fix: preserve timeout across HTTP retries in requests_utils

haystack/components/classifiers/zero_shot_document_classifier.py

@@ -159,6 +159,8 @@ def to_dict(self) -> dict[str, Any]:
             self,
             labels=self.labels,
             model=self.huggingface_pipeline_kwargs["model"],
+            classification_field=self.classification_field,
+            multi_label=self.multi_label,
             huggingface_pipeline_kwargs=self.huggingface_pipeline_kwargs,
             token=self.token,
             multi_label=self.multi_label,

haystack/components/joiners/document_joiner.py

@@ -171,7 +171,7 @@ def _concatenate(document_lists: list[list[Document]]) -> list[Document]:
         for doc in itertools.chain.from_iterable(document_lists):
             docs_per_id[doc.id].append(doc)
         for docs in docs_per_id.values():
-            doc_with_best_score = max(docs, key=lambda doc: doc.score if doc.score else -inf)
+            doc_with_best_score = max(docs, key=lambda doc: doc.score if doc.score is not None else -inf)
             output.append(doc_with_best_score)
         return output
 

haystack/utils/requests_utils.py

@@ -67,9 +67,10 @@ def request_with_retry(
         after=after_log(logger, logging.DEBUG),
     )
     def run() -> httpx.Response:
-        timeout = kwargs.pop("timeout", 10)
+        request_kwargs = dict(kwargs)
+        timeout = request_kwargs.pop("timeout", 10)
         with httpx.Client() as client:
-            res = client.request(**kwargs, timeout=timeout)
+            res = client.request(**request_kwargs, timeout=timeout)
 
             if res.status_code in status_codes_to_retry:
                 # We raise only for the status codes that must trigger a retry
@@ -177,9 +178,10 @@ async def example_5xx():
         after=after_log(logger, logging.DEBUG),
     )
     async def run() -> httpx.Response:
-        timeout = kwargs.pop("timeout", 10)
+        request_kwargs = dict(kwargs)
+        timeout = request_kwargs.pop("timeout", 10)
         async with httpx.AsyncClient() as client:
-            res = await client.request(**kwargs, timeout=timeout)
+            res = await client.request(**request_kwargs, timeout=timeout)
 
             if res.status_code in status_codes_to_retry:
                 # We raise only for the status codes that must trigger a retry

releasenotes/notes/Fix-DocumentJoiner-concatenate-mode-treating-score-0.0-as-missing-784af68479fb54bc.yaml

@@ -0,0 +1,5 @@
+---
+fixes:
+  - |
+    Fixed `DocumentJoiner` in `concatenate` mode treating documents with a score of `0.0` as unscored when deduplicating by ID.
+    Duplicate documents with a zero score could lose to documents with negative or missing scores.

releasenotes/notes/Fix-TransformersZeroShotDocumentClassifier-serialization-of-classification_field-and-multi_label-441a88b5195b8a8c.yaml

@@ -0,0 +1,4 @@
+---
+fixes:
+  - |
+    `TransformersZeroShotDocumentClassifier.to_dict()` now serializes `classification_field` and `multi_label`, so pipeline dump/load round-trips preserve the configured classification behavior.

releasenotes/notes/preserve-timeout-across-retries-a590b478d89d435b.yaml

@@ -0,0 +1,5 @@
+---
+fixes:
+  - |
+    Preserve user-specified timeouts across retries in `request_with_retry` and
+    `async_request_with_retry`.

test/components/classifiers/test_zero_shot_document_classifier.py

@@ -36,6 +36,8 @@ def test_to_dict(self):
             "init_parameters": {
                 "model": "cross-encoder/nli-deberta-v3-xsmall",
                 "labels": ["positive", "negative"],
+                "classification_field": None,
+                "multi_label": False,
                 "token": {"env_vars": ["HF_API_TOKEN", "HF_TOKEN"], "strict": False, "type": "env_var"},
                 "multi_label": True,
                 "classification_field": "title",
@@ -47,6 +49,17 @@ def test_to_dict(self):
             },
         }
 
+    def test_to_dict_from_dict_round_trip(self):
+        component = TransformersZeroShotDocumentClassifier(
+            model="cross-encoder/nli-deberta-v3-xsmall",
+            labels=["a", "b"],
+            classification_field="title",
+            multi_label=True,
+        )
+        restored = TransformersZeroShotDocumentClassifier.from_dict(component.to_dict())
+        assert restored.classification_field == "title"
+        assert restored.multi_label is True
+
     def test_from_dict(self, del_hf_env_vars):
         data = {
             "type": "haystack.components.classifiers.zero_shot_document_classifier.TransformersZeroShotDocumentClassifier",  # noqa: E501

test/components/joiners/test_document_joiner.py

@@ -146,6 +146,17 @@ def test_run_with_concatenate_join_mode_and_duplicate_documents(self):
             output["documents"], key=lambda d: d.id
         )
 
+    def test_concatenate_keeps_highest_score_for_zero_and_negative_scores(self):
+        joiner = DocumentJoiner(sort_by_score=False)
+        documents_1 = [Document(id="dup", content="no score")]
+        documents_2 = [Document(id="dup", content="zero score", score=0.0)]
+        documents_3 = [Document(id="dup", content="negative score", score=-0.1)]
+
+        output = joiner.run([documents_1, documents_2, documents_3])
+        assert len(output["documents"]) == 1
+        assert output["documents"][0].content == "zero score"
+        assert output["documents"][0].score == 0.0
+
     def test_run_with_merge_join_mode(self):
         joiner = DocumentJoiner(join_mode="merge", weights=[1.5, 0.5])
         documents_1 = [Document(content="a", score=1.0), Document(content="b", score=2.0)]