fix: fix kreuzberg for 4.7 breaking changes (#3110)

* fix: fix kreuzberg for 4.7 breaking changes

* simplify

Author: anakin87

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

@@ -37,6 +37,7 @@
     _is_batch_error,
     _serialize_annotations,
     _serialize_images,
+    _serialize_keywords,
     _serialize_warnings,
 )
 
@@ -266,13 +267,15 @@ def _build_extraction_metadata(result: ExtractionResult) -> dict[str, Any]:
         Build metadata dict from an `ExtractionResult`.
 
         Flattens kreuzberg's metadata fields and enriches with top-level result attributes.
-
-        Fields already present in `result.metadata` (`quality_score`,
-        `output_format`, `keywords`) are passed through as-is - they
-        don't need separate serialization. None values are filtered out.
         """
         meta: dict[str, Any] = {k: v for k, v in result.metadata.items() if v is not None}
 
+        if result.output_format:
+            meta["output_format"] = result.output_format
+        if result.quality_score:
+            meta["quality_score"] = result.quality_score
+        if result.extracted_keywords:
+            meta["keywords"] = _serialize_keywords(result.extracted_keywords)
         if result.processing_warnings:
             meta["processing_warnings"] = _serialize_warnings(result.processing_warnings)
         if result.detected_languages:

integrations/kreuzberg/src/haystack_integrations/components/converters/kreuzberg/utils.py

@@ -49,6 +49,19 @@ def _serialize_warnings(warnings: list[Any]) -> list[dict[str, str]]:
     return serialized
 
 
+def _serialize_keywords(keywords: list[Any]) -> list[dict[str, Any]]:
+    """Serialize kreuzberg `ExtractedKeyword` objects to plain dicts (PyO3 objects aren't picklable)."""
+    return [
+        {
+            "text": k.text,
+            "score": k.score,
+            "algorithm": k.algorithm,
+            "positions": list(k.positions) if k.positions is not None else None,
+        }
+        for k in keywords
+    ]
+
+
 def _serialize_annotations(annotations: list[Any]) -> list[dict[str, Any]]:
     """Serialize PDF annotations to plain dicts."""
     serialized = []

integrations/kreuzberg/tests/test_converter.py

@@ -21,6 +21,7 @@
 from haystack_integrations.components.converters.kreuzberg import KreuzbergConverter
 from haystack_integrations.components.converters.kreuzberg.utils import (
     _is_batch_error,
+    _serialize_keywords,
     _serialize_warnings,
 )
 
@@ -522,6 +523,26 @@ def test_helper_serialize_warnings_with_objects() -> None:
     assert result == [{"source": "parser", "message": "skipped element"}]
 
 
+def test_helper_serialize_keywords() -> None:
+    k = MagicMock(spec=["text", "score", "algorithm", "positions"])
+    k.text = "haystack"
+    k.score = 0.87
+    k.algorithm = "yake"
+    k.positions = [(0, 8), (42, 50)]
+    assert _serialize_keywords([k]) == [
+        {"text": "haystack", "score": 0.87, "algorithm": "yake", "positions": [(0, 8), (42, 50)]}
+    ]
+
+
+def test_helper_serialize_keywords_with_none_positions() -> None:
+    k = MagicMock(spec=["text", "score", "algorithm", "positions"])
+    k.text = "kreuzberg"
+    k.score = 0.5
+    k.algorithm = "yake"
+    k.positions = None
+    assert _serialize_keywords([k]) == [{"text": "kreuzberg", "score": 0.5, "algorithm": "yake", "positions": None}]
+
+
 def test_build_config_skips_auto_language_detection_when_already_set() -> None:
     config = ExtractionConfig(language_detection=LanguageDetectionConfig(enabled=False))
     converter = KreuzbergConverter(config=config)

integrations/kreuzberg/tests/test_converter_integration.py

@@ -207,7 +207,6 @@ def test_metadata_keyword_extraction() -> None:
     )
     result = converter.run(sources=[FIXTURES_DIR / "sample.pdf"])
     doc = _docs(result)[0]
-    # Keywords flow through from result.metadata as-is (plain dicts)
     assert "keywords" in doc.meta
     keywords = doc.meta["keywords"]
     assert len(keywords) == 3