[CI] Ignore ruff D209, rejoin the docstrings it split, and silence incompatible-rule warnings (#57212)

Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
This commit is contained in:
Harry Mellor
2026-09-16 10:57:24 -07:00
committed by GitHub
co-authored by Claude Opus 5
parent 1085b64425
commit 651a88c09a
693 changed files with 2088 additions and 4171 deletions
@@ -465,8 +465,7 @@ def test_system_message_warns_on_video(video_content):
def test_json_schema_response_format_missing_schema():
"""When response_format type is 'json_schema' but the json_schema field
is not provided, request construction should raise a validation error
so the API returns 400 instead of 500.
"""
so the API returns 400 instead of 500."""
with pytest.raises(Exception, match="json_schema.*must be provided"):
ChatCompletionRequest(
model=MODEL_NAME,
@@ -520,8 +519,7 @@ def test_structured_outputs_structural_tag_invalid(structural_tag):
@pytest.mark.parametrize("field_name", ["prompt_logprobs", "top_logprobs"])
def test_non_numeric_logprobs_rejected(field_name):
"""A non-numeric logprobs value must be a clean 400 validation error, not a
TypeError from the mode='before' comparison (which surfaces as HTTP 500).
"""
TypeError from the mode='before' comparison (which surfaces as HTTP 500)."""
with pytest.raises(VLLMValidationError, match=f"`{field_name}` must be an integer"):
ChatCompletionRequest(
model=MODEL_NAME,
@@ -81,8 +81,7 @@ async def test_chat_logit_bias_invalid(client):
@pytest.mark.asyncio
async def test_chat_logit_bias_non_integer_key(client):
"""Test that a non-integer logit_bias key is rejected with a clean,
informative error instead of a raw 'invalid literal for int()' message.
"""
informative error instead of a raw 'invalid literal for int()' message."""
with pytest.raises(openai.BadRequestError) as excinfo:
await client.chat.completions.create(
model=MODEL_NAME,
@@ -102,8 +101,7 @@ async def test_chat_logit_bias_non_integer_key(client):
@pytest.mark.asyncio
async def test_chat_logit_bias_non_numeric_value(client):
"""Test that a non-numeric logit_bias value is rejected with a message
that names the specific offending token, not just a generic TypeError.
"""
that names the specific offending token, not just a generic TypeError."""
with pytest.raises(openai.BadRequestError) as excinfo:
await client.chat.completions.create(
model=MODEL_NAME,
@@ -122,8 +120,7 @@ async def test_chat_logit_bias_non_numeric_value(client):
@pytest.mark.asyncio
async def test_chat_logit_bias_multiple_non_integer_keys(client):
"""Test that ALL invalid logit_bias keys are reported together,
not just the first one encountered.
"""
not just the first one encountered."""
with pytest.raises(openai.BadRequestError) as excinfo:
await client.chat.completions.create(
model=MODEL_NAME,
@@ -54,8 +54,7 @@ def stop_sign_image_url():
async def test_translategemma_extra_lang_code_fields(client: openai.AsyncOpenAI):
"""En -> es translation through TranslateGemma's bundled chat template's
``text`` content branch, which depends on ``source_lang_code`` /
``target_lang_code`` extra fields being preserved on the content part.
"""
``target_lang_code`` extra fields being preserved on the content part."""
completion = await client.chat.completions.create(
model=MODEL_NAME,
messages=[
@@ -99,8 +98,7 @@ async def test_translategemma_image_extra_lang_code_fields(
``source_lang_code`` / ``target_lang_code`` extras must survive parsing
and end up alongside the ``{"type": "image"}`` placeholder so the
template's image branch (``content["type"] == 'image'``) renders the
right OCR-translation prompt.
"""
right OCR-translation prompt."""
completion = await client.chat.completions.create(
model=MODEL_NAME,
messages=[
@@ -191,8 +191,7 @@ async def test_logprob_token_ids_stream_with_no_top_k(server):
async def test_logprob_token_ids_default_behavior_unchanged(server):
"""Without `logprob_token_ids`, the response carries the natural top-k
most-likely tokens. This guards against the new field accidentally
changing the default-path output.
"""
changing the default-path output."""
async with server.get_async_client() as client:
resp = await client.chat.completions.create(
model=MODEL_NAME,
@@ -1406,8 +1406,7 @@ async def test_serving_chat_did_set_correct_cache_salt(model_type):
@pytest.mark.asyncio
async def test_serving_chat_data_parallel_rank_extraction():
"""Test that data_parallel_rank is properly extracted from header and
passed to engine.
"""
passed to engine."""
mock_engine = MagicMock(spec=AsyncLLM)
mock_engine.errored = False
mock_engine.model_config = MockModelConfig()
@@ -1729,8 +1728,7 @@ class TestServingChatWithHarmony:
@pytest.mark.asyncio
async def test_system_message_without_tools(self, serving_chat, stream):
"""Leading system message produces a developer message with
DeveloperContent (# Instructions header).
"""
DeveloperContent (# Instructions header)."""
messages = [
{"role": "system", "content": "You are a helpful assistant."},
{"role": "user", "content": "Hello"},
@@ -1752,8 +1750,7 @@ class TestServingChatWithHarmony:
@pytest.mark.asyncio
async def test_system_message_with_tools(self, serving_chat, stream, weather_tools):
"""Leading system message is folded into the developer message
alongside tool definitions.
"""
alongside tool definitions."""
messages = [
{"role": "system", "content": "You are a helpful assistant."},
{"role": "user", "content": "What's the weather?"},
@@ -2168,8 +2165,7 @@ class TestServingChatWithHarmony:
@pytest.mark.asyncio
async def test_tool_choice_validation_without_parser():
"""Test that tool_choice='required' or named tool without tool_parser
returns an appropriate error message.
"""
returns an appropriate error message."""
mock_engine = MagicMock(spec=AsyncLLM)
mock_engine.errored = False
mock_engine.model_config = MockModelConfig()
@@ -151,8 +151,7 @@ async def client(request, server, server_with_auto_reasoning_config):
@pytest.mark.parametrize("client", ["default", "auto_config"], indirect=True)
async def test_thinking_token_budget_mixed_requests(client: openai.AsyncOpenAI):
"""Test that mixed requests (some with thinking_token_budget, some without)
complete successfully without errors.
"""
complete successfully without errors."""
response_with_budget = await client.chat.completions.create(
model=MODEL_NAME,
messages=MESSAGES,
@@ -521,8 +521,7 @@ async def test_completion_error_stream():
def test_json_schema_response_format_missing_schema():
"""When response_format type is 'json_schema' but the json_schema field
is not provided, request construction should raise a validation error
so the API returns 400 instead of 500.
"""
so the API returns 400 instead of 500."""
with pytest.raises(Exception, match="json_schema.*must be provided"):
CompletionRequest(
model=MODEL_NAME,
@@ -585,8 +584,7 @@ def test_negative_prompt_token_ids_flat():
def test_logprobs_minus_one_allowed():
"""logprobs=-1 means "return all logprobs". The sampling layer and the chat
top_logprobs / prompt_logprobs validators all accept -1, so the completion
logprobs validator must accept it too instead of rejecting it as negative.
"""
logprobs validator must accept it too instead of rejecting it as negative."""
request = CompletionRequest(
model=MODEL_NAME,
prompt="Test prompt",
@@ -746,8 +744,7 @@ class TestCompletionPromptListLimit:
@pytest.mark.parametrize("field_name", ["prompt_logprobs", "logprobs"])
def test_non_numeric_logprobs_rejected(field_name):
"""A non-numeric logprobs value must be a clean 400 validation error, not a
TypeError from the mode='before' comparison (which surfaces as HTTP 500).
"""
TypeError from the mode='before' comparison (which surfaces as HTTP 500)."""
with pytest.raises(VLLMValidationError, match=f"`{field_name}` must be an integer"):
CompletionRequest(
model=MODEL_NAME,
@@ -43,8 +43,7 @@ def _system():
class TestResponseInputToHarmonyRenderParity:
"""Each test drives the same conversation through both APIs and asserts
identical Harmony messages and rendered token sequences.
"""
identical Harmony messages and rendered token sequences."""
# -----------------------------------------------------------------------
# Single-message cases
@@ -52,8 +51,7 @@ class TestResponseInputToHarmonyRenderParity:
def test_developer_message(self):
"""Both APIs must render developer messages identically using
DeveloperContent (with the '# Instructions' header).
"""
DeveloperContent (with the '# Instructions' header)."""
chat_msgs = parse_chat_input_to_harmony_message(
{"role": "developer", "content": "Be concise."}
)
@@ -248,8 +246,7 @@ class TestResponseInputToHarmonyRenderParity:
def test_reasoning_combined_with_function_call(self):
"""Chat API packs reasoning + tool_calls into one dict; responses API
represents them as two separate items. Both must produce the same two
Harmony messages in the same order: analysis then commentary.
"""
Harmony messages in the same order: analysis then commentary."""
chat_msgs = parse_chat_input_to_harmony_message(
{
"role": "assistant",
@@ -532,8 +529,7 @@ class TestResponseInputToHarmonyRenderParity:
def test_completed_turns_drop_reasoning(self):
"""Validates that reasoning from completed turns is dropped, while
reasoning from the current in-progress tool-call turn is preserved
in both chat completions and responses previous_input_messages.
"""
in both chat completions and responses previous_input_messages."""
first_turn_reasoning = "FIRST_TURN_REASONING"
second_turn_reasoning = "SECOND_TURN_REASONING"
@@ -323,8 +323,7 @@ class TestCommonParseInputToHarmonyMessage:
def test_system_message(self, parse_function):
"""Test parsing system messages, which are parsed into developer messages
with DeveloperContent.
"""
with DeveloperContent."""
chat_msg = {
"role": "system",
"content": "You are a helpful assistant",
@@ -12,8 +12,7 @@ from vllm.entrypoints.openai.responses.protocol import ResponsesRequest
def test_function_call_dict_converted_to_object():
"""Test that function_call dictionaries are correctly parsed into
ResponseFunctionToolCall objects.
"""
ResponseFunctionToolCall objects."""
# Create a request with function_call as dict
request_data = {
"model": "gpt-oss",
@@ -330,8 +329,7 @@ def test_validator_handles_empty_iterator():
def test_assistant_string_content_stays_easyinput():
"""EasyInput assistant message with plain string content is not
coerced into a ResponseOutputMessage.
"""
coerced into a ResponseOutputMessage."""
request_data = {
"model": "test-model",
"input": [
@@ -353,8 +351,7 @@ def test_assistant_string_content_stays_easyinput():
def test_assistant_output_style_content_coerced():
"""Assistant message whose content is output-message-shaped (list of
output_text items) should be coerced to ResponseOutputMessage.
"""
output_text items) should be coerced to ResponseOutputMessage."""
request_data = {
"model": "test-model",
"input": [
@@ -67,8 +67,7 @@ class TestResponseInputToHarmonyMessage:
def test_system_message(self):
"""System messages carry developer instructions and must be rendered
as developer messages with DeveloperContent.
"""
as developer messages with DeveloperContent."""
msg = response_input_to_harmony(
{"type": "message", "role": "system", "content": "Be helpful."},
prev_responses=[],
@@ -90,8 +89,7 @@ class TestResponseInputToHarmonyMessage:
def test_developer_message_gets_instructions_prefix(self):
"""Developer messages must use DeveloperContent which adds the
'# Instructions' header the model was trained on.
"""
'# Instructions' header the model was trained on."""
msg = response_input_to_harmony(
{"type": "message", "role": "developer", "content": "Be concise."},
prev_responses=[],
@@ -121,8 +119,7 @@ class TestResponseInputToHarmonyMessage:
def test_developer_message_array_content_concatenated(self):
"""Array content in developer messages is flattened and rendered
via DeveloperContent with the '# Instructions' header.
"""
via DeveloperContent with the '# Instructions' header."""
msg = response_input_to_harmony(
{
"type": "message",
@@ -213,8 +210,7 @@ class TestResponseInputToHarmonyMessage:
def test_function_call_output_uses_most_recent_matching_call(self):
"""When multiple prev_responses share a call_id, the last one wins
because the search is reversed.
"""
because the search is reversed."""
earlier = ResponseFunctionToolCall(
id="fc_old",
call_id="call_test",
@@ -839,13 +839,11 @@ class TestConstructChatMessagesCombinePolicy:
class TestConstructInputMessagesInstructionsLeak:
"""Regression tests for #37697: instructions from a prior response
should NOT leak through previous_response_id.
"""
should NOT leak through previous_response_id."""
def test_old_instructions_stripped_from_prev_msg(self):
"""System message in prev_msg must be dropped so the new request's
instructions are the only system message in the conversation.
"""
instructions are the only system message in the conversation."""
prev = [
{"role": "system", "content": "old instructions"},
{"role": "user", "content": "What is 2+2?"},
@@ -862,8 +860,7 @@ class TestConstructInputMessagesInstructionsLeak:
def test_no_instructions_in_new_request(self):
"""If the new request has no instructions, old ones should still
be stripped -- they must not carry over.
"""
be stripped -- they must not carry over."""
prev = [
{"role": "system", "content": "old instructions"},
{"role": "user", "content": "Hi"},
@@ -1163,8 +1163,7 @@ class TestHarmonyPreambleStreaming:
def test_preamble_done_emits_text_done_events(self) -> None:
"""Completed preamble should emit text done + content_part done +
output_item done, same shape as final channel.
"""
output_item done, same shape as final channel."""
from vllm.entrypoints.openai.responses.streaming_events import (
emit_previous_item_done_events,
)
@@ -1185,8 +1184,7 @@ class TestHarmonyPreambleStreaming:
def test_commentary_with_recipient_no_preamble_done(self) -> None:
"""Commentary + recipient='functions.X' should route to function call
done, not preamble done.
"""
done, not preamble done."""
from vllm.entrypoints.openai.responses.streaming_events import (
emit_previous_item_done_events,
)
@@ -1332,8 +1330,7 @@ def _mock_parser_with_reasoning(serving, delta_sequence: list[DeltaMessage]):
class TestStreamingReasoningToContentTransition:
"""Tests for _process_simple_streaming_events reasoning-to-content
transition, specifically the fix for mixed deltas that carry both
reasoning and content simultaneously.
"""
reasoning and content simultaneously."""
@pytest.mark.asyncio
async def test_mixed_delta_reasoning_and_content_emits_reasoning_delta(
@@ -1343,8 +1340,7 @@ class TestStreamingReasoningToContentTransition:
and content set (e.g. reasoning end and content start in the same
chunk), the trailing reasoning text must be emitted as a
ResponseReasoningTextDeltaEvent and included in the
ResponseReasoningTextDoneEvent text.
"""
ResponseReasoningTextDoneEvent text."""
monkeypatch.setattr(envs, "VLLM_USE_EXPERIMENTAL_PARSER_CONTEXT", False)
serving = _make_serving_instance_with_reasoning()
@@ -1413,8 +1409,7 @@ class TestStreamingReasoningToContentTransition:
self, monkeypatch
):
"""When the transition from reasoning to content is clean (no mixed
delta), no extra reasoning delta event should be emitted.
"""
delta), no extra reasoning delta event should be emitted."""
monkeypatch.setattr(envs, "VLLM_USE_EXPERIMENTAL_PARSER_CONTEXT", False)
serving = _make_serving_instance_with_reasoning()
@@ -1475,8 +1470,7 @@ class TestStreamingReasoningToContentTransition:
async def test_reasoning_only_stream_no_content(self, monkeypatch):
"""When the stream has only reasoning deltas and no content, the
reasoning done event should be emitted at finalization with the
full accumulated text, and no text delta events should appear.
"""
full accumulated text, and no text delta events should appear."""
monkeypatch.setattr(envs, "VLLM_USE_EXPERIMENTAL_PARSER_CONTEXT", False)
serving = _make_serving_instance_with_reasoning()
@@ -143,8 +143,7 @@ async def test_streaming_output_consistency(client: OpenAI, model_name: str):
@pytest.mark.parametrize("model_name", [MODEL_NAME])
async def test_streaming_logprobs(client: OpenAI, model_name: str):
"""Test that streaming with logprobs returns valid logprob data on
output_text.delta events and that top_logprobs has the requested count.
"""
output_text.delta events and that top_logprobs has the requested count."""
response = await client.responses.create(
model=model_name,
input="Say hello.",
@@ -124,8 +124,7 @@ class TestProcessorCompoundDeltas:
def test_split_name_and_args_same_index(self):
"""Regression: parsers like KimiK2 emit name and args as separate
DeltaToolCalls at the same index within one DeltaMessage.
"""
DeltaToolCalls at the same index within one DeltaMessage."""
tc_name = _make_tool_call(0, name="get_weather")
tc_args = _make_tool_call(0, arguments='{"city":"SF"}')
delta = DeltaMessage(tool_calls=[tc_name, tc_args])
@@ -141,8 +140,7 @@ class TestProcessorCompoundDeltas:
def test_reasoning_to_content_transition(self):
"""Regression: the old special case in emit_delta handled this;
now split_delta handles it generically.
"""
now split_delta handles it generically."""
processor = SimpleStreamingEventProcessor()
_run_through_processor(processor, DeltaMessage(reasoning="think"))
assert processor.state.current_state == _StateType.REASONING
@@ -25,8 +25,7 @@ def _model_config() -> Mock:
def test_completion_flag_forwarded_to_tok_params():
"""build_tok_params must forward return_token_offsets, defaulting to
False (zero behavioral change for existing callers) and coercing JSON
null to False via the bool() guard.
"""
null to False via the bool() guard."""
cfg = _model_config()
default = CompletionRequest(model="m", prompt="hi")
@@ -41,8 +40,7 @@ def test_completion_flag_forwarded_to_tok_params():
def test_chat_flag_forwarded_to_tok_params():
"""Chat build_tok_params has its own (max_completion_tokens) branch, so
its return_token_offsets forwarding is verified independently.
"""
its return_token_offsets forwarding is verified independently."""
cfg = _model_config()
messages = [{"role": "user", "content": "hi"}]
@@ -66,8 +64,7 @@ def test_generate_request_token_offsets_default_none():
def test_generate_request_token_offsets_survive_json_round_trip():
"""GenerateRequest crosses the disagg serialization boundary; the
tuple[int, int] offsets must survive model_dump and re-validate.
"""
tuple[int, int] offsets must survive model_dump and re-validate."""
req = GenerateRequest(
token_ids=[10, 20],
sampling_params=SamplingParams(),