Merge branch 'main' into content-extension

tests/integration/fixtures/common.py

@@ -270,7 +270,7 @@ def openai_client(client_with_models):
 
 @pytest.fixture(params=["openai_client", "client_with_models"])
 def compat_client(request, client_with_models):
-    if isinstance(client_with_models, LlamaStackAsLibraryClient):
+    if request.param == "openai_client" and isinstance(client_with_models, LlamaStackAsLibraryClient):
         # OpenAI client expects a server, so unless we also rewrite OpenAI client's requests
         # to go via the Stack library client (which itself rewrites requests to be served inline),
         # we cannot do this.

tests/integration/non_ci/responses/fixtures/test_cases/responses.yaml

@@ -137,7 +137,7 @@ test_response_multi_turn_tool_execution:
         server_url: "<FILLED_BY_TEST_RUNNER>"
       output: "yes"
     - case_id: "experiment_results_lookup"
-      input: "I need to get the results for the 'boiling_point' experiment. First, get the experiment ID for 'boiling_point', then use that ID to get the experiment results. Tell me what you found."
+      input: "I need to get the results for the 'boiling_point' experiment. First, get the experiment ID for 'boiling_point', then use that ID to get the experiment results. Tell me the boiling point in Celsius."
       tools:
       - type: mcp
         server_label: "localmcp"
@@ -149,7 +149,7 @@ test_response_multi_turn_tool_execution_streaming:
   test_params:
     case:
     - case_id: "user_permissions_workflow"
-      input: "Help me with this security check: First, get the user ID for 'charlie', then get the permissions for that user ID, and finally check if that user can access 'secret_file.txt'. Stream your progress as you work through each step."
+      input: "Help me with this security check: First, get the user ID for 'charlie', then get the permissions for that user ID, and finally check if that user can access 'secret_file.txt'. Stream your progress as you work through each step. Return only one tool call per step. Summarize the final result with a single 'yes' or 'no' response."
       tools:
       - type: mcp
         server_label: "localmcp"
@@ -157,7 +157,7 @@ test_response_multi_turn_tool_execution_streaming:
       stream: true
       output: "no"
     - case_id: "experiment_analysis_streaming"
-      input: "I need a complete analysis: First, get the experiment ID for 'chemical_reaction', then get the results for that experiment, and tell me if the yield was above 80%. Please stream your analysis process."
+      input: "I need a complete analysis: First, get the experiment ID for 'chemical_reaction', then get the results for that experiment, and tell me if the yield was above 80%. Return only one tool call per step.  Please stream your analysis process."
       tools:
       - type: mcp
         server_label: "localmcp"

tests/integration/non_ci/responses/test_responses.py

@@ -363,6 +363,9 @@ def test_response_non_streaming_file_search_empty_vector_store(request, compat_c
     ids=case_id_generator,
 )
 def test_response_non_streaming_mcp_tool(request, compat_client, text_model_id, case):
+    if not isinstance(compat_client, LlamaStackAsLibraryClient):
+        pytest.skip("in-process MCP server is only supported in library client")
+
     with make_mcp_server() as mcp_server_info:
         tools = case["tools"]
         for tool in tools:
@@ -381,12 +384,18 @@ def test_response_non_streaming_mcp_tool(request, compat_client, text_model_id,
         assert list_tools.type == "mcp_list_tools"
         assert list_tools.server_label == "localmcp"
         assert len(list_tools.tools) == 2
-        assert {t.name for t in list_tools.tools} == {"get_boiling_point", "greet_everyone"}
+        assert {t.name for t in list_tools.tools} == {
+            "get_boiling_point",
+            "greet_everyone",
+        }
 
         call = response.output[1]
         assert call.type == "mcp_call"
         assert call.name == "get_boiling_point"
-        assert json.loads(call.arguments) == {"liquid_name": "myawesomeliquid", "celsius": True}
+        assert json.loads(call.arguments) == {
+            "liquid_name": "myawesomeliquid",
+            "celsius": True,
+        }
         assert call.error is None
         assert "-100" in call.output
 
@@ -485,8 +494,11 @@ def test_response_non_streaming_multi_turn_image(request, compat_client, text_mo
     responses_test_cases["test_response_multi_turn_tool_execution"]["test_params"]["case"],
     ids=case_id_generator,
 )
-def test_response_non_streaming_multi_turn_tool_execution(request, compat_client, text_model_id, case):
+def test_response_non_streaming_multi_turn_tool_execution(compat_client, text_model_id, case):
     """Test multi-turn tool execution where multiple MCP tool calls are performed in sequence."""
+    if not isinstance(compat_client, LlamaStackAsLibraryClient):
+        pytest.skip("in-process MCP server is only supported in library client")
+
     with make_mcp_server(tools=dependency_tools()) as mcp_server_info:
         tools = case["tools"]
         # Replace the placeholder URL with the actual server URL
@@ -541,8 +553,11 @@ def test_response_non_streaming_multi_turn_tool_execution(request, compat_client
     responses_test_cases["test_response_multi_turn_tool_execution_streaming"]["test_params"]["case"],
     ids=case_id_generator,
 )
-async def test_response_streaming_multi_turn_tool_execution(request, compat_client, text_model_id, case):
+def test_response_streaming_multi_turn_tool_execution(compat_client, text_model_id, case):
     """Test streaming multi-turn tool execution where multiple MCP tool calls are performed in sequence."""
+    if not isinstance(compat_client, LlamaStackAsLibraryClient):
+        pytest.skip("in-process MCP server is only supported in library client")
+
     with make_mcp_server(tools=dependency_tools()) as mcp_server_info:
         tools = case["tools"]
         # Replace the placeholder URL with the actual server URL
@@ -572,6 +587,105 @@ async def test_response_streaming_multi_turn_tool_execution(request, compat_clie
             f"Last chunk should be response.completed, got {chunks[-1].type}"
         )
 
+        # Verify tool call streaming events are present
+        chunk_types = [chunk.type for chunk in chunks]
+
+        # Should have function call arguments delta events for tool calls
+        delta_events = [chunk for chunk in chunks if chunk.type == "response.function_call_arguments.delta"]
+        done_events = [chunk for chunk in chunks if chunk.type == "response.function_call_arguments.done"]
+
+        # Should have output item events for tool calls
+        item_added_events = [chunk for chunk in chunks if chunk.type == "response.output_item.added"]
+        item_done_events = [chunk for chunk in chunks if chunk.type == "response.output_item.done"]
+
+        # Verify we have substantial streaming activity (not just batch events)
+        assert len(chunks) > 10, f"Expected rich streaming with many events, got only {len(chunks)} chunks"
+
+        # Since this test involves MCP tool calls, we should see streaming events
+        assert len(delta_events) > 0, f"Expected function_call_arguments.delta events, got chunk types: {chunk_types}"
+        assert len(done_events) > 0, f"Expected function_call_arguments.done events, got chunk types: {chunk_types}"
+
+        # Should have output item events for function calls
+        assert len(item_added_events) > 0, f"Expected response.output_item.added events, got chunk types: {chunk_types}"
+        assert len(item_done_events) > 0, f"Expected response.output_item.done events, got chunk types: {chunk_types}"
+
+        # Verify delta events have proper structure
+        for delta_event in delta_events:
+            assert hasattr(delta_event, "delta"), "Delta event should have 'delta' field"
+            assert hasattr(delta_event, "item_id"), "Delta event should have 'item_id' field"
+            assert hasattr(delta_event, "sequence_number"), "Delta event should have 'sequence_number' field"
+            assert delta_event.delta, "Delta should not be empty"
+
+        # Verify done events have proper structure
+        for done_event in done_events:
+            assert hasattr(done_event, "arguments"), "Done event should have 'arguments' field"
+            assert hasattr(done_event, "item_id"), "Done event should have 'item_id' field"
+            assert done_event.arguments, "Final arguments should not be empty"
+
+        # Verify output item added events have proper structure
+        for added_event in item_added_events:
+            assert hasattr(added_event, "item"), "Added event should have 'item' field"
+            assert hasattr(added_event, "output_index"), "Added event should have 'output_index' field"
+            assert hasattr(added_event, "sequence_number"), "Added event should have 'sequence_number' field"
+            assert hasattr(added_event, "response_id"), "Added event should have 'response_id' field"
+            assert added_event.item.type in ["function_call", "mcp_call"], "Added item should be a tool call"
+            assert added_event.item.status == "in_progress", "Added item should be in progress"
+            assert added_event.response_id, "Response ID should not be empty"
+            assert isinstance(added_event.output_index, int), "Output index should be integer"
+            assert added_event.output_index >= 0, "Output index should be non-negative"
+
+        # Verify output item done events have proper structure
+        for done_event in item_done_events:
+            assert hasattr(done_event, "item"), "Done event should have 'item' field"
+            assert hasattr(done_event, "output_index"), "Done event should have 'output_index' field"
+            assert hasattr(done_event, "sequence_number"), "Done event should have 'sequence_number' field"
+            assert hasattr(done_event, "response_id"), "Done event should have 'response_id' field"
+            assert done_event.item.type in ["function_call", "mcp_call"], "Done item should be a tool call"
+            # Note: MCP calls don't have a status field, only function calls do
+            if done_event.item.type == "function_call":
+                assert done_event.item.status == "completed", "Function call should be completed"
+            assert done_event.response_id, "Response ID should not be empty"
+            assert isinstance(done_event.output_index, int), "Output index should be integer"
+            assert done_event.output_index >= 0, "Output index should be non-negative"
+
+        # Group function call argument events by item_id (these should have proper tracking)
+        function_call_events_by_item_id = {}
+        for chunk in chunks:
+            if hasattr(chunk, "item_id") and chunk.type in [
+                "response.function_call_arguments.delta",
+                "response.function_call_arguments.done",
+            ]:
+                item_id = chunk.item_id
+                if item_id not in function_call_events_by_item_id:
+                    function_call_events_by_item_id[item_id] = []
+                function_call_events_by_item_id[item_id].append(chunk)
+
+        for item_id, related_events in function_call_events_by_item_id.items():
+            # Should have at least one delta and one done event for a complete function call
+            delta_events = [e for e in related_events if e.type == "response.function_call_arguments.delta"]
+            done_events = [e for e in related_events if e.type == "response.function_call_arguments.done"]
+
+            assert len(delta_events) > 0, f"Item {item_id} should have at least one delta event"
+            assert len(done_events) == 1, f"Item {item_id} should have exactly one done event"
+
+            # Verify all events have the same item_id
+            for event in related_events:
+                assert event.item_id == item_id, f"Event should have consistent item_id {item_id}, got {event.item_id}"
+
+        # Basic pairing check: each output_item.added should be followed by some activity
+        # (but we can't enforce strict 1:1 pairing due to the complexity of multi-turn scenarios)
+        assert len(item_added_events) > 0, "Should have at least one output_item.added event"
+
+        # Verify response_id consistency across all events
+        response_ids = set()
+        for chunk in chunks:
+            if hasattr(chunk, "response_id"):
+                response_ids.add(chunk.response_id)
+            elif hasattr(chunk, "response") and hasattr(chunk.response, "id"):
+                response_ids.add(chunk.response.id)
+
+        assert len(response_ids) == 1, f"All events should reference the same response_id, found: {response_ids}"
+
         # Get the final response from the last chunk
         final_chunk = chunks[-1]
         if hasattr(final_chunk, "response"):
@@ -634,7 +748,7 @@ async def test_response_streaming_multi_turn_tool_execution(request, compat_clie
         },
     ],
 )
-def test_response_text_format(request, compat_client, text_model_id, text_format):
+def test_response_text_format(compat_client, text_model_id, text_format):
     if isinstance(compat_client, LlamaStackAsLibraryClient):
         pytest.skip("Responses API text format is not yet supported in library client.")
 
@@ -653,7 +767,7 @@ def test_response_text_format(request, compat_client, text_model_id, text_format
 
 
 @pytest.fixture
-def vector_store_with_filtered_files(request, compat_client, text_model_id, tmp_path_factory):
+def vector_store_with_filtered_files(compat_client, text_model_id, tmp_path_factory):
     """Create a vector store with multiple files that have different attributes for filtering tests."""
     if isinstance(compat_client, LlamaStackAsLibraryClient):
         pytest.skip("Responses API file search is not yet supported in library client.")
@@ -713,7 +827,9 @@ def vector_store_with_filtered_files(request, compat_client, text_model_id, tmp_
 
         # Attach file to vector store with attributes
         file_attach_response = compat_client.vector_stores.files.create(
-            vector_store_id=vector_store.id, file_id=file_response.id, attributes=file_data["attributes"]
+            vector_store_id=vector_store.id,
+            file_id=file_response.id,
+            attributes=file_data["attributes"],
         )
 
         # Wait for attachment

tests/integration/vector_io/test_openai_vector_stores.py

@@ -10,10 +10,11 @@ import uuid
 from io import BytesIO
 
 import pytest
-from llama_stack_client import BadRequestError, LlamaStackClient
+from llama_stack_client import BadRequestError
 from openai import BadRequestError as OpenAIBadRequestError
 
 from llama_stack.apis.vector_io import Chunk
+from llama_stack.core.library_client import LlamaStackAsLibraryClient
 
 logger = logging.getLogger(__name__)
 
@@ -476,9 +477,6 @@ def test_openai_vector_store_attach_file(compat_client_with_empty_stores, client
     """Test OpenAI vector store attach file."""
     skip_if_provider_doesnt_support_openai_vector_stores(client_with_models)
 
-    if isinstance(compat_client_with_empty_stores, LlamaStackClient):
-        pytest.skip("Vector Store Files attach is not yet supported with LlamaStackClient")
-
     compat_client = compat_client_with_empty_stores
 
     # Create a vector store
@@ -527,9 +525,6 @@ def test_openai_vector_store_attach_files_on_creation(compat_client_with_empty_s
     """Test OpenAI vector store attach files on creation."""
     skip_if_provider_doesnt_support_openai_vector_stores(client_with_models)
 
-    if isinstance(compat_client_with_empty_stores, LlamaStackClient):
-        pytest.skip("Vector Store Files attach is not yet supported with LlamaStackClient")
-
     compat_client = compat_client_with_empty_stores
 
     # Create some files and attach them to the vector store
@@ -583,9 +578,6 @@ def test_openai_vector_store_list_files(compat_client_with_empty_stores, client_
     """Test OpenAI vector store list files."""
     skip_if_provider_doesnt_support_openai_vector_stores(client_with_models)
 
-    if isinstance(compat_client_with_empty_stores, LlamaStackClient):
-        pytest.skip("Vector Store Files list is not yet supported with LlamaStackClient")
-
     compat_client = compat_client_with_empty_stores
 
     # Create a vector store
@@ -598,16 +590,20 @@ def test_openai_vector_store_list_files(compat_client_with_empty_stores, client_
             file_buffer.name = f"openai_test_{i}.txt"
             file = compat_client.files.create(file=file_buffer, purpose="assistants")
 
-        compat_client.vector_stores.files.create(
+        response = compat_client.vector_stores.files.create(
             vector_store_id=vector_store.id,
             file_id=file.id,
         )
+        assert response is not None
+        assert response.status == "completed", (
+            f"Failed to attach file {file.id} to vector store {vector_store.id}: {response=}"
+        )
         file_ids.append(file.id)
 
     files_list = compat_client.vector_stores.files.list(vector_store_id=vector_store.id)
     assert files_list
     assert files_list.object == "list"
-    assert files_list.data
+    assert files_list.data is not None
     assert not files_list.has_more
     assert len(files_list.data) == 3
     assert set(file_ids) == {file.id for file in files_list.data}
@@ -643,12 +639,13 @@ def test_openai_vector_store_list_files_invalid_vector_store(compat_client_with_
     """Test OpenAI vector store list files with invalid vector store ID."""
     skip_if_provider_doesnt_support_openai_vector_stores(client_with_models)
 
-    if isinstance(compat_client_with_empty_stores, LlamaStackClient):
-        pytest.skip("Vector Store Files list is not yet supported with LlamaStackClient")
-
     compat_client = compat_client_with_empty_stores
+    if isinstance(compat_client, LlamaStackAsLibraryClient):
+        errors = ValueError
+    else:
+        errors = (BadRequestError, OpenAIBadRequestError)
 
-    with pytest.raises((BadRequestError, OpenAIBadRequestError)):
+    with pytest.raises(errors):
         compat_client.vector_stores.files.list(vector_store_id="abc123")
 
 
@@ -656,9 +653,6 @@ def test_openai_vector_store_retrieve_file_contents(compat_client_with_empty_sto
     """Test OpenAI vector store retrieve file contents."""
     skip_if_provider_doesnt_support_openai_vector_stores(client_with_models)
 
-    if isinstance(compat_client_with_empty_stores, LlamaStackClient):
-        pytest.skip("Vector Store Files retrieve contents is not yet supported with LlamaStackClient")
-
     compat_client = compat_client_with_empty_stores
 
     # Create a vector store
@@ -686,9 +680,15 @@ def test_openai_vector_store_retrieve_file_contents(compat_client_with_empty_sto
         file_id=file.id,
     )
 
-    assert file_contents
-    assert file_contents.content[0]["type"] == "text"
-    assert file_contents.content[0]["text"] == test_content.decode("utf-8")
+    assert file_contents is not None
+    assert len(file_contents.content) == 1
+    content = file_contents.content[0]
+
+    # llama-stack-client returns a model, openai-python is a badboy and returns a dict
+    if not isinstance(content, dict):
+        content = content.model_dump()
+    assert content["type"] == "text"
+    assert content["text"] == test_content.decode("utf-8")
     assert file_contents.filename == file_name
     assert file_contents.attributes == attributes
 
@@ -697,9 +697,6 @@ def test_openai_vector_store_delete_file(compat_client_with_empty_stores, client
     """Test OpenAI vector store delete file."""
     skip_if_provider_doesnt_support_openai_vector_stores(client_with_models)
 
-    if isinstance(compat_client_with_empty_stores, LlamaStackClient):
-        pytest.skip("Vector Store Files list is not yet supported with LlamaStackClient")
-
     compat_client = compat_client_with_empty_stores
 
     # Create a vector store
@@ -752,9 +749,6 @@ def test_openai_vector_store_delete_file_removes_from_vector_store(compat_client
     """Test OpenAI vector store delete file removes from vector store."""
     skip_if_provider_doesnt_support_openai_vector_stores(client_with_models)
 
-    if isinstance(compat_client_with_empty_stores, LlamaStackClient):
-        pytest.skip("Vector Store Files attach is not yet supported with LlamaStackClient")
-
     compat_client = compat_client_with_empty_stores
 
     # Create a vector store
@@ -793,9 +787,6 @@ def test_openai_vector_store_update_file(compat_client_with_empty_stores, client
     """Test OpenAI vector store update file."""
     skip_if_provider_doesnt_support_openai_vector_stores(client_with_models)
 
-    if isinstance(compat_client_with_empty_stores, LlamaStackClient):
-        pytest.skip("Vector Store Files update is not yet supported with LlamaStackClient")
-
     compat_client = compat_client_with_empty_stores
 
     # Create a vector store
@@ -841,9 +832,6 @@ def test_create_vector_store_files_duplicate_vector_store_name(compat_client_wit
     """
     skip_if_provider_doesnt_support_openai_vector_stores(client_with_models)
 
-    if isinstance(compat_client_with_empty_stores, LlamaStackClient):
-        pytest.skip("Vector Store Files create is not yet supported with LlamaStackClient")
-
     compat_client = compat_client_with_empty_stores
 
     # Create a vector store with files