Files
hermes-agent/tests/agent/test_switch_model_rollback.py
ethernet 890bbbda1f Merge remote-tracking branch 'origin/main' into ethie/pm-clean
# Conflicts:
#	apps/desktop/e2e/archived-hidden-session-recoverable.spec.ts
#	apps/desktop/e2e/bot-chat-message-agent-friendly-name.spec.ts
#	apps/desktop/e2e/bot-mailbox-unreadable-ticket.spec.ts
#	apps/desktop/e2e/bot-mode-roster-localized.spec.ts
#	apps/desktop/e2e/bot-mode-row-click-mirrors-registry.spec.ts
#	apps/desktop/e2e/bot-mode-tab-shows-bot-name.spec.ts
#	apps/desktop/e2e/bot-roster-group-row-organisation.spec.ts
#	apps/desktop/e2e/bot-roster-ignores-infra-dirs.spec.ts
#	apps/desktop/e2e/bot-roster-timestamp-meta.spec.ts
#	apps/desktop/e2e/bot-roster-user-sections.spec.ts
#	apps/desktop/e2e/bot-routines-pane-narrow.spec.ts
#	apps/desktop/e2e/bot-row-open-recent-session.spec.ts
#	apps/desktop/e2e/bot-tile-ignores-ambient-composer-model.spec.ts
#	apps/desktop/e2e/group-composer-auto-grow.spec.ts
#	apps/desktop/e2e/group-create-gate-remote-roster.spec.ts
#	apps/desktop/e2e/group-prompt-renamed-primary-handle.spec.ts
#	apps/desktop/e2e/hosted-room-backend-continuity.spec.ts
#	apps/desktop/e2e/hosted-room-legacy-store-migration.spec.ts
#	apps/desktop/e2e/settings-scope-chips-bot-title.spec.ts
#	apps/desktop/e2e/worktree-branch-status.spec.ts
#	apps/desktop/electron/backend-probes.test.ts
#	apps/desktop/electron/connection-apply.test.ts
#	apps/desktop/electron/desktop-electron-pin.test.ts
#	apps/desktop/electron/desktop-uninstall.test.ts
#	apps/desktop/electron/gateway-file-download-transport.test.ts
#	apps/desktop/electron/gateway-stop-before-update.test.ts
#	apps/desktop/electron/github-api-auth.test.ts
#	apps/desktop/electron/registry-primary-profile-scope.test.ts
#	apps/desktop/electron/update-api-check.test.ts
#	apps/desktop/electron/update-handoff-marker.test.ts
#	apps/desktop/electron/venv-blocker-scan.test.ts
#	apps/desktop/scripts/after-extract.test.mjs
#	apps/desktop/scripts/local-pack-publish.test.mjs
#	apps/desktop/scripts/tasks-scroll.test.mjs
#	apps/desktop/src/app/settings/model-settings.test.tsx
#	apps/desktop/src/app/updates-overlay.blockers.test.tsx
#	apps/desktop/src/components/desktop-install-overlay.test.tsx
#	apps/desktop/src/lib/update-copy.test.ts
#	scripts/ci/check_os_marker_fakes.py
#	tests-js/desktop-mac-usage-descriptions.test.ts
#	tests-js/node-engine-alignment.test.ts
#	tests/agent/lsp/test_install_and_lint_fixes.py
#	tests/agent/test_command_token_source.py
#	tests/agent/test_compression_boundary_hook.py
#	tests/agent/test_create_openai_client_ssl_verify.py
#	tests/agent/test_custom_provider_ca_probes.py
#	tests/agent/test_endpoint_blackhole.py
#	tests/agent/test_estimator_parity.py
#	tests/agent/test_in_place_compaction.py
#	tests/agent/test_moa_loop_mode.py
#	tests/agent/test_model_metadata.py
#	tests/agent/test_skill_session_platform_gate.py
#	tests/agent/test_skill_utils.py
#	tests/agent/test_ssl_ca_guard.py
#	tests/computer_use/test_doctor.py
#	tests/cron/test_codex_execution_paths.py
#	tests/cron/test_cron_bot_chat_delivery.py
#	tests/cron/test_cron_script.py
#	tests/cron/test_media_delivery_parity.py
#	tests/cron/test_misfire_catchup.py
#	tests/cron/test_parallel_pool.py
#	tests/cron/test_recurring_eagain_redispatch.py
#	tests/gateway/test_choice_picker.py
#	tests/gateway/test_control_socket_windows_live.py
#	tests/gateway/test_dingtalk.py
#	tests/gateway/test_feishu.py
#	tests/gateway/test_feishu_onboard.py
#	tests/gateway/test_gateway_shutdown.py
#	tests/gateway/test_matrix.py
#	tests/gateway/test_model_command_custom_providers.py
#	tests/gateway/test_reasoning_command.py
#	tests/gateway/test_runtime_footer.py
#	tests/gateway/test_session.py
#	tests/gateway/test_session_hygiene.py
#	tests/gateway/test_status.py
#	tests/gateway/test_teams.py
#	tests/gateway/test_turn_lease.py
#	tests/gateway/test_whatsapp_connect.py
#	tests/hermes_cli/test_approvals_command.py
#	tests/hermes_cli/test_auth_store_lock_concurrent.py
#	tests/hermes_cli/test_backup.py
#	tests/hermes_cli/test_banner_git_state.py
#	tests/hermes_cli/test_certifi_repair.py
#	tests/hermes_cli/test_cmd_update.py
#	tests/hermes_cli/test_compat_manifest_targets.py
#	tests/hermes_cli/test_computer_use_cli.py
#	tests/hermes_cli/test_cpr_local_leak.py
#	tests/hermes_cli/test_dashboard_auth_gate.py
#	tests/hermes_cli/test_dashboard_procs_kill_grace.py
#	tests/hermes_cli/test_desktop_lifecycle_windows_live.py
#	tests/hermes_cli/test_doctor.py
#	tests/hermes_cli/test_doctor_command_install.py
#	tests/hermes_cli/test_fleet_config_migration_windows_live.py
#	tests/hermes_cli/test_gateway.py
#	tests/hermes_cli/test_gateway_platform_gating.py
#	tests/hermes_cli/test_gateway_restart_loop.py
#	tests/hermes_cli/test_gateway_task_probe.py
#	tests/hermes_cli/test_gateway_wsl.py
#	tests/hermes_cli/test_gui_command.py
#	tests/hermes_cli/test_install_cua_driver.py
#	tests/hermes_cli/test_kanban_db.py
#	tests/hermes_cli/test_lazy_command_exports.py
#	tests/hermes_cli/test_lazy_refresh_venv_repair.py
#	tests/hermes_cli/test_linux_desktop_entry.py
#	tests/hermes_cli/test_local_runtime.py
#	tests/hermes_cli/test_local_runtime_updates.py
#	tests/hermes_cli/test_managed_uv.py
#	tests/hermes_cli/test_mcp_reload_confirm_gate.py
#	tests/hermes_cli/test_nous_subscription.py
#	tests/hermes_cli/test_npm_engine.py
#	tests/hermes_cli/test_personality_none.py
#	tests/hermes_cli/test_pet_toggle.py
#	tests/hermes_cli/test_plan_reconciliation_windows_live.py
#	tests/hermes_cli/test_plugin_event_bus.py
#	tests/hermes_cli/test_plugin_manifest_v2.py
#	tests/hermes_cli/test_plugin_packs.py
#	tests/hermes_cli/test_plugins_cmd.py
#	tests/hermes_cli/test_plugins_cmd_enable_disable_nested.py
#	tests/hermes_cli/test_process_identity.py
#	tests/hermes_cli/test_profiles.py
#	tests/hermes_cli/test_profiles_sidebar_cache.py
#	tests/hermes_cli/test_pty_bridge.py
#	tests/hermes_cli/test_resolve_turn_limit.py
#	tests/hermes_cli/test_serve_runtime_inventory.py
#	tests/hermes_cli/test_session_vacuum_config.py
#	tests/hermes_cli/test_set_config_value.py
#	tests/hermes_cli/test_signal_handler_kanban_worker.py
#	tests/hermes_cli/test_slash_confirm_windows.py
#	tests/hermes_cli/test_stale_pid_guard.py
#	tests/hermes_cli/test_startup_fast_guards.py
#	tests/hermes_cli/test_status.py
#	tests/hermes_cli/test_telegram_managed_bot.py
#	tests/hermes_cli/test_tools_config.py
#	tests/hermes_cli/test_update_apply_shallow_count.py
#	tests/hermes_cli/test_update_autostash.py
#	tests/hermes_cli/test_update_concurrent_quarantine.py
#	tests/hermes_cli/test_update_fetch_failure_classifier.py
#	tests/hermes_cli/test_update_fleet_probe_resume_token.py
#	tests/hermes_cli/test_update_handoff_backend_reap.py
#	tests/hermes_cli/test_update_handoff_desktop_rebuild.py
#	tests/hermes_cli/test_update_head_moved_gate.py
#	tests/hermes_cli/test_update_host_obligation.py
#	tests/hermes_cli/test_update_import_guard.py
#	tests/hermes_cli/test_update_interrupted_recovery.py
#	tests/hermes_cli/test_update_inventory.py
#	tests/hermes_cli/test_update_launchd_unloaded_gateway.py
#	tests/hermes_cli/test_update_missing_configured_deps.py
#	tests/hermes_cli/test_update_modified_notice.py
#	tests/hermes_cli/test_update_multiplex_migration_hook.py
#	tests/hermes_cli/test_update_no_gateway_restart.py
#	tests/hermes_cli/test_update_orphan_backend_reap.py
#	tests/hermes_cli/test_update_parked_branch_guard.py
#	tests/hermes_cli/test_update_post_pull_syntax_guard.py
#	tests/hermes_cli/test_update_receipt.py
#	tests/hermes_cli/test_update_self_lock.py
#	tests/hermes_cli/test_update_shim_fail_closed.py
#	tests/hermes_cli/test_update_shim_self_lock.py
#	tests/hermes_cli/test_update_sqlite_remediation.py
#	tests/hermes_cli/test_update_stale_dashboard.py
#	tests/hermes_cli/test_update_stale_virtualenv.py
#	tests/hermes_cli/test_update_venv_health.py
#	tests/hermes_cli/test_update_venv_ownership_preflight.py
#	tests/hermes_cli/test_update_wedged_gateway.py
#	tests/hermes_cli/test_update_yes_flag.py
#	tests/hermes_cli/test_update_zip_two_phase.py
#	tests/hermes_cli/test_urllib_security.py
#	tests/hermes_cli/test_ux_messages_auth_config.py
#	tests/hermes_cli/test_ux_messages_startup.py
#	tests/hermes_cli/test_venv_holder_classifier.py
#	tests/hermes_cli/test_verify_console_scripts.py
#	tests/hermes_cli/test_verify_core_dependencies.py
#	tests/hermes_cli/test_web_server.py
#	tests/hermes_cli/test_web_server_console_ws.py
#	tests/hermes_cli/test_web_server_ws_ping.py
#	tests/hermes_cli/test_web_ui_build.py
#	tests/hermes_state/test_fts_rebuild_admission.py
#	tests/hermes_state/test_hermes_state.py
#	tests/plugins/memory/test_memory_lazy_install.py
#	tests/plugins/test_google_meet_plugin.py
#	tests/plugins/test_langfuse_plugin.py
#	tests/plugins/test_security_guidance_plugin.py
#	tests/plugins/test_transform_llm_output_hook.py
#	tests/scripts/desktop_update/test_desktop_update_windows_gateway_flag.py
#	tests/scripts/desktop_update/test_desktop_update_windows_python_handoff.py
#	tests/scripts/desktop_update/test_desktop_update_windows_timestamp.py
#	tests/scripts/install/test_install_clone_throttle_fallback.py
#	tests/scripts/install/test_install_lockfile_churn.py
#	tests/scripts/install/test_install_no_initial_commit.py
#	tests/scripts/install/test_install_sh_browser_install.py
#	tests/scripts/install/test_install_sh_node_prerelease.py
#	tests/scripts/install/test_install_sh_symlink_stomp.py
#	tests/scripts/install/test_install_sh_uv_lock_config.py
#	tests/scripts/install/test_install_unmerged_index.py
#	tests/scripts/test_contributor_map.py
#	tests/scripts/test_run_tests_parallel.py
#	tests/skills/test_competitor_news_monitor_skill.py
#	tests/skills/test_document_to_action_items_skill.py
#	tests/skills/test_google_workspace_setup.py
#	tests/skills/test_google_workspace_setup_deps.py
#	tests/skills/test_grounded_citations_skill.py
#	tests/skills/test_ip_as_logo_skill.py
#	tests/skills/test_live_dashboard_skill.py
#	tests/skills/test_mcp_oauth_remote_gateway_skill.py
#	tests/skills/test_office_document_skills.py
#	tests/skills/test_openclaw_migration.py
#	tests/skills/test_product_price_monitor_skill.py
#	tests/skills/test_scrollcraft_skill.py
#	tests/skills/test_setup_wizard_generator_skill.py
#	tests/skills/test_weekly_review_planning_skill.py
#	tests/test_engines_satisfiable.py
#	tests/test_fast_safe_load.py
#	tests/test_hermes_bootstrap.py
#	tests/test_hermes_constants.py
#	tests/test_hermes_logging.py
#	tests/test_managed_runtime_resolution.py
#	tests/test_model_tools_async_bridge.py
#	tests/test_packaging_build_guard.py
#	tests/test_packaging_metadata.py
#	tests/test_yaml_indent_consistency.py
#	tests/tools/test_approval_timeout_overflow.py
#	tests/tools/test_base_environment.py
#	tests/tools/test_bot_mode_dm.py
#	tests/tools/test_browser_chromium_check.py
#	tests/tools/test_browser_hardening.py
#	tests/tools/test_browser_homebrew_paths.py
#	tests/tools/test_browser_npx_warmup.py
#	tests/tools/test_browser_orphan_reaper.py
#	tests/tools/test_browser_real_profile.py
#	tests/tools/test_browser_use_cli.py
#	tests/tools/test_clipboard.py
#	tests/tools/test_code_execution.py
#	tests/tools/test_code_execution_modes.py
#	tests/tools/test_code_execution_windows_env.py
#	tests/tools/test_computer_use.py
#	tests/tools/test_delegate_liveness_timeout.py
#	tests/tools/test_execute_code_approval_cluster.py
#	tests/tools/test_execution_flag_detection.py
#	tests/tools/test_fal_common.py
#	tests/tools/test_file_operations.py
#	tests/tools/test_file_tools.py
#	tests/tools/test_file_tools_cwd_resolution.py
#	tests/tools/test_file_tools_live.py
#	tests/tools/test_lazy_deps.py
#	tests/tools/test_lazy_deps_durable_target.py
#	tests/tools/test_lazy_deps_managed.py
#	tests/tools/test_local_env_blocklist.py
#	tests/tools/test_local_tempdir.py
#	tests/tools/test_macos_protected_search.py
#	tests/tools/test_mcp_npx_cached_bin.py
#	tests/tools/test_oneshot_completion_linger.py
#	tests/tools/test_process_registry.py
#	tests/tools/test_read_file_schema_gating.py
#	tests/tools/test_skill_improvements.py
#	tests/tools/test_skills_sync.py
#	tests/tools/test_termux_api_detection.py
#	tests/tools/test_tirith_security.py
#	tests/tools/test_transcription_tools.py
#	tests/tools/test_tts_streaming.py
#	tests/tools/test_wake_word.py
#	tests/tui_gateway/test_compute_host_borrowed_lease.py
#	tests/tui_gateway/test_compute_host_turn_protocol.py
#	tests/tui_gateway/test_isolated_orphan_activity.py
#	tests/tui_gateway/test_protocol.py
#	tests/tui_gateway/test_slash_worker_profile_home.py
#	tests/tui_gateway/test_subprocess_encoding.py
#	tests/tui_gateway/test_tui_gateway_server.py
#	ui-tui/src/__tests__/terminalParity.test.ts
#	ui-tui/src/__tests__/termuxComposerLayout.test.ts
#	ui-tui/src/__tests__/textInputFastEcho.test.ts
2026-09-23 07:02:44 -04:00

181 lines
6.9 KiB
Python

"""Regression test for #33175: switch_model() must roll back to the pre-swap
state if the client rebuild raises.
Before the fix, ``agent.model`` and ``agent.provider`` were assigned BEFORE
the client rebuild was attempted, with no try/except to restore them on
failure. An exception during ``build_anthropic_client`` / OpenAI client
construction left the agent with the new model+provider name but the OLD
client — producing HTTP 400s like "claude-sonnet-4-6 is not supported on
openai-codex" on the next turn.
These tests exercise both branches (openai_chat_completions and
anthropic_messages) and assert that every mutated field returns to its
pre-swap value when the rebuild raises.
"""
from unittest.mock import MagicMock, patch
import pytest
from run_agent import AIAgent
def _make_agent_openrouter():
"""Agent on openrouter (openai-compatible) with sentinel client + kwargs."""
agent = AIAgent.__new__(AIAgent)
agent.provider = "openrouter"
agent.model = "x-ai/grok-4"
agent.base_url = "https://openrouter.ai/api/v1"
agent.api_key = "or-key-original"
agent.api_mode = "chat_completions"
agent.client = MagicMock(name="OriginalOpenRouterClient")
agent._client_kwargs = {
"api_key": "or-key-original",
"base_url": "https://openrouter.ai/api/v1",
}
agent.context_compressor = None
agent._anthropic_api_key = ""
agent._anthropic_base_url = None
agent._anthropic_client = None
agent._is_anthropic_oauth = False
agent._cached_system_prompt = "cached"
agent._primary_runtime = {}
agent._fallback_activated = False
agent._fallback_index = 0
agent._fallback_chain = []
agent._fallback_model = None
agent._config_context_length = None
agent.runtime_capabilities = {"native_compaction": False}
return agent
def _make_agent_anthropic():
"""Agent on native anthropic with a sentinel anthropic client."""
agent = AIAgent.__new__(AIAgent)
agent.provider = "anthropic"
agent.model = "claude-sonnet-4-5"
agent.base_url = "https://api.anthropic.com"
agent.api_key = "sk-ant-original"
agent.api_mode = "anthropic_messages"
agent.client = None
agent._client_kwargs = {}
agent.context_compressor = None
agent._anthropic_api_key = "sk-ant-original"
agent._anthropic_base_url = "https://api.anthropic.com"
agent._anthropic_client = MagicMock(name="OriginalAnthropicClient")
agent._is_anthropic_oauth = False
agent._cached_system_prompt = "cached"
agent._primary_runtime = {}
agent._fallback_activated = False
agent._fallback_index = 0
agent._fallback_chain = []
agent._fallback_model = None
agent._config_context_length = None
agent.runtime_capabilities = {"native_compaction": False}
return agent
def test_openai_client_rebuild_failure_rolls_back_to_original_state():
"""When OpenAI client construction fails, every mutated field must restore."""
agent = _make_agent_openrouter()
original_client = agent.client
original_kwargs = dict(agent._client_kwargs)
# _create_openai_client raises mid-swap (simulates bad key / network error)
def boom(*_a, **_kw):
raise RuntimeError("simulated client build failure")
agent._create_openai_client = boom
with patch("hermes_cli.timeouts.get_provider_request_timeout", return_value=None):
with pytest.raises(RuntimeError, match="simulated client build failure"):
agent.switch_model(
new_model="openai/gpt-5",
new_provider="openai-codex",
api_key="codex-key-new",
base_url="https://chatgpt.com/backend-api/codex/responses",
api_mode="chat_completions",
)
# Core invariant: agent state is unchanged from before the call
assert agent.model == "x-ai/grok-4"
assert agent.provider == "openrouter"
assert agent.base_url == "https://openrouter.ai/api/v1"
assert agent.api_mode == "chat_completions"
assert agent.api_key == "or-key-original"
assert agent.client is original_client
assert agent._client_kwargs == original_kwargs
assert agent.runtime_capabilities == {"native_compaction": False}
def test_anthropic_client_rebuild_failure_rolls_back_to_original_state():
"""When build_anthropic_client raises, every mutated field must restore."""
agent = _make_agent_anthropic()
original_anthropic_client = agent._anthropic_client
original_anthropic_key = agent._anthropic_api_key
original_anthropic_base = agent._anthropic_base_url
with (
patch(
"agent.anthropic_adapter.build_anthropic_client",
side_effect=RuntimeError("simulated anthropic build failure"),
),
patch(
"agent.anthropic_credentials.resolve_anthropic_token",
return_value="sk-ant-resolved",
),
patch("agent.anthropic_credentials._is_oauth_token", return_value=False),
patch("hermes_cli.timeouts.get_provider_request_timeout", return_value=None),
):
with pytest.raises(RuntimeError, match="simulated anthropic build failure"):
agent.switch_model(
new_model="claude-opus-4-6",
new_provider="opencode-zen",
api_key="zen-key-new",
base_url="https://opencode.example/v1",
api_mode="anthropic_messages",
)
# Anthropic-specific state restored
assert agent._anthropic_client is original_anthropic_client
assert agent._anthropic_api_key == original_anthropic_key
assert agent._anthropic_base_url == original_anthropic_base
# Core state also restored
assert agent.model == "claude-sonnet-4-5"
assert agent.provider == "anthropic"
assert agent.base_url == "https://api.anthropic.com"
assert agent.api_mode == "anthropic_messages"
assert agent.api_key == "sk-ant-original"
def test_cross_branch_anthropic_to_openai_rebuild_failure_rolls_back():
"""Switching from anthropic_messages to chat_completions: failure must
restore the anthropic state, not leave the agent half-converted."""
agent = _make_agent_anthropic()
original_anthropic_client = agent._anthropic_client
def boom(*_a, **_kw):
raise RuntimeError("openai client failed")
agent._create_openai_client = boom
with patch("hermes_cli.timeouts.get_provider_request_timeout", return_value=None):
with pytest.raises(RuntimeError, match="openai client failed"):
agent.switch_model(
new_model="x-ai/grok-4",
new_provider="openrouter",
api_key="or-key-new",
base_url="https://openrouter.ai/api/v1",
api_mode="chat_completions",
)
# Anthropic client preserved (not nulled by the openai branch)
assert agent._anthropic_client is original_anthropic_client
assert agent.model == "claude-sonnet-4-5"
assert agent.provider == "anthropic"
assert agent.api_mode == "anthropic_messages"
assert agent.base_url == "https://api.anthropic.com"