mirror of
https://github.com/NousResearch/hermes-agent.git
synced 2026-07-22 16:25:58 +00:00
c80b244b52
The per-session record store is now the ONLY cwd mechanism. Deleted: - env.cwd_owner stamping + prev_owner threading (terminal_tool): the shared env no longer carries ownership metadata at all - _resolve_command_cwd's env/prev_owner params: resolution is workdir > session record > config/override default - file_tools._live_cwd_if_owned + _get_live_tracking_cwd: path resolution never consults the shared env's live cwd - file_tools._last_known_cwd + _remember_last_known_cwd + _last_known_cwd_for: the #26211 preserved-anchor registry is subsumed by the session record, which never lived on the env and therefore cannot be lost to env cleanup. The _get_file_ops stale-cache rescue now writes the record instead. - env recreation (both _get_file_ops and terminal_tool) seeds the fresh env from override > session record > config Why no transition fallback: the legacy state was process-local and in-memory exactly like the record store — after a restart both start empty, and within a running process every legacy write site has been dual-writing the record since step 1. There is no populated-legacy/ empty-record state to fall back for. Tests updated to drive the record store instead of the deleted mechanism; the cross-session isolation suite now asserts the same behavior contracts (no leak, cd isolation, #26211 persistence) against the new architecture, plus a new "session C inherits nothing" case that the old ownership guard could not express.
335 lines
13 KiB
Python
335 lines
13 KiB
Python
"""Regression tests for task/session cwd propagation in terminal_tool."""
|
|
|
|
import json
|
|
from types import SimpleNamespace
|
|
|
|
import tools.terminal_tool as terminal_tool
|
|
|
|
|
|
def _minimal_terminal_config(cwd="/default"):
|
|
return {
|
|
"env_type": "local",
|
|
"cwd": cwd,
|
|
"timeout": 60,
|
|
"lifetime_seconds": 3600,
|
|
}
|
|
|
|
|
|
def test_foreground_command_uses_registered_task_cwd_for_existing_environment(monkeypatch):
|
|
"""ACP can update task cwd after the local env exists; foreground must honor it."""
|
|
calls = []
|
|
|
|
class FakeEnv:
|
|
env = {}
|
|
|
|
def execute(self, command, **kwargs):
|
|
calls.append((command, kwargs))
|
|
return {"output": "ok", "returncode": 0}
|
|
|
|
task_id = "acp-session-1"
|
|
monkeypatch.setattr(terminal_tool, "_active_environments", {task_id: FakeEnv()})
|
|
monkeypatch.setattr(terminal_tool, "_last_activity", {})
|
|
monkeypatch.setattr(terminal_tool, "_task_env_overrides", {task_id: {"cwd": "/workspace/acp"}})
|
|
monkeypatch.setattr(terminal_tool, "_get_env_config", lambda: _minimal_terminal_config())
|
|
monkeypatch.setattr(
|
|
terminal_tool,
|
|
"_check_all_guards",
|
|
lambda command, env_type, **kwargs: {"approved": True},
|
|
)
|
|
|
|
result = json.loads(terminal_tool.terminal_tool(command="pwd", task_id=task_id))
|
|
|
|
assert result["exit_code"] == 0
|
|
assert calls == [("pwd", {"timeout": 60, "cwd": "/workspace/acp", "bounded_capture": True})]
|
|
|
|
|
|
def test_explicit_workdir_still_wins_over_registered_task_cwd(monkeypatch):
|
|
calls = []
|
|
|
|
class FakeEnv:
|
|
env = {}
|
|
|
|
def execute(self, command, **kwargs):
|
|
calls.append(kwargs)
|
|
return {"output": "ok", "returncode": 0}
|
|
|
|
task_id = "acp-session-1"
|
|
monkeypatch.setattr(terminal_tool, "_active_environments", {task_id: FakeEnv()})
|
|
monkeypatch.setattr(terminal_tool, "_last_activity", {})
|
|
monkeypatch.setattr(terminal_tool, "_task_env_overrides", {task_id: {"cwd": "/workspace/acp"}})
|
|
monkeypatch.setattr(terminal_tool, "_get_env_config", lambda: _minimal_terminal_config())
|
|
monkeypatch.setattr(
|
|
terminal_tool,
|
|
"_check_all_guards",
|
|
lambda command, env_type, **kwargs: {"approved": True},
|
|
)
|
|
|
|
result = json.loads(
|
|
terminal_tool.terminal_tool(
|
|
command="pwd",
|
|
task_id=task_id,
|
|
workdir="/explicit/workdir",
|
|
)
|
|
)
|
|
|
|
assert result["exit_code"] == 0
|
|
assert calls == [{"timeout": 60, "cwd": "/explicit/workdir", "bounded_capture": True}]
|
|
|
|
|
|
def test_foreground_command_prefers_recorded_session_cwd_over_init_time_cwd(monkeypatch):
|
|
"""A prior `cd` records the session cwd; terminal_tool must honor it."""
|
|
calls = []
|
|
|
|
class FakeEnv:
|
|
env = {}
|
|
cwd = "/workspace/live"
|
|
|
|
def execute(self, command, **kwargs):
|
|
calls.append((command, kwargs))
|
|
return {"output": "ok", "returncode": 0}
|
|
|
|
task_id = "session-live-cwd"
|
|
monkeypatch.setattr(terminal_tool, "_active_environments", {task_id: FakeEnv()})
|
|
monkeypatch.setattr(terminal_tool, "_last_activity", {})
|
|
monkeypatch.setattr(terminal_tool, "_session_cwd", {})
|
|
monkeypatch.setattr(terminal_tool, "_task_env_overrides", {task_id: {"cwd": "/workspace/init"}})
|
|
monkeypatch.setattr(terminal_tool, "_get_env_config", lambda: _minimal_terminal_config(cwd="/workspace/init"))
|
|
monkeypatch.setattr(terminal_tool, "_start_cleanup_thread", lambda: None)
|
|
monkeypatch.setattr(terminal_tool, "_resolve_container_task_id", lambda value: value or "default")
|
|
monkeypatch.setattr(
|
|
terminal_tool,
|
|
"_check_all_guards",
|
|
lambda command, env_type, **kwargs: {"approved": True},
|
|
)
|
|
# The prior command's completed `cd` recorded the session cwd.
|
|
terminal_tool.record_session_cwd(task_id, "/workspace/live")
|
|
|
|
result = json.loads(terminal_tool.terminal_tool(command="pwd", task_id=task_id))
|
|
|
|
assert result["exit_code"] == 0
|
|
assert calls == [("pwd", {"timeout": 60, "cwd": "/workspace/live", "bounded_capture": True})]
|
|
|
|
|
|
def test_background_command_prefers_recorded_session_cwd_over_init_time_cwd(monkeypatch):
|
|
"""Background process launches must also use the recorded session cwd."""
|
|
|
|
class FakeEnv:
|
|
env = {}
|
|
cwd = "/workspace/live"
|
|
|
|
class FakeRegistry:
|
|
def __init__(self):
|
|
self.calls = []
|
|
self.pending_watchers = []
|
|
|
|
def spawn_local(self, **kwargs):
|
|
self.calls.append(kwargs)
|
|
return SimpleNamespace(id="proc_test", pid=1234)
|
|
|
|
import tools.process_registry as process_registry_mod
|
|
|
|
registry = FakeRegistry()
|
|
task_id = "session-live-cwd-bg"
|
|
monkeypatch.setattr(terminal_tool, "_active_environments", {task_id: FakeEnv()})
|
|
monkeypatch.setattr(terminal_tool, "_last_activity", {})
|
|
monkeypatch.setattr(terminal_tool, "_session_cwd", {})
|
|
monkeypatch.setattr(terminal_tool, "_task_env_overrides", {task_id: {"cwd": "/workspace/init"}})
|
|
monkeypatch.setattr(terminal_tool, "_get_env_config", lambda: _minimal_terminal_config(cwd="/workspace/init"))
|
|
monkeypatch.setattr(terminal_tool, "_start_cleanup_thread", lambda: None)
|
|
monkeypatch.setattr(terminal_tool, "_resolve_container_task_id", lambda value: value or "default")
|
|
monkeypatch.setattr(
|
|
terminal_tool,
|
|
"_check_all_guards",
|
|
lambda command, env_type, **kwargs: {"approved": True},
|
|
)
|
|
monkeypatch.setattr(process_registry_mod, "process_registry", registry)
|
|
terminal_tool.record_session_cwd(task_id, "/workspace/live")
|
|
|
|
result = json.loads(
|
|
terminal_tool.terminal_tool(
|
|
command="sleep 1",
|
|
task_id=task_id,
|
|
background=True,
|
|
)
|
|
)
|
|
|
|
assert result["exit_code"] == 0
|
|
# session_key falls back to the raw task_id when no gateway contextvar is set
|
|
# (it doesn't propagate to tool-worker threads), so process.kill / stop can
|
|
# still find and terminate this background process.
|
|
assert registry.calls == [{
|
|
"command": "sleep 1",
|
|
"cwd": "/workspace/live",
|
|
"task_id": task_id,
|
|
"session_key": task_id,
|
|
"env_vars": {},
|
|
"use_pty": False,
|
|
}]
|
|
|
|
|
|
def test_registering_cwd_override_updates_session_record(monkeypatch):
|
|
"""An ACP ``update_cwd`` (re-)registered mid-session must win over a
|
|
previously ``cd``-ed session cwd.
|
|
|
|
Registration writes the session record directly, so an explicit ACP
|
|
project-root change takes effect on the next command, as the editor
|
|
client expects.
|
|
"""
|
|
|
|
class FakeEnv:
|
|
env = {}
|
|
cwd = "/workspace/old"
|
|
|
|
task_id = "acp-session-update"
|
|
fake_env = FakeEnv()
|
|
monkeypatch.setattr(terminal_tool, "_active_environments", {task_id: fake_env})
|
|
monkeypatch.setattr(terminal_tool, "_task_env_overrides", {})
|
|
monkeypatch.setattr(terminal_tool, "_session_cwd", {})
|
|
# The session had cd'd somewhere before the editor switched project roots.
|
|
terminal_tool.record_session_cwd(task_id, "/workspace/old")
|
|
|
|
terminal_tool.register_task_env_overrides(task_id, {"cwd": "/workspace/new"})
|
|
|
|
# The live env mirror still updates (legacy env seeding) …
|
|
assert fake_env.cwd == "/workspace/new"
|
|
# … and the session record — what commands actually resolve against — too.
|
|
assert terminal_tool.get_session_cwd(task_id) == "/workspace/new"
|
|
assert terminal_tool._resolve_command_cwd(
|
|
workdir=None, default_cwd="/workspace/config", session_key=task_id
|
|
) == "/workspace/new"
|
|
|
|
|
|
def test_registering_cwd_override_noop_when_no_live_env(monkeypatch):
|
|
"""Registering an override before the env exists must not crash; the cwd
|
|
is applied at env creation time instead."""
|
|
monkeypatch.setattr(terminal_tool, "_active_environments", {})
|
|
monkeypatch.setattr(terminal_tool, "_task_env_overrides", {})
|
|
|
|
# Should not raise even though no env is cached yet.
|
|
terminal_tool.register_task_env_overrides("acp-session-pending", {"cwd": "/workspace/new"})
|
|
|
|
assert terminal_tool._task_env_overrides["acp-session-pending"] == {"cwd": "/workspace/new"}
|
|
|
|
|
|
def test_registering_non_cwd_override_leaves_live_env_cwd_untouched(monkeypatch):
|
|
"""A non-cwd override (e.g. a per-task Modal image) must not disturb the
|
|
live env's cwd."""
|
|
|
|
class FakeEnv:
|
|
env = {}
|
|
cwd = "/workspace/keep"
|
|
|
|
task_id = "rl-rollout-1"
|
|
fake_env = FakeEnv()
|
|
monkeypatch.setattr(terminal_tool, "_active_environments", {task_id: fake_env})
|
|
monkeypatch.setattr(terminal_tool, "_task_env_overrides", {})
|
|
|
|
terminal_tool.register_task_env_overrides(task_id, {"modal_image": "custom:latest"})
|
|
|
|
assert fake_env.cwd == "/workspace/keep"
|
|
|
|
|
|
def test_stale_env_cwd_from_different_session_is_ignored(monkeypatch):
|
|
"""A different session's `cd` left env.cwd pointing at its checkout.
|
|
|
|
The terminal env is shared (collapsed to "default"), so env.cwd tracks the
|
|
LAST session that ran a command. When session B claims the env after
|
|
session A left it in A's worktree, the first command must NOT run in A's
|
|
leftover cwd — it must fall through to the config/override cwd (this
|
|
session's own workspace).
|
|
"""
|
|
calls = []
|
|
|
|
class FakeEnv:
|
|
env = {}
|
|
cwd = "/home/user/src/hermes-desktop-tipc/apps/desktop"
|
|
cwd_owner = "session-A-key"
|
|
|
|
def execute(self, command, **kwargs):
|
|
calls.append((command, kwargs))
|
|
return {"output": "ok", "returncode": 0}
|
|
|
|
task_id = "session-B"
|
|
monkeypatch.setattr(terminal_tool, "_active_environments", {"default": FakeEnv()})
|
|
monkeypatch.setattr(terminal_tool, "_last_activity", {})
|
|
monkeypatch.setattr(terminal_tool, "_task_env_overrides", {})
|
|
monkeypatch.setattr(terminal_tool, "_get_env_config", lambda: _minimal_terminal_config(cwd="/home/user/src/hermes-agent"))
|
|
monkeypatch.setattr(terminal_tool, "_start_cleanup_thread", lambda: None)
|
|
monkeypatch.setattr(terminal_tool, "_resolve_container_task_id", lambda value: "default")
|
|
monkeypatch.setattr(
|
|
terminal_tool,
|
|
"_check_all_guards",
|
|
lambda command, env_type, **kwargs: {"approved": True},
|
|
)
|
|
|
|
result = json.loads(terminal_tool.terminal_tool(command="pwd", task_id=task_id))
|
|
|
|
assert result["exit_code"] == 0
|
|
# The command must run in the config cwd (hermes-agent), NOT the stale
|
|
# env.cwd left by session A (hermes-desktop-tipc).
|
|
assert calls == [("pwd", {"timeout": 60, "cwd": "/home/user/src/hermes-agent", "bounded_capture": True})]
|
|
|
|
|
|
def test_same_session_recorded_cwd_survives_across_commands(monkeypatch):
|
|
"""In-session `cd` state survives: the record written by one command is
|
|
used by the next command in the same session."""
|
|
calls = []
|
|
|
|
class FakeEnv:
|
|
env = {}
|
|
cwd = "/workspace/deep"
|
|
|
|
def execute(self, command, **kwargs):
|
|
calls.append((command, kwargs))
|
|
return {"output": "ok", "returncode": 0}
|
|
|
|
env = FakeEnv()
|
|
task_id = "session-X"
|
|
monkeypatch.setattr(terminal_tool, "_active_environments", {"default": env})
|
|
monkeypatch.setattr(terminal_tool, "_last_activity", {})
|
|
monkeypatch.setattr(terminal_tool, "_task_env_overrides", {})
|
|
monkeypatch.setattr(terminal_tool, "_session_cwd", {})
|
|
monkeypatch.setattr(terminal_tool, "_get_env_config", lambda: _minimal_terminal_config(cwd="/workspace/config"))
|
|
monkeypatch.setattr(terminal_tool, "_start_cleanup_thread", lambda: None)
|
|
monkeypatch.setattr(terminal_tool, "_resolve_container_task_id", lambda value: "default")
|
|
monkeypatch.setattr(
|
|
terminal_tool,
|
|
"_check_all_guards",
|
|
lambda command, env_type, **kwargs: {"approved": True},
|
|
)
|
|
|
|
# First command runs in the config cwd (no record yet) and afterwards
|
|
# mirrors the env's post-command cwd into the session record.
|
|
result = json.loads(terminal_tool.terminal_tool(command="pwd", task_id=task_id))
|
|
assert result["exit_code"] == 0
|
|
assert calls[0] == ("pwd", {"timeout": 60, "cwd": "/workspace/config", "bounded_capture": True})
|
|
assert terminal_tool.get_session_cwd(task_id) == "/workspace/deep"
|
|
|
|
# Second command in the same session trusts the record.
|
|
result = json.loads(terminal_tool.terminal_tool(command="pwd", task_id=task_id))
|
|
assert result["exit_code"] == 0
|
|
assert calls[1] == ("pwd", {"timeout": 60, "cwd": "/workspace/deep", "bounded_capture": True})
|
|
|
|
|
|
def test_safe_getcwd_returns_real_cwd(monkeypatch):
|
|
monkeypatch.setattr(terminal_tool.os, "getcwd", lambda: "/home/user/project")
|
|
assert terminal_tool._safe_getcwd() == "/home/user/project"
|
|
|
|
|
|
def test_safe_getcwd_falls_back_to_terminal_cwd_when_cwd_deleted(monkeypatch):
|
|
def _boom():
|
|
raise FileNotFoundError("[Errno 2] No such file or directory")
|
|
|
|
monkeypatch.setattr(terminal_tool.os, "getcwd", _boom)
|
|
monkeypatch.setenv("TERMINAL_CWD", "/srv/work")
|
|
assert terminal_tool._safe_getcwd() == "/srv/work"
|
|
|
|
|
|
def test_safe_getcwd_falls_back_to_home_when_no_terminal_cwd(monkeypatch):
|
|
def _boom():
|
|
raise FileNotFoundError()
|
|
|
|
monkeypatch.setattr(terminal_tool.os, "getcwd", _boom)
|
|
monkeypatch.delenv("TERMINAL_CWD", raising=False)
|
|
monkeypatch.setattr(terminal_tool.os.path, "expanduser", lambda p: "/home/me")
|
|
assert terminal_tool._safe_getcwd() == "/home/me"
|