mirror of
https://github.com/NousResearch/hermes-agent.git
synced 2026-07-31 19:16:29 +00:00
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.
140 lines
5.7 KiB
Python
140 lines
5.7 KiB
Python
"""Regression tests for issue #8340.
|
|
|
|
When a user command backgrounds a child process (``cmd &``, ``setsid cmd &
|
|
disown``, etc.), the backgrounded grandchild inherits the write-end of our
|
|
stdout pipe via fork(). Before the fix, the drain thread's blocking
|
|
``for line in proc.stdout`` would never see EOF until that grandchild
|
|
closed the pipe — causing the terminal tool to hang for the full lifetime
|
|
of the backgrounded service (indefinitely for a uvicorn server).
|
|
|
|
The fix switches ``_drain()`` to select()-based non-blocking reads and
|
|
stops draining shortly after bash exits even if the pipe hasn't EOF'd.
|
|
"""
|
|
import subprocess
|
|
import time
|
|
|
|
import pytest
|
|
|
|
from tools.environments.local import LocalEnvironment
|
|
|
|
|
|
def _pkill(pattern: str) -> None:
|
|
subprocess.run(f"pkill -9 -f {pattern!r} 2>/dev/null", shell=True)
|
|
|
|
|
|
@pytest.fixture
|
|
def local_env():
|
|
env = LocalEnvironment(cwd="/tmp")
|
|
try:
|
|
yield env
|
|
finally:
|
|
env.cleanup()
|
|
|
|
|
|
class TestBackgroundChildDoesNotHang:
|
|
"""Regression guard for issue #8340."""
|
|
|
|
def test_plain_background_returns_promptly(self, local_env):
|
|
"""``cmd &`` with no output redirection must not hang on pipe inherit."""
|
|
marker = "hermes_8340_plain_bg"
|
|
cmd = f'python3 -c "import time; time.sleep(60)" & echo {marker}'
|
|
try:
|
|
t0 = time.monotonic()
|
|
result = local_env.execute(cmd, timeout=15)
|
|
elapsed = time.monotonic() - t0
|
|
|
|
assert elapsed < 10.0, ( # hang under guard is 15s+; loose bound rides out runner stalls
|
|
f"terminal_tool hung for {elapsed:.1f}s — drain thread "
|
|
f"is still blocking on backgrounded child's inherited pipe fd"
|
|
)
|
|
assert result["returncode"] == 0
|
|
assert marker in result["output"]
|
|
finally:
|
|
_pkill("time.sleep(60)")
|
|
|
|
def test_setsid_disown_pattern_returns_promptly(self, local_env):
|
|
"""The exact pattern from the issue: setsid ... & disown."""
|
|
cmd = (
|
|
'setsid python3 -c "import time; time.sleep(60)" '
|
|
'> /dev/null 2>&1 < /dev/null & disown; echo started'
|
|
)
|
|
try:
|
|
t0 = time.monotonic()
|
|
result = local_env.execute(cmd, timeout=15)
|
|
elapsed = time.monotonic() - t0
|
|
|
|
assert elapsed < 10.0, f"setsid+disown path hung for {elapsed:.1f}s"
|
|
assert result["returncode"] == 0
|
|
assert "started" in result["output"]
|
|
finally:
|
|
_pkill("time.sleep(60)")
|
|
|
|
|
|
def test_default_capture_is_full_fidelity_for_internal_consumers(
|
|
self, local_env
|
|
):
|
|
"""Default execute() (no bounded_capture) must return complete output.
|
|
|
|
Internal consumers — file-operation ``cat`` reads that feed the patch
|
|
engine, code-execution RPC reads, log reads — rely on full-fidelity
|
|
capture. Bounding them at tool_output.max_bytes would CORRUPT files
|
|
on read-modify-write (#64435 review finding), so only the foreground
|
|
terminal tool opts in via bounded_capture=True.
|
|
"""
|
|
# ~200 KB — four times the default 50 KB cap.
|
|
command = (
|
|
"python3 -c \"import sys; "
|
|
"sys.stdout.write('START-MARK\\n' + ('y' * 200000) + '\\nEND-MARK')\""
|
|
)
|
|
|
|
result = local_env.execute(command, timeout=10)
|
|
|
|
assert result["returncode"] == 0
|
|
assert "[OUTPUT TRUNCATED" not in result["output"]
|
|
assert result["output"].startswith("START-MARK")
|
|
assert result["output"].endswith("END-MARK")
|
|
assert len(result["output"]) > 200000
|
|
|
|
|
|
def test_utf8_multibyte_across_read_boundary(self, local_env):
|
|
"""Multibyte UTF-8 characters straddling a 4096-byte ``os.read()`` boundary
|
|
must be decoded correctly via the incremental decoder — not lost to a
|
|
``UnicodeDecodeError`` fallback. Regression for a bug in the first draft
|
|
of the fix where a strict ``bytes.decode('utf-8')`` on each raw chunk
|
|
wiped the entire buffer as soon as any chunk split a multi-byte char.
|
|
"""
|
|
# 10000 "日" chars = 30000 bytes — guaranteed to cross multiple 4096
|
|
# read boundaries, and most boundaries will land in the middle of the
|
|
# 3-byte UTF-8 encoding of U+65E5.
|
|
cmd = (
|
|
'python3 -c \'import sys; '
|
|
'sys.stdout.buffer.write(chr(0x65e5).encode("utf-8") * 10000); '
|
|
'sys.stdout.buffer.write(b"\\n")\''
|
|
)
|
|
result = local_env.execute(cmd, timeout=10)
|
|
assert result["returncode"] == 0
|
|
# All 10000 characters must survive the round-trip
|
|
assert result["output"].count("\u65e5") == 10000, (
|
|
f"lost multibyte chars across read boundaries: got "
|
|
f"{result['output'].count(chr(0x65e5))} / 10000"
|
|
)
|
|
# And the "[binary output detected ...]" fallback must NOT fire
|
|
assert "binary output detected" not in result["output"]
|
|
|
|
def test_invalid_utf8_uses_replacement_not_fallback(self, local_env):
|
|
"""Truly invalid byte sequences must be substituted with U+FFFD (matching
|
|
the pre-fix ``errors='replace'`` behaviour of the old ``TextIOWrapper``
|
|
drain), not clobber the entire buffer with a fallback placeholder.
|
|
"""
|
|
# Write a deliberate invalid UTF-8 lead byte sandwiched between valid ASCII
|
|
cmd = (
|
|
'python3 -c \'import sys; '
|
|
'sys.stdout.buffer.write(b"before "); '
|
|
'sys.stdout.buffer.write(b"\\xff\\xfe"); '
|
|
'sys.stdout.buffer.write(b" after\\n")\''
|
|
)
|
|
result = local_env.execute(cmd, timeout=15)
|
|
assert result["returncode"] == 0
|
|
assert "before" in result["output"]
|
|
assert "after" in result["output"]
|
|
assert "binary output detected" not in result["output"]
|