Files
multi-agent-mux/tests/test_issue78_hints_codex.py
T

325 lines
12 KiB
Python

"""Issue #78: in_use_hints lists lsof holders, and codex purge removes
tui-thread-reference-capabilities/<uuid>.
"""
from __future__ import annotations
import json
import os
import sqlite3
import subprocess
import sys
import uuid as uuidlib
from pathlib import Path
import pytest
REPO_ROOT = Path(__file__).resolve().parent.parent
SKILLS = REPO_ROOT / ".agents" / "skills"
if str(SKILLS) not in sys.path:
sys.path.insert(0, str(SKILLS))
from lib_py.agents.base import DiscoveryContext, file_holders # noqa: E402
from lib_py.agents.registry import get_adapter # noqa: E402
pytestmark = pytest.mark.usefixtures("own_workspace_root")
def _clean_env(extra: dict | None = None) -> dict:
env = {
k: v
for k, v in os.environ.items()
if not k.startswith("XDG_")
and not k.startswith("HERDR_")
and k
not in (
"CLAUDE_CODE_SESSION_ID",
"ANTIGRAVITY_CONVERSATION_ID",
"HERMES_SESSION_ID",
"GROK_SESSION_ID",
"OPENCODE_SESSION_ID",
"CLAUDE_CONFIG_DIR",
"CLAUDE_PROJECT_DIR",
"CODEX_HOME",
)
}
env["PYTHONPATH"] = str(SKILLS)
if extra:
env.update(extra)
return env
def _write_fake_lsof(bindir: Path, mapping: dict[str, list[int]]) -> Path:
"""Fake lsof that prints mapped PIDs when any argument matches a key substring."""
bindir.mkdir(parents=True, exist_ok=True)
cases = []
for needle, pids in mapping.items():
echoes = "\n".join(f" echo {p}" for p in pids)
cases.append(f" *{needle}*)\n{echoes}\n found=1 ;;")
body = "\n".join(cases)
exe = bindir / "lsof"
exe.write_text(
"#!/bin/sh\n"
"found=0\n"
'for a in "$@"; do\n'
' case "$a" in\n'
f"{body}\n"
" esac\n"
"done\n"
'[ "$found" = "1" ] && exit 0\n'
"exit 1\n",
encoding="utf-8",
)
exe.chmod(0o755)
return exe
def _materialize_artifact(agent: str, conv: str, ctx: DiscoveryContext) -> str:
adapter = get_adapter(agent)
if agent == "agy":
# Hold the presence lock so agy's in_use_paths (_artifact_paths) is exercised too.
paths = adapter._artifact_paths(conv, ctx)
target = paths[-1]
else:
target = adapter.artifact_path(conv, ctx)
Path(target).parent.mkdir(parents=True, exist_ok=True)
Path(target).write_text("{}\n", encoding="utf-8")
return target
@pytest.mark.parametrize("agent", ["claude", "agy", "codex", "grok", "hermes"])
def test_in_use_hints_list_lsof_holders(tmp_path, monkeypatch, agent):
home = tmp_path / "home"
ws = tmp_path / "ws"
bindir = tmp_path / "bin"
no_lsof_bin = tmp_path / "no-lsof-bin"
for d in (home, ws, bindir, no_lsof_bin):
d.mkdir(parents=True, exist_ok=True)
# Populate no_lsof_bin with symlinks to basic tools (ps) but not lsof.
for tool in ("ps", "sh", "bash"):
for sysdir in ("/bin", "/usr/bin"):
cand = Path(sysdir) / tool
if cand.exists() and not (no_lsof_bin / tool).exists():
(no_lsof_bin / tool).symlink_to(cand)
conv = str(uuidlib.uuid4())
monkeypatch.setenv("HOME", str(home))
monkeypatch.setenv("HOME_DIR", str(home))
monkeypatch.setenv("CODEX_HOME", str(home / ".codex"))
ctx = DiscoveryContext(workspace=str(ws), agent_name=agent, home_dir=str(home))
target = _materialize_artifact(agent, conv, ctx)
adapter = get_adapter(agent)
holder = subprocess.Popen(["sleep", "300"])
try:
_write_fake_lsof(bindir, {conv: [holder.pid]})
env = _clean_env({
"HOME": str(home),
"HOME_DIR": str(home),
"CODEX_HOME": str(home / ".codex"),
"PATH": f"{bindir}:{os.environ.get('PATH', '')}",
})
res = subprocess.run(
[sys.executable, "-m", "lib_py.agents", "in-use", agent, conv, str(ws)],
capture_output=True,
text=True,
env=env,
timeout=30,
)
assert res.returncode == 4, res.stdout + res.stderr
assert json.loads(res.stdout) == {"agent": agent, "in_use": True}
assert f"ps -axww | grep {conv}" in res.stderr
assert f"held open by pid {holder.pid}:" in res.stderr
assert "sleep 300" in res.stderr
assert "not stopped by MAM" in res.stderr
monkeypatch.setenv("PATH", env["PATH"])
hints = adapter.in_use_hints(conv, ctx)
if agent == "claude":
assert hints[0].startswith("a claude process whose session file under ")
assert f"ps -axww | grep {conv}" in hints[1]
else:
assert f"ps -axww | grep {conv}" in hints[0]
assert any(f"held open by pid {holder.pid}:" in h and "sleep 300" in h for h in hints)
assert target in adapter.in_use_paths(conv, ctx)
# Control: without lsof on PATH, file_holders returns [] and hints match the pre-#78 lines.
monkeypatch.setenv("PATH", str(no_lsof_bin))
assert file_holders([target]) == []
hints_no_lsof = adapter.in_use_hints(conv, ctx)
assert not any("held open by pid" in h for h in hints_no_lsof)
assert any(f"ps -axww | grep {conv}" in h for h in hints_no_lsof)
finally:
holder.kill()
holder.wait()
def test_in_use_hints_opencode_skip_service(tmp_path, monkeypatch):
home = tmp_path / "home"
ws = tmp_path / "ws"
bindir = tmp_path / "bin"
for d in (home, ws, bindir):
d.mkdir(parents=True, exist_ok=True)
db = home / ".local" / "share" / "opencode" / "opencode.db"
db.parent.mkdir(parents=True, exist_ok=True)
db.write_text("", encoding="utf-8")
adapter = get_adapter("opencode")
monkeypatch.setattr(adapter, "_bin", lambda ctx: None)
monkeypatch.setenv("HOME", str(home))
monkeypatch.setenv("HOME_DIR", str(home))
ctx = DiscoveryContext(workspace=str(ws), agent_name="opencode", home_dir=str(home))
service_proc = subprocess.Popen(
[sys.executable, "-c", "import time; time.sleep(300)", "serve", "--service"]
)
holder_proc = subprocess.Popen(["sleep", "300"])
try:
_write_fake_lsof(bindir, {"opencode.db": [service_proc.pid, holder_proc.pid]})
monkeypatch.setenv("PATH", f"{bindir}:{os.environ.get('PATH', '')}")
hints = adapter.in_use_hints("ses_test78", ctx)
assert hints[0] == "any process with it on its command line (an editor, a grep, ...): ps -axww | grep ses_test78"
assert any(f"held open by pid {holder_proc.pid}:" in h and "sleep 300" in h for h in hints)
assert not any(f"pid {service_proc.pid}" in h or "serve --service" in h for h in hints)
assert hints[-1] == f"another opencode session holding {db}: lsof {db}"
finally:
for p in (service_proc, holder_proc):
p.kill()
p.wait()
def test_stop_purge_refusal_prints_holders(mam_sandbox, mock_herdr, monkeypatch):
conv = str(uuidlib.uuid4())
name = "issue78-reviewer-claude"
ws = mam_sandbox
ctx = DiscoveryContext(workspace=str(ws), agent_name="claude", home_dir=str(mam_sandbox))
transcript = Path(get_adapter("claude").artifact_path(conv, ctx))
transcript.parent.mkdir(parents=True, exist_ok=True)
transcript.write_text(json.dumps({"sessionId": conv}) + "\n", encoding="utf-8")
lib = mam_sandbox / ".agents" / "skills" / "lib.sh"
yaml_path = mam_sandbox / ".mam" / "agent-sessions.yaml"
mutation = (
f"d['herdr_sessions'] = [{{'name': {name!r}, 'status': 'running', 'agent': 'claude', "
f"'role': 'reviewer', 'claude_session_id_own': {conv!r}, "
f"'pane': {{'cwd': {os.path.realpath(str(ws))!r}}}}}]\n"
)
res_init = subprocess.run(
["bash", "-c", f'source "{lib}" && atomic_dump_yaml "{yaml_path}"'],
input=mutation,
capture_output=True,
text=True,
env=_clean_env({
"HOME": str(mam_sandbox),
"HOME_DIR": str(mam_sandbox),
"WORKSPACE_ROOT": str(mam_sandbox),
"AGENT_SESSIONS_YAML": str(yaml_path),
}),
)
assert res_init.returncode == 0, res_init.stdout + res_init.stderr
holder = subprocess.Popen(["sleep", "300"])
try:
bindir = mam_sandbox / ".local" / "bin"
_write_fake_lsof(bindir, {conv: [holder.pid]})
stop_sh = mam_sandbox / ".agents" / "skills" / "multi-agent-mux-stop" / "scripts" / "stop_session.sh"
env = _clean_env({
"HOME": str(mam_sandbox),
"HOME_DIR": str(mam_sandbox),
"CLAUDE_PROJECT_DIR": str(mam_sandbox / ".claude" / "projects"),
"WORKSPACE_ROOT": str(mam_sandbox),
"AGENT_SESSIONS_YAML": str(yaml_path),
"PATH": f"{bindir}:{os.environ.get('PATH', '')}",
})
res = subprocess.run(
["bash", str(stop_sh), "--session", name, "--agent", "claude", "--purge-conversation", "--yes"],
capture_output=True,
text=True,
cwd=str(mam_sandbox),
env=env,
timeout=60,
)
assert res.returncode == 4, res.stdout + res.stderr
assert f"held open by pid {holder.pid}:" in res.stderr
assert "sleep 300" in res.stderr
assert transcript.exists()
assert name in yaml_path.read_text(encoding="utf-8")
finally:
holder.kill()
holder.wait()
@pytest.mark.parametrize("delete_ok", [True, False], ids=["delete_ok", "delete_fails"])
def test_codex_purge_removes_tui_thread_capability(tmp_path, monkeypatch, delete_ok):
home = tmp_path / "home"
codex_home = tmp_path / "codex-home"
ws = tmp_path / "ws"
bindir = tmp_path / "bin"
for d in (home, codex_home, ws, bindir):
d.mkdir(parents=True, exist_ok=True)
conv = str(uuidlib.uuid4())
other = str(uuidlib.uuid4())
rollout = codex_home / "sessions" / "2026" / "10" / "28" / f"rollout-2026-10-28T00-00-00-{conv}.jsonl"
rollout.parent.mkdir(parents=True, exist_ok=True)
rollout.write_text(json.dumps({"payload": {"id": conv, "cwd": str(ws)}}) + "\n", encoding="utf-8")
cap_dir = codex_home / "tui-thread-reference-capabilities"
cap_dir.mkdir(parents=True, exist_ok=True)
cap_mine = cap_dir / conv
cap_other = cap_dir / other
cap_mine.write_text("cap-mine", encoding="utf-8")
cap_other.write_text("cap-other", encoding="utf-8")
state_db = codex_home / "state_5.sqlite"
conn = sqlite3.connect(str(state_db))
conn.execute("CREATE TABLE threads (id TEXT PRIMARY KEY, rollout_path TEXT, cwd TEXT)")
conn.execute("INSERT INTO threads VALUES (?, ?, ?)", (conv, str(rollout), str(ws)))
conn.execute("INSERT INTO threads VALUES (?, ?, ?)", (other, "", str(ws)))
conn.commit()
conn.close()
fake_codex = bindir / "codex"
if delete_ok:
# Simulate `codex delete <uuid> --force` removing rollout + DB row, leaving tui-thread-reference-capabilities.
fake_codex.write_text(
f"#!{sys.executable}\n"
"import os, sqlite3, sys\n"
"if len(sys.argv) >= 3 and sys.argv[1] == 'delete':\n"
f" p = {str(rollout)!r}\n"
" if os.path.exists(p):\n"
" os.remove(p)\n"
f" conn = sqlite3.connect({str(state_db)!r})\n"
" conn.execute('DELETE FROM threads WHERE id=?', (sys.argv[2],))\n"
" conn.commit()\n"
" conn.close()\n"
" sys.exit(0)\n"
"sys.exit(1)\n",
encoding="utf-8",
)
else:
fake_codex.write_text("#!/bin/sh\nexit 1\n", encoding="utf-8")
fake_codex.chmod(0o755)
monkeypatch.setenv("HOME", str(home))
monkeypatch.setenv("HOME_DIR", str(home))
monkeypatch.setenv("CODEX_HOME", str(codex_home))
monkeypatch.setenv("PATH", f"{bindir}:{os.environ.get('PATH', '')}")
adapter = get_adapter("codex")
ctx = DiscoveryContext(workspace=str(ws), agent_name="codex", home_dir=str(home))
assert str(cap_mine) in adapter.residual_artifacts(conv, ctx)
purged = adapter.purge_artifacts(conv, ctx)
assert str(cap_mine) in purged
assert not cap_mine.exists()
assert cap_other.exists()
assert not rollout.exists()
assert adapter.residual_artifacts(conv, ctx) == []
assert str(cap_other) in adapter.residual_artifacts(other, ctx)