hermes-agent/tests/tools/test_browser_eval_supervisor_path.py
Teknium 39975613b1
test: prune wave 2 + speed fixes — 28,106 → 19,757 test functions, suite wall 315s → 294s
Second, deeper pass over tools/gateway/hermes_cli plus first pass over
the trees wave 1 missed (acp, acp_adapter, skills, computer_use, docker,
dashboard, conformance, monitoring, secret_sources, hermes_state,
providers). Same rubric as wave 1 (AGENTS.md test policy); security,
alternation/caching invariants, issue-number regressions, and E2E kept.

Real test-quality fixes found and rooted out along the way:
- tests/tools/test_command_guards.py made real auxiliary-LLM HTTPS calls
  (DEFAULT_CONFIG smart-approval leaked in) — pinned approval
  mode=manual via autouse fixture: 17.4s → 0.4s.
- test_model_switch_custom_providers.py / test_user_providers_model_switch.py
  silently probed live provider catalogs (~2s/test) — stubbed
  cached_provider_model_ids/provider_model_ids/fetch_api_models.
- test_telegram_noise_filter.py: 15-platform copy-paste matrix over
  shared gateway.run logic → 3 representative platforms (55s → 3.9s).
- test_gateway_shutdown.py: stop()'s 5s interrupt-deadline loop spun on
  MagicMock agents — interrupt.side_effect now clears _running_agents
  (22s → 1.0s).
- test_gateway_inactivity_timeout.py poll-harness timings shrunk 3-5x
  (24s → 1.1s); test_mcp_stability.py backoff/SIGTERM-grace sleeps
  patched (15.4s → 2.5s); test_async_delegation.py negative-drain wait
  5s → 0.5s.
- test_telegram_init_deadline.py: loop-block margin restored to 1.0s
  with rationale comment — the watchdog-dump assertion needs the loop
  blocked well past deadline+grace under parallel load (flaked once in
  the 40-worker verification run at a 0.2s margin).

Verification: full hermetic suite via scripts/run_tests.sh —
2,438 files, 21,718 tests passed, 0 failed, 293.9s wall.
Suite totals vs original baseline: 46,820 → 19,757 test functions
(−57.8%), wall 583.5s → 293.9s (−50%), subprocess CPU 13,564s → 11,623s.
2026-07-29 13:39:40 -07:00

305 lines
10 KiB
Python

"""Unit tests for the supervisor-WS fast path in browser_console / _browser_eval.
These exercise the dispatch logic in ``tools.browser_tool._browser_eval`` and
the response shaping in ``CDPSupervisor.evaluate_runtime`` using mocks — no
real browser, no real WebSocket. Real-CDP coverage lives in
``tests/tools/test_browser_supervisor.py`` (gated on Chrome being installed).
"""
from __future__ import annotations
import json
from unittest.mock import MagicMock
import pytest
# ---------------------------------------------------------------------------
# Fast-path dispatch: tools.browser_tool._browser_eval
# ---------------------------------------------------------------------------
@pytest.fixture(autouse=True)
def _disable_camofox(monkeypatch):
"""Force the non-camofox path so our supervisor branch is reached."""
import tools.browser_tool as bt
monkeypatch.setattr(bt, "_is_camofox_mode", lambda: False)
monkeypatch.setattr(bt, "_last_session_key", lambda task_id: "test-task")
def _patch_supervisor(monkeypatch, supervisor):
"""Wire SUPERVISOR_REGISTRY.get to return ``supervisor`` for any task_id."""
import tools.browser_supervisor as bs
registry = MagicMock()
registry.get.return_value = supervisor
monkeypatch.setattr(bs, "SUPERVISOR_REGISTRY", registry)
return registry
class TestBrowserEvalSupervisorPath:
"""The supervisor fast path replaces the agent-browser subprocess hop."""
def test_primitive_result_routes_through_supervisor(self, monkeypatch):
import tools.browser_tool as bt
sup = MagicMock()
sup.evaluate_runtime.return_value = {
"ok": True,
"result": 42,
"result_type": "number",
}
_patch_supervisor(monkeypatch, sup)
# If the subprocess path is hit we want a loud failure.
monkeypatch.setattr(
bt, "_run_browser_command",
lambda *a, **kw: pytest.fail("subprocess path must not run when supervisor is healthy"),
)
out = json.loads(bt._browser_eval("1 + 41"))
assert out["success"] is True
assert out["result"] == 42
assert out["method"] == "cdp_supervisor"
sup.evaluate_runtime.assert_called_once_with("1 + 41")
def test_json_string_result_is_parsed(self, monkeypatch):
"""Match agent-browser semantics: JSON-string results get parsed."""
import tools.browser_tool as bt
sup = MagicMock()
sup.evaluate_runtime.return_value = {
"ok": True,
"result": '{"a": 1, "b": [2, 3]}',
"result_type": "string",
}
_patch_supervisor(monkeypatch, sup)
monkeypatch.setattr(
bt, "_run_browser_command",
lambda *a, **kw: pytest.fail("subprocess path must not run"),
)
out = json.loads(bt._browser_eval('JSON.stringify({a:1,b:[2,3]})'))
assert out["success"] is True
assert out["result"] == {"a": 1, "b": [2, 3]}
# result_type reflects the parsed Python type, not the raw JS type.
assert out["result_type"] == "dict"
def test_subprocess_reference_chain_error_becomes_guidance(self, monkeypatch):
"""The CLI subprocess can't retry with returnByValue=False, so the
cryptic 'Object reference chain is too long' CDP error must be turned
into actionable guidance instead of surfaced raw."""
import tools.browser_tool as bt
# No supervisor → subprocess path runs.
_patch_supervisor(monkeypatch, None)
def _fake_subprocess(task_id, cmd, args):
assert cmd == "eval"
return {
"success": False,
"error": "Runtime.evaluate failed: Object reference chain is too long",
}
monkeypatch.setattr(bt, "_run_browser_command", _fake_subprocess)
out = json.loads(bt._browser_eval("document.body"))
assert out["success"] is False
# Raw protocol error must NOT leak through.
assert "reference chain" not in out["error"].lower()
# Actionable guidance instead.
assert "primitive" in out["error"].lower()
assert "DOM node" in out["error"] or "dom node" in out["error"].lower()
# ---------------------------------------------------------------------------
# Response shaping: CDPSupervisor.evaluate_runtime
# ---------------------------------------------------------------------------
def _make_supervisor_with_cdp(cdp_response):
"""Build a CDPSupervisor instance that mocks ``_cdp`` to return ``cdp_response``.
Bypasses ``__init__`` entirely so we don't need a real WS connection. We
set just the state ``evaluate_runtime`` reads.
"""
import asyncio
import threading
from tools.browser_supervisor import CDPSupervisor
sup = object.__new__(CDPSupervisor)
sup._state_lock = threading.Lock()
sup._active = True
sup._page_session_id = "test-session-id"
# Build a real running event loop on a background thread so
# asyncio.run_coroutine_threadsafe has somewhere to dispatch.
loop = asyncio.new_event_loop()
def _runner():
asyncio.set_event_loop(loop)
loop.run_forever()
thread = threading.Thread(target=_runner, daemon=True)
thread.start()
async def _fake_cdp(method, params=None, *, session_id=None, timeout=10.0):
return cdp_response
sup._cdp = _fake_cdp # type: ignore[method-assign]
sup._loop = loop
sup._thread = thread
return sup
def _stop_supervisor(sup):
sup._loop.call_soon_threadsafe(sup._loop.stop)
sup._thread.join(timeout=2)
class TestEvaluateRuntimeResponseShaping:
"""CDPSupervisor.evaluate_runtime decodes the Runtime.evaluate response correctly."""
def test_primitive_value(self):
sup = _make_supervisor_with_cdp({
"id": 1,
"result": {"result": {"type": "number", "value": 42}},
})
try:
out = sup.evaluate_runtime("1 + 41")
assert out == {"ok": True, "result": 42, "result_type": "number"}
finally:
_stop_supervisor(sup)
def test_object_value_returned_by_value(self):
sup = _make_supervisor_with_cdp({
"id": 1,
"result": {
"result": {
"type": "object",
"value": {"foo": "bar", "n": 7},
}
},
})
try:
out = sup.evaluate_runtime('({foo:"bar", n:7})')
assert out["ok"] is True
assert out["result"] == {"foo": "bar", "n": 7}
assert out["result_type"] == "object"
finally:
_stop_supervisor(sup)
def test_no_session_attached_returns_error(self):
import asyncio
import threading
from tools.browser_supervisor import CDPSupervisor
sup = object.__new__(CDPSupervisor)
sup._state_lock = threading.Lock()
sup._active = True
sup._page_session_id = None # ← attach hasn't happened yet
loop = asyncio.new_event_loop()
thread = threading.Thread(
target=lambda: (asyncio.set_event_loop(loop), loop.run_forever()),
daemon=True,
)
thread.start()
sup._loop = loop
try:
out = sup.evaluate_runtime("1+1")
assert out["ok"] is False
assert "session" in out["error"].lower()
finally:
loop.call_soon_threadsafe(loop.stop)
thread.join(timeout=2)
def _make_supervisor_with_cdp_fn(cdp_fn):
"""Like ``_make_supervisor_with_cdp`` but lets the test supply a coroutine
function as ``_cdp`` so behaviour can vary by params (e.g. returnByValue).
"""
import asyncio
import threading
from tools.browser_supervisor import CDPSupervisor
sup = object.__new__(CDPSupervisor)
sup._state_lock = threading.Lock()
sup._active = True
sup._page_session_id = "test-session-id"
loop = asyncio.new_event_loop()
def _runner():
asyncio.set_event_loop(loop)
loop.run_forever()
thread = threading.Thread(target=_runner, daemon=True)
thread.start()
sup._cdp = cdp_fn # type: ignore[method-assign]
sup._loop = loop
sup._thread = thread
return sup
class TestEvaluateRuntimeDomNodeCrashRetry:
"""returnByValue=True on a DOM node fails CDP serialization with 'Object
reference chain is too long'. evaluate_runtime must retry with
returnByValue=False and return the node's description instead of crashing.
"""
def test_reference_chain_crash_retries_without_by_value(self):
calls = []
async def _fake_cdp(method, params=None, *, session_id=None, timeout=10.0):
by_value = (params or {}).get("returnByValue")
calls.append(by_value)
if by_value:
# Mirror _read_loop turning a top-level CDP error into a RuntimeError.
raise RuntimeError(
"CDP error on id=7: {'code': -32000, "
"'message': 'Object reference chain is too long'}"
)
# returnByValue=False: Chrome returns the node's description, no value.
return {
"id": 8,
"result": {
"result": {
"type": "object",
"subtype": "node",
"description": "body",
}
},
}
sup = _make_supervisor_with_cdp_fn(_fake_cdp)
try:
out = sup.evaluate_runtime("document.body")
assert out["ok"] is True
assert out["result"] == "body"
assert out["result_type"] == "object"
# First call by_value=True (crashed), retried with by_value=False.
assert calls == [True, False]
finally:
_stop_supervisor(sup)
def test_unrelated_error_does_not_retry(self):
calls = []
async def _fake_cdp(method, params=None, *, session_id=None, timeout=10.0):
calls.append((params or {}).get("returnByValue"))
raise RuntimeError("CDP error on id=3: {'message': 'Target closed'}")
sup = _make_supervisor_with_cdp_fn(_fake_cdp)
try:
out = sup.evaluate_runtime("document.body")
assert out["ok"] is False
assert "Target closed" in out["error"]
# No retry for unrelated failures — exactly one call.
assert calls == [True]
finally:
_stop_supervisor(sup)