytestream_to_document_stream(bs)
+        assert ds.name == "chosen.pdf"
+
+    def test_guesses_extension_from_mime_type(self) -> None:
+        mime = "application/pdf"
+        expected_ext = mimetypes.guess_extension(mime)
+        bs = ByteStream(data=b"data", meta={"file_path": "report"}, mime_type=mime)
+        ds = _bytestream_to_document_stream(bs)
+        assert ds.name == f"report{expected_ext}"
+
+    def test_keeps_extension_when_present(self) -> None:
+        # mime_type should not override an already-present extension
+        bs = ByteStream(data=b"data", meta={"file_path": "report.pdf"}, mime_type="text/plain")
+        ds = _bytestream_to_document_stream(bs)
+        assert ds.name == "report.pdf"
+
+    def test_no_meta_no_mime_type(self) -> None:
+        bs = ByteStream(data=b"data")
+        ds = _bytestream_to_document_stream(bs)
+        assert ds.name == "document"
+
+    def test_no_meta_with_mime_type(self) -> None:
+        mime = "application/pdf"
+        expected_ext = mimetypes.guess_extension(mime)
+        bs = ByteStream(data=b"data", mime_type=mime)
+        ds = _bytestream_to_document_stream(bs)
+        assert ds.name == f"document{expected_ext}"
+
+    def test_empty_meta_no_mime_type(self) -> None:
+        bs = ByteStream(data=b"data", meta={})
+        ds = _bytestream_to_document_stream(bs)
+        assert ds.name == "document"
+
+    def test_returns_document_stream_with_bytesio(self) -> None:
+        bs = ByteStream(data=b"hello", meta={"file_path": "f.pdf"})
+        ds = _bytestream_to_document_stream(bs)
+        assert isinstance(ds, DocumentStream)
+        assert isinstance(ds.stream, BytesIO)
+
+    def test_unknown_mime_type_keeps_base_name(self) -> None:
+        # mimetypes.guess_extension returns None for unknown types, so the name stays as-is.
+        assert mimetypes.guess_extension("application/x-totally-made-up-type") is None
+        bs = ByteStream(
+            data=b"data",
+            meta={"file_path": "report"},
+            mime_type="application/x-totally-made-up-type",
+        )
+        ds = _bytestream_to_document_stream(bs)
+        assert ds.name == "report"
+
+
+class TestMetaExtractor:
+    def test_extract_chunk_meta_wraps_export_json_dict(self) -> None:
+        chunk = MagicMock()
+        chunk.export_json_dict.return_value = {"some": "dict"}
+
+        result = MetaExtractor().extract_chunk_meta(chunk=chunk)
+
+        assert result == {"dl_meta": {"some": "dict"}}
+        chunk.export_json_dict.assert_called_once_with()
+
+    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"}
+
+        result = MetaExtractor().extract_dl_doc_meta(dl_doc=dl_doc)
+
+        assert result == {"dl_meta": {"origin": {"filename": "foo.pdf", "mimetype": "application/pdf"}}}
+        dl_doc.origin.model_dump.assert_called_once_with(exclude_none=True)
+
+    def test_extract_dl_doc_meta_without_origin(self) -> None:
+        dl_doc = MagicMock()
+        dl_doc.origin = None
+
+        result = MetaExtractor().extract_dl_doc_meta(dl_doc=dl_doc)
+
+        assert result == {}
+
+
+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()