remove synthetic id fallback, key accumulator by index

Author: SyedShahmeerAli12

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

@@ -165,10 +165,10 @@ def _convert_ollama_response_to_chatmessage(ollama_response: ChatResponse) -> Ch
     tool_calls: list[ToolCall] = []
 
     if ollama_tool_calls := ollama_message.get("tool_calls"):
-        for idx, ollama_tc in enumerate(ollama_tool_calls):
+        for ollama_tc in ollama_tool_calls:
             tool_calls.append(
                 ToolCall(
-                    id=ollama_tc.get("id") or f"call_{idx}",
+                    id=ollama_tc.get("id"),
                     tool_name=ollama_tc["function"]["name"],
                     arguments=ollama_tc["function"]["arguments"],
                 )
@@ -209,7 +209,7 @@ def _build_chunk(
             tool_calls_list.append(
                 ToolCallDelta(
                     index=tool_call_index,
-                    id=tool_call.get("id") or f"call_{tool_call_index}",
+                    id=tool_call.get("id"),
                     tool_name=tool_call["function"]["name"],
                     arguments=json.dumps(tool_call["function"]["arguments"])
                     if tool_call["function"]["arguments"]
@@ -372,10 +372,11 @@ def _handle_streaming_response(
         component_info = ComponentInfo.from_component(self)
         chunks: list[StreamingChunk] = []
 
-        # Accumulators
-        arg_by_id: dict[str, str] = {}
-        name_by_id: dict[str, str] = {}
-        id_order: list[str] = []
+        # Accumulators keyed by tool_call.index (always unique per call, even for repeated tool names)
+        arg_by_index: dict[str, str] = {}
+        name_by_index: dict[str, str] = {}
+        id_by_index: dict[str, str | None] = {}
+        index_order: list[str] = []
         tool_call_index: int = 0
 
         # track reasoning and content blocks to correctly set start=True on the first chunk of each block
@@ -402,15 +403,14 @@ def _handle_streaming_response(
 
             if chunk.tool_calls:
                 for tool_call in chunk.tool_calls:
-                    # id is always set by _build_chunk (either from the server or synthetic "call_N").
-                    # Fall back to tool_name only as a last resort for callers that bypass _build_chunk.
-                    tool_call_id = tool_call.id or tool_call.tool_name or ""
+                    key = str(tool_call.index)
                     args = tool_call.arguments or ""
 
-                    if tool_call_id not in id_order:
-                        id_order.append(tool_call_id)
-                        name_by_id[tool_call_id] = tool_call.tool_name or ""
-                    arg_by_id[tool_call_id] = args
+                    if key not in index_order:
+                        index_order.append(key)
+                        name_by_index[key] = tool_call.tool_name or ""
+                        id_by_index[key] = tool_call.id
+                    arg_by_index[key] = args
 
             if callback:
                 callback(chunk)
@@ -423,10 +423,10 @@ def _handle_streaming_response(
             reasoning += c.reasoning.reasoning_text if c.reasoning else ""
 
         tool_calls = []
-        for tool_call_id in id_order:
-            arguments: str = arg_by_id.get(tool_call_id, "")
+        for key in index_order:
+            arguments: str = arg_by_index.get(key, "")
             tool_calls.append(
-                ToolCall(id=tool_call_id, tool_name=name_by_id[tool_call_id], arguments=json.loads(arguments))
+                ToolCall(id=id_by_index[key], tool_name=name_by_index[key], arguments=json.loads(arguments))
             )
 
         # We can't use _convert_streaming_chunks_to_chat_message because
@@ -453,10 +453,11 @@ async def _handle_streaming_response_async(
         component_info = ComponentInfo.from_component(self)
         chunks: list[StreamingChunk] = []
 
-        # Accumulators
-        arg_by_id: dict[str, str] = {}
-        name_by_id: dict[str, str] = {}
-        id_order: list[str] = []
+        # Accumulators keyed by tool_call.index (always unique per call, even for repeated tool names)
+        arg_by_index: dict[str, str] = {}
+        name_by_index: dict[str, str] = {}
+        id_by_index: dict[str, str | None] = {}
+        index_order: list[str] = []
         tool_call_index: int = 0
 
         # track reasoning and content blocks to correctly set start=True on the first chunk of each block
@@ -484,13 +485,14 @@ async def _handle_streaming_response_async(
 
             if chunk.tool_calls:
                 for tool_call in chunk.tool_calls:
-                    tool_call_id = tool_call.id or tool_call.tool_name or ""
+                    key = str(tool_call.index)
                     args = tool_call.arguments or ""
 
-                    if tool_call_id not in id_order:
-                        id_order.append(tool_call_id)
-                        name_by_id[tool_call_id] = tool_call.tool_name or ""
-                    arg_by_id[tool_call_id] = args
+                    if key not in index_order:
+                        index_order.append(key)
+                        name_by_index[key] = tool_call.tool_name or ""
+                        id_by_index[key] = tool_call.id
+                    arg_by_index[key] = args
 
             if callback is not None:
                 await callback(chunk)
@@ -505,10 +507,10 @@ async def _handle_streaming_response_async(
             reasoning += c.reasoning.reasoning_text if c.reasoning else ""
 
         tool_calls = []
-        for tool_call_id in id_order:
-            arguments: str = arg_by_id.get(tool_call_id, "")
+        for key in index_order:
+            arguments: str = arg_by_index.get(key, "")
             tool_calls.append(
-                ToolCall(id=tool_call_id, tool_name=name_by_id[tool_call_id], arguments=json.loads(arguments))
+                ToolCall(id=id_by_index[key], tool_name=name_by_index[key], arguments=json.loads(arguments))
             )
 
         # We can't use _convert_streaming_chunks_to_chat_message because

integrations/ollama/tests/test_chat_generator.py

@@ -190,7 +190,6 @@ def test_convert_ollama_response_to_chatmessage_with_tools(self):
         assert observed.role == "assistant"
         assert observed.text is None
         assert observed.tool_call == ToolCall(
-            id="call_0",
             tool_name="get_current_weather",
             arguments={"format": "celsius", "location": "Paris, FR"},
         )
@@ -219,8 +218,8 @@ def test_convert_ollama_response_to_chatmessage_with_repeated_tool(self):
         observed = _convert_ollama_response_to_chatmessage(ollama_response)
 
         assert len(observed.tool_calls) == 2
-        assert observed.tool_calls[0] == ToolCall(id="call_0", tool_name="weather", arguments={"city": "Paris"})
-        assert observed.tool_calls[1] == ToolCall(id="call_1", tool_name="weather", arguments={"city": "London"})
+        assert observed.tool_calls[0] == ToolCall(tool_name="weather", arguments={"city": "Paris"})
+        assert observed.tool_calls[1] == ToolCall(tool_name="weather", arguments={"city": "London"})
 
     def test_build_chunk(self):
         generator = OllamaChatGenerator()
@@ -414,10 +413,10 @@ def test_callback(chunk: StreamingChunk):
         assert result["replies"][0].text is None
         assert result["replies"][0].tool_calls[0].tool_name == "calculator"
         assert result["replies"][0].tool_calls[0].arguments == {"expression": "7 * (4 + 2)"}
-        assert result["replies"][0].tool_calls[0].id == "call_1"
+        assert result["replies"][0].tool_calls[0].id is None
         assert result["replies"][0].tool_calls[1].tool_name == "factorial"
         assert result["replies"][0].tool_calls[1].arguments == {"n": 5}
-        assert result["replies"][0].tool_calls[1].id == "call_2"
+        assert result["replies"][0].tool_calls[1].id is None
         assert result["replies"][0].meta["finish_reason"] == "stop"
         assert result["replies"][0].meta["model"] == "qwen3:0.6b"
 
@@ -430,7 +429,7 @@ def test_callback(chunk: StreamingChunk):
         expected = {
             "index": 1,
             "arguments": '{"expression": "7 * (4 + 2)"}',
-            "id": "call_1",
+            "id": None,
             "tool_name": "calculator",
         }
         # We add extra to the expected dict if it exists in the result for comparison
@@ -443,7 +442,7 @@ def test_callback(chunk: StreamingChunk):
             "index": 2,
             "tool_name": "factorial",
             "arguments": '{"n": 5}',
-            "id": "call_2",
+            "id": None,
         }
         # We add extra to the expected dict if it exists in the result for comparison
         # This was added in PR https://github.com/deepset-ai/haystack/pull/10018 and released in Haystack 2.20.0
@@ -503,10 +502,10 @@ def test_handle_streaming_response_repeated_tool_calls(self):
         assert len(result["replies"][0].tool_calls) == 2
         assert result["replies"][0].tool_calls[0].tool_name == "weather"
         assert result["replies"][0].tool_calls[0].arguments == {"city": "Paris"}
-        assert result["replies"][0].tool_calls[0].id == "call_1"
+        assert result["replies"][0].tool_calls[0].id is None
         assert result["replies"][0].tool_calls[1].tool_name == "weather"
         assert result["replies"][0].tool_calls[1].arguments == {"city": "London"}
-        assert result["replies"][0].tool_calls[1].id == "call_2"
+        assert result["replies"][0].tool_calls[1].id is None
 
     def test_handle_streaming_response_tool_calls_with_thinking(self):
         ollama_chunks = [
@@ -622,7 +621,7 @@ def test_callback(chunk: StreamingChunk):
         assert result["replies"][0].text is None
         assert result["replies"][0].tool_calls[0].tool_name == "add_two_numbers"
         assert result["replies"][0].tool_calls[0].arguments == {"a": 2, "b": 2}
-        assert result["replies"][0].tool_calls[0].id == "call_1"
+        assert result["replies"][0].tool_calls[0].id is None
         assert result["replies"][0].reasoning.reasoning_text == "Okay, the user is asking 2 plus 2."
         assert result["replies"][0].meta["finish_reason"] == "stop"
         assert result["replies"][0].meta["model"] == "qwen3:0.6b"
@@ -651,7 +650,7 @@ def test_callback(chunk: StreamingChunk):
         expected = {
             "index": 1,
             "arguments": '{"a": 2, "b": 2}',
-            "id": "call_1",
+            "id": None,
             "tool_name": "add_two_numbers",
         }
         serialized_dict = streaming_chunks[12].tool_calls[0].to_dict()