mirror of
https://github.com/bytedance/deer-flow.git
synced 2026-11-09 04:20:34 +00:00
aaca3eeaeb
* refactor(agent): ground runtime decisions in structured evidence Use producer status and structured errors for tool recovery, preserve memory context without lexical confirmation, and remove predetermined receipt and skill conclusions. Require explicit MCP discovery and authenticated schedule trial actions. Preserve authorization, ownership, isolation, and bounded execution; update documentation and regression coverage. * test(sandbox): expect read_file initialization error callback * fix(agent): address evidence review follow-ups * fix(agent): preserve retry budgets for OpenAI transport errors * fix(agent): expose tool execution status in model requests Project validated execution metadata and exact tool text into request-only JSON so status survives provider serialization. Preserve stored results and receipt hashes, cover SDK and graph boundaries, and migrate replay input hashes without changing recorded responses. --------- Co-authored-by: Willem Jiang <willem.jiang@gmail.com>
97 lines
3.9 KiB
Python
97 lines
3.9 KiB
Python
"""A serialized MCP routing index must not break middleware construction.
|
|
|
|
``McpRoutingMiddleware._normalize_index`` documents that it accepts "arbitrary
|
|
serialized routing data, not only the output of the builder". Two shapes the
|
|
loader can produce nevertheless escaped the coercion:
|
|
|
|
- an entry that is not a mapping (``str`` / ``list`` / ``None``) raised
|
|
``AttributeError`` on ``.get``;
|
|
- a non-finite ``priority`` raised ``OverflowError`` from ``int()`` — JSON
|
|
accepts ``Infinity``, so a hand-edited or foreign index can carry one.
|
|
|
|
Construction happens once per run, so either shape turns every model call in
|
|
that process into an exception instead of the entry being ignored.
|
|
"""
|
|
|
|
from __future__ import annotations
|
|
|
|
import pytest
|
|
from langchain_core.messages import HumanMessage
|
|
from langchain_core.tools import tool
|
|
|
|
from deerflow.agents.middlewares.mcp_routing_middleware import McpRoutingMiddleware
|
|
from deerflow.tools.builtins.tool_search import assemble_deferred_tools, build_mcp_routing_middleware
|
|
from deerflow.tools.mcp_metadata import tag_mcp_routing, tag_mcp_tool
|
|
|
|
normalize = McpRoutingMiddleware._normalize_index
|
|
|
|
|
|
@pytest.mark.parametrize("bad_entry", ["just-a-string", ["keyword"], None, 7, 1.5])
|
|
def test_non_mapping_entries_are_skipped(bad_entry):
|
|
"""The malformed entry is dropped; its valid sibling still normalizes."""
|
|
index = {"tool.bad": bad_entry, "tool.good": {"priority": 2, "keywords": ["search"]}}
|
|
|
|
assert normalize(index) == {"tool.good": (2, ("search",))}
|
|
|
|
|
|
@pytest.mark.parametrize("priority", [float("inf"), float("-inf")])
|
|
def test_non_finite_priority_falls_back_to_zero(priority):
|
|
"""int() refuses infinity, and JSON accepts it as a literal."""
|
|
assert normalize({"tool.a": {"priority": priority, "keywords": ["k"]}}) == {"tool.a": (0, ("k",))}
|
|
|
|
|
|
def test_nan_priority_falls_back_to_zero():
|
|
assert normalize({"tool.a": {"priority": float("nan"), "keywords": ["k"]}}) == {"tool.a": (0, ("k",))}
|
|
|
|
|
|
def test_valid_entries_are_unchanged():
|
|
"""Control: the builder's own output still normalizes exactly as before."""
|
|
index = {
|
|
"tool.a": {"priority": 3, "keywords": ["search", "find"]},
|
|
"tool.b": {"priority": 0, "keywords": ["read"]},
|
|
}
|
|
|
|
assert normalize(index) == {"tool.a": (3, ("search", "find")), "tool.b": (0, ("read",))}
|
|
|
|
|
|
def test_entry_without_keywords_is_dropped():
|
|
"""Existing rule: an entry that cannot match text is not indexed."""
|
|
assert normalize({"tool.a": {"priority": 1, "keywords": []}}) == {}
|
|
assert normalize({"tool.a": {"priority": 1, "keywords": "search"}}) == {}
|
|
|
|
|
|
def test_empty_name_is_dropped():
|
|
assert normalize({"": {"priority": 1, "keywords": ["k"]}}) == {}
|
|
|
|
|
|
@pytest.mark.parametrize(
|
|
("priority", "expected"),
|
|
[
|
|
pytest.param(float("inf"), 0, id="positive-infinity"),
|
|
pytest.param(float("-inf"), 0, id="negative-infinity"),
|
|
pytest.param(float("nan"), 0, id="nan"),
|
|
pytest.param(None, 0, id="none"),
|
|
pytest.param("invalid", 0, id="invalid-string"),
|
|
pytest.param([], 0, id="invalid-list"),
|
|
pytest.param("7", 7, id="numeric-string"),
|
|
pytest.param(7.5, 7, id="finite-float"),
|
|
pytest.param(-2, -2, id="negative-int"),
|
|
pytest.param(True, 1, id="bool"),
|
|
],
|
|
)
|
|
def test_builder_and_serialized_index_share_priority_coercion(priority, expected):
|
|
@tool
|
|
def routed_search(query: str) -> str:
|
|
"""Search test data."""
|
|
return query
|
|
|
|
routing = {"mode": "prefer", "priority": priority, "keywords": ["search"]}
|
|
tag_mcp_routing(tag_mcp_tool(routed_search), routing)
|
|
tools, setup = assemble_deferred_tools([routed_search], enabled=True)
|
|
|
|
middleware = build_mcp_routing_middleware(tools, setup, top_k=1)
|
|
|
|
assert isinstance(middleware, McpRoutingMiddleware)
|
|
assert middleware._routing_index == normalize({"routed_search": routing}) == {"routed_search": (expected, ("search",))}
|
|
assert middleware._matched_names({"messages": [HumanMessage(content="search")]}) == []
|