fix: replace in-place dataclass mutations with dataclasses.replace() (#3112)

* fix: replace in-place dataclass mutations with dataclasses.replace()

Fixes DeprecationWarnings triggered by direct attribute assignment on
Haystack dataclass instances (Document, StreamingChunk, ChatMessage).
Affected components:

- fastembed ranker: doc.score
- google_vertex document embedder: doc.embedding
- nvidia ranker: doc.score
- ollama document embedder: doc.embedding (sync + async)
- ollama chat generator: chunk.start (sync + async), chat_msg._meta

Part of deepset-ai/haystack#10956

* fix(nvidia): sort imports in ranker.py

Author: SyedShahmeerAli12

integrations/fastembed/src/haystack_integrations/components/rankers/fastembed/ranker.py

@@ -2,6 +2,7 @@
 #
 # SPDX-License-Identifier: Apache-2.0
 
+from dataclasses import replace
 from typing import Any
 
 from haystack import Document, component, default_from_dict, default_to_dict, logging
@@ -198,10 +199,6 @@ def run(self, query: str, documents: list[Document], top_k: int | None = None) -
         # Sort the list of tuples by the score in descending order
         sorted_doc_scores = sorted(doc_scores, key=lambda x: x[1], reverse=True)
 
-        # Get the top_k documents
-        top_k_documents = []
-        for doc, score in sorted_doc_scores[:top_k]:
-            doc.score = score
-            top_k_documents.append(doc)
+        top_k_documents = [replace(doc, score=score) for doc, score in sorted_doc_scores[:top_k]]
 
         return {"documents": top_k_documents}

integrations/google_vertex/src/haystack_integrations/components/embedders/google_vertex/document_embedder.py

@@ -1,5 +1,6 @@
 import math
 import time
+from dataclasses import replace
 from typing import Any, Literal, Optional
 
 import vertexai
@@ -271,8 +272,7 @@ def run(self, documents: list[Document]):
             i += batch_size
             batch_number += 1
 
-        for doc, embeddings in zip(documents, all_embeddings):
-            doc.embedding = embeddings
+        documents = [replace(doc, embedding=emb) for doc, emb in zip(documents, all_embeddings)]
 
         return {"documents": documents}
 

integrations/nvidia/src/haystack_integrations/components/rankers/nvidia/ranker.py

@@ -4,6 +4,7 @@
 
 import os
 import warnings
+from dataclasses import replace
 from typing import Any
 
 from haystack import Document, component, default_from_dict, default_to_dict, logging
@@ -236,11 +237,8 @@ def run(self, query: str, documents: list[Document], top_k: int | None = None) -
 
         # rank result is list[{index: int, logit: float}] sorted by logit
         sorted_indexes_and_scores = self.backend.rank(query_text=query_text, document_texts=document_texts)
-        sorted_documents = []
-        for item in sorted_indexes_and_scores[:top_k]:
-            # mutate (don't copy) the document because we're only updating the score
-            doc = documents[item["index"]]
-            doc.score = item["logit"]
-            sorted_documents.append(doc)
+        sorted_documents = [
+            replace(documents[item["index"]], score=item["logit"]) for item in sorted_indexes_and_scores[:top_k]
+        ]
 
         return {"documents": sorted_documents}

integrations/ollama/src/haystack_integrations/components/embedders/ollama/document_embedder.py

@@ -1,4 +1,5 @@
 import asyncio
+from dataclasses import replace
 from typing import Any
 
 from haystack import Document, component
@@ -209,8 +210,7 @@ def run(
             texts_to_embed=texts_to_embed, batch_size=self.batch_size, generation_kwargs=generation_kwargs
         )
 
-        for doc, emb in zip(documents, embeddings, strict=True):
-            doc.embedding = emb
+        documents = [replace(doc, embedding=emb) for doc, emb in zip(documents, embeddings, strict=True)]
 
         return {"documents": documents, "meta": {"model": self.model}}
 
@@ -245,7 +245,6 @@ async def run_async(
             texts_to_embed=texts_to_embed, batch_size=self.batch_size, generation_kwargs=generation_kwargs
         )
 
-        for doc, emb in zip(documents, embeddings, strict=True):
-            doc.embedding = emb
+        documents = [replace(doc, embedding=emb) for doc, emb in zip(documents, embeddings, strict=True)]
 
         return {"documents": documents, "meta": {"model": self.model}}

integrations/ollama/src/haystack_integrations/components/generators/ollama/chat/chat_generator.py

@@ -1,5 +1,6 @@
 import json
 from collections.abc import AsyncIterator, Callable, Iterator
+from dataclasses import replace
 from typing import Any, Literal
 
 from haystack import component, default_from_dict, default_to_dict
@@ -174,9 +175,12 @@ def _convert_ollama_response_to_chatmessage(ollama_response: ChatResponse) -> Ch
 
     reasoning = ollama_message.get("thinking", None)
 
-    chat_msg = ChatMessage.from_assistant(text=text or None, tool_calls=tool_calls, reasoning=reasoning)
-
-    chat_msg._meta = _convert_ollama_meta_to_openai_format(response_dict)
+    chat_msg = ChatMessage.from_assistant(
+        text=text or None,
+        tool_calls=tool_calls,
+        reasoning=reasoning,
+        meta=_convert_ollama_meta_to_openai_format(response_dict),
+    )
 
     return chat_msg
 
@@ -379,10 +383,9 @@ def _handle_streaming_response(
             chunk = _build_chunk(
                 chunk_response=raw, component_info=component_info, index=index, tool_call_index=tool_call_index
             )
-            chunks.append(chunk)
-
             start = index == 0 or bool(chunk.tool_calls)
-            chunk.start = start
+            chunk = replace(chunk, start=start)
+            chunks.append(chunk)
 
             if chunk.tool_calls:
                 for tool_call in chunk.tool_calls:
@@ -463,10 +466,9 @@ async def _handle_streaming_response_async(
             chunk = _build_chunk(
                 chunk_response=raw, component_info=component_info, index=index, tool_call_index=tool_call_index
             )
-            chunks.append(chunk)
-
             start = index == 0 or bool(chunk.tool_calls)
-            chunk.start = start
+            chunk = replace(chunk, start=start)
+            chunks.append(chunk)
 
             if chunk.tool_calls:
                 for tool_call in chunk.tool_calls: