Skip to content

Use Claude Code, Codex, VSCode, Pi, and OpenCode (and 6 other harnesses) for free (1.3B+ free tokens) from your terminal, app, IDE, or phone, and now from the browser with native browser sessions (multi-harness + multi-model) like OpenClaw (voice supported + ToS friendly)

Python
40.9 KB
Raw Blame
import asyncio
import logging
from collections.abc import AsyncIterator, Mapping
from dataclasses import replace
from pathlib import Path
from unittest.mock import AsyncMock, MagicMock, patch

import pytest

from free_claude_code.application.connected_accounts import (
    ConnectedAccountState,
    ConnectedAccountStatus,
)
from free_claude_code.application.errors import ApplicationUnavailableError
from free_claude_code.application.model_metadata import ProviderModelInfo
from free_claude_code.config.admin.credentials import CredentialSubmission
from free_claude_code.config.admin.persistence import PreparedAdminUpdate
from free_claude_code.config.api_keys import normalize_api_keys
from free_claude_code.config.loader import ManagedConfigSnapshot
from free_claude_code.config.settings import Settings
from free_claude_code.core.anthropic.models import MessagesRequest
from free_claude_code.core.openai_responses import OpenAIResponsesRequest
from free_claude_code.core.reasoning import DEFAULT_REASONING_POLICY, ReasoningPolicy
from free_claude_code.core.stream_recovery import ContinuationSeed
from free_claude_code.messaging.command_context import StopOutcome
from free_claude_code.messaging.platforms.ports import (
    InboundMessageHandler,
    MessagingPlatformComponents,
    MessagingStartupNotice,
)
from free_claude_code.providers.base import BaseProvider
from free_claude_code.providers.credential_validation import (
    CredentialCheck,
    CredentialStatus,
)
from free_claude_code.providers.runtime import ProviderRuntime
from free_claude_code.runtime.application import ApplicationRuntime
from free_claude_code.runtime.configuration import ConfigurationService
from free_claude_code.runtime.provider_manager import ProviderRuntimeManager
from tests.providers.support import make_provider_config


class TrackingRuntime(ProviderRuntime):
    def __init__(self, settings: Settings, admission_registry) -> None:
        super().__init__(settings, admission_registry)
        self.cleanup_calls = 0

    async def cleanup(self) -> None:
        self.cleanup_calls += 1
        await super().cleanup()


class AdminModelProvider(BaseProvider):
    def __init__(
        self,
        model_infos: frozenset[ProviderModelInfo] = frozenset(),
        *,
        error: Exception | None = None,
    ) -> None:
        super().__init__(
            make_provider_config(
                api_key="test",
                base_url="https://provider.invalid/v1",
            )
        )
        self._model_infos = model_infos
        self._error = error

    async def cleanup(self) -> None:
        return None

    async def list_model_infos(self) -> frozenset[ProviderModelInfo]:
        if self._error is not None:
            raise self._error
        return self._model_infos

    async def stream_messages(
        self,
        request: MessagesRequest,
        input_tokens: int = 0,
        *,
        request_id: str | None = None,
        response_model: str | None = None,
        reasoning: ReasoningPolicy = DEFAULT_REASONING_POLICY,
        request_headers: Mapping[str, str] | None = None,
        model_info: ProviderModelInfo | None = None,
        continuation: ContinuationSeed | None = None,
    ) -> AsyncIterator[str]:
        if False:
            yield ""

    async def stream_responses(
        self,
        request: OpenAIResponsesRequest,
        input_tokens: int = 0,
        *,
        request_id: str | None = None,
        response_model: str | None = None,
        reasoning: ReasoningPolicy = DEFAULT_REASONING_POLICY,
        request_headers: Mapping[str, str] | None = None,
        model_info: ProviderModelInfo | None = None,
        continuation: ContinuationSeed | None = None,
    ) -> AsyncIterator[str]:
        if False:
            yield ""


class TrackingFactory:
    def __init__(self) -> None:
        self.runtimes: list[TrackingRuntime] = []
        self.fail = False
        self.events: list[str] = []

    def __call__(self, settings: Settings, admission_registry) -> ProviderRuntime:
        self.events.append(f"construct:{settings.model}")
        if self.fail:
            raise RuntimeError("candidate failed")
        runtime = TrackingRuntime(settings, admission_registry)
        self.runtimes.append(runtime)
        return runtime


class TrackingTranscriber:
    def __init__(self, events: list[str]) -> None:
        self.events = events
        self.close_calls = 0

    async def transcribe(self, file_path: Path) -> str:
        assert isinstance(file_path, Path)
        return "transcribed"

    async def close(self) -> None:
        self.close_calls += 1
        self.events.append("transcriber.close")


class FailingTranscriber(TrackingTranscriber):
    async def close(self) -> None:
        await super().close()
        raise RuntimeError("transcriber close failed")


class CancelledTranscriber(TrackingTranscriber):
    async def close(self) -> None:
        await super().close()
        raise asyncio.CancelledError


class CancellingOnceTranscriber(TrackingTranscriber):
    async def close(self) -> None:
        await super().close()
        if self.close_calls == 1:
            raise asyncio.CancelledError


class TrackingMessagingRuntime:
    name = "tracking"

    def __init__(
        self,
        events: list[str],
        *,
        fail_quiesce_once: bool = False,
        fail_close_once: bool = False,
    ) -> None:
        self.events = events
        self.fail_quiesce_once = fail_quiesce_once
        self.fail_close_once = fail_close_once

    async def start(self) -> None:
        self.events.append("messaging.start")

    async def quiesce(self) -> None:
        self.events.append("messaging.quiesce")
        if self.fail_quiesce_once:
            self.fail_quiesce_once = False
            raise RuntimeError("quiesce failed")

    async def close(self) -> None:
        self.events.append("messaging.close")
        if self.fail_close_once:
            self.fail_close_once = False
            raise RuntimeError("close failed")

    def on_message(self, handler: InboundMessageHandler) -> None:
        assert callable(handler)

    @property
    def is_connected(self) -> bool:
        return True


class PersistentlyFailingMessagingRuntime(TrackingMessagingRuntime):
    def __init__(self, events: list[str]) -> None:
        super().__init__(events)
        self.fail_quiesce = True

    async def quiesce(self) -> None:
        self.events.append("messaging.quiesce")
        if self.fail_quiesce:
            raise RuntimeError("quiesce failed")


def _settings(model: str, *, port: int = 8082) -> Settings:
    return Settings().model_copy(update={"model": model, "port": port})


def _prepared(
    settings: Settings,
    tmp_path,
    *,
    pending_fields: tuple[str, ...] = (),
) -> PreparedAdminUpdate:
    return PreparedAdminUpdate(
        target_values={"MODEL": settings.model},
        settings=settings,
        errors=(),
        pending_fields=pending_fields,
        path=tmp_path / ".env",
        expected=ManagedConfigSnapshot(settings, {}, {}, {}, tmp_path / ".env"),
    )


def _applied_response(pending_fields: tuple[str, ...] = ()) -> dict[str, object]:
    return {
        "applied": True,
        "valid": True,
        "errors": [],
        "env_preview": "MODEL=updated\n",
        "path": ".env",
        "pending_fields": list(pending_fields),
    }


def _runtime_with_admin_provider(
    provider: BaseProvider,
) -> tuple[ApplicationRuntime, ProviderRuntimeManager]:
    settings = _settings("nvidia_nim/fallback").model_copy(
        update={"nvidia_nim_api_keys": normalize_api_keys("test-key")}
    )
    manager = ProviderRuntimeManager(
        settings,
        runtime_factory=lambda snapshot, admission_registry: ProviderRuntime(
            snapshot,
            admission_registry,
            {("nvidia_nim", 1): provider},
        ),
    )
    return ApplicationRuntime(
        manager, configuration=AsyncMock(spec=ConfigurationService), transcriber=None
    ), manager


@pytest.mark.asyncio
async def test_provider_check_caches_and_returns_sorted_models() -> None:
    provider = AdminModelProvider(
        frozenset(
            {
                ProviderModelInfo("vendor/model-b"),
                ProviderModelInfo("vendor/model-a"),
            }
        )
    )
    runtime, manager = _runtime_with_admin_provider(provider)

    result = await runtime.test_provider("nvidia_nim")

    assert result == {
        "provider_id": "nvidia_nim",
        "ok": True,
        "models": ["vendor/model-a", "vendor/model-b"],
    }
    assert manager.cached_model_ids()["nvidia_nim"] == frozenset(
        {"vendor/model-a", "vendor/model-b"}
    )
    await manager.close()


@pytest.mark.asyncio
async def test_provider_check_returns_stable_failure_for_application_error() -> None:
    provider = AdminModelProvider(
        error=ApplicationUnavailableError(
            "NVIDIA_NIM_API_KEY is not set. Add it in the Admin UI."
        )
    )
    runtime, manager = _runtime_with_admin_provider(provider)

    result = await runtime.test_provider("nvidia_nim")

    assert result == {
        "provider_id": "nvidia_nim",
        "ok": False,
        "message": "The provider is unavailable. Check its configuration and access.",
    }
    assert "error_type" not in result
    await manager.close()


@pytest.mark.asyncio
async def test_provider_check_never_returns_unrecognized_credentials(
    caplog: pytest.LogCaptureFixture,
) -> None:
    secret = "CREDENTIAL[unrecognized-format-987654321]"
    provider = AdminModelProvider(
        error=RuntimeError(f"Provider rejected credential {secret}")
    )
    runtime, manager = _runtime_with_admin_provider(provider)

    with caplog.at_level(logging.WARNING):
        result = await runtime.test_provider("nvidia_nim")

    assert result == {
        "provider_id": "nvidia_nim",
        "ok": False,
        "message": "Could not load models. Check the provider's settings and retry.",
    }
    log_text = " | ".join(record.getMessage() for record in caplog.records)
    assert "provider=nvidia_nim" in log_text
    assert isinstance(result["message"], str)
    assert result["message"] in log_text
    assert secret not in log_text
    assert "error_type" not in result
    await manager.close()


@pytest.mark.asyncio
async def test_stop_all_maps_messaging_outcome_to_application_count() -> None:
    manager = ProviderRuntimeManager(_settings("nvidia_nim/model"))
    runtime = ApplicationRuntime(
        manager, configuration=AsyncMock(spec=ConfigurationService), transcriber=None
    )
    workflow = MagicMock()
    workflow.stop_all_tasks = AsyncMock(
        return_value=StopOutcome(
            cancelled_count=3,
            status_feedback_scopes=frozenset(),
            fallback_required=False,
        )
    )
    runtime._messaging._messaging_workflow = workflow

    result = await runtime.stop_all()

    assert result is not None
    assert result.cancelled_count == 3
    workflow.stop_all_tasks.assert_awaited_once()
    await manager.close()


@pytest.mark.asyncio
async def test_provider_apply_constructs_before_commit_then_publishes(tmp_path) -> None:
    factory = TrackingFactory()
    manager = ProviderRuntimeManager(
        _settings("nvidia_nim/old"),
        runtime_factory=factory,
    )
    runtime = ApplicationRuntime(
        manager, configuration=AsyncMock(spec=ConfigurationService), transcriber=None
    )
    prepared = _prepared(_settings("nvidia_nim/new"), tmp_path)
    factory.events.clear()

    def commit(_prepared_update: PreparedAdminUpdate) -> dict[str, object]:
        factory.events.append("commit")
        assert manager.current_generation_id == 1
        return _applied_response()

    with (
        patch.object(
            runtime._configuration,
            "prepare",
            return_value=prepared,
        ),
        patch.object(
            runtime._configuration,
            "commit",
            side_effect=commit,
        ),
    ):
        result = await runtime.apply_admin_config({"MODEL": "nvidia_nim/new"})

    assert factory.events == ["construct:nvidia_nim/new", "commit"]
    assert manager.current_generation_id == 2
    assert manager.current_settings().model == "nvidia_nim/new"
    assert result["restart"] == {
        "required": False,
        "automatic": False,
        "admin_url": None,
        "fields": [],
    }
    await manager.close()


@pytest.mark.asyncio
async def test_candidate_failure_never_commits_and_preserves_current(tmp_path) -> None:
    factory = TrackingFactory()
    manager = ProviderRuntimeManager(
        _settings("nvidia_nim/old"),
        runtime_factory=factory,
    )
    runtime = ApplicationRuntime(
        manager, configuration=AsyncMock(spec=ConfigurationService), transcriber=None
    )
    prepared = _prepared(_settings("nvidia_nim/new"), tmp_path)
    factory.fail = True

    with (
        patch.object(
            runtime._configuration,
            "prepare",
            return_value=prepared,
        ),
        patch.object(runtime._configuration, "commit") as commit,
        pytest.raises(RuntimeError, match="candidate failed"),
    ):
        await runtime.apply_admin_config({"MODEL": "nvidia_nim/new"})

    commit.assert_not_called()
    assert manager.current_generation_id == 1
    assert manager.current_settings().model == "nvidia_nim/old"
    await manager.close()


@pytest.mark.asyncio
async def test_persistence_failure_closes_candidate_and_preserves_current(
    tmp_path,
) -> None:
    factory = TrackingFactory()
    manager = ProviderRuntimeManager(
        _settings("nvidia_nim/old"),
        runtime_factory=factory,
    )
    runtime = ApplicationRuntime(
        manager, configuration=AsyncMock(spec=ConfigurationService), transcriber=None
    )
    prepared = _prepared(_settings("nvidia_nim/new"), tmp_path)

    with (
        patch.object(
            runtime._configuration,
            "prepare",
            return_value=prepared,
        ),
        patch.object(
            runtime._configuration,
            "commit",
            side_effect=OSError("disk full"),
        ),
        pytest.raises(OSError, match="disk full"),
    ):
        await runtime.apply_admin_config({"MODEL": "nvidia_nim/new"})

    assert manager.current_generation_id == 1
    assert factory.runtimes[0].cleanup_calls == 0
    assert factory.runtimes[1].cleanup_calls == 1
    await manager.close()


@pytest.mark.asyncio
async def test_restart_required_apply_commits_without_hot_publication(tmp_path) -> None:
    factory = TrackingFactory()
    manager = ProviderRuntimeManager(
        _settings("nvidia_nim/old"),
        runtime_factory=factory,
    )
    restart = MagicMock(return_value=None)
    runtime = ApplicationRuntime(
        manager,
        configuration=AsyncMock(spec=ConfigurationService),
        transcriber=None,
        restart_callback=restart,
    )
    prepared = _prepared(
        _settings("nvidia_nim/old", port=9090),
        tmp_path,
        pending_fields=("PORT",),
    )
    with patch.object(
        runtime._configuration, "admin_values", AsyncMock(return_value={})
    ):
        instance_id = (await runtime.admin_status())["instance_id"]

    with (
        patch.object(
            runtime._configuration,
            "prepare",
            return_value=prepared,
        ),
        patch.object(
            runtime._configuration,
            "commit",
            return_value=_applied_response(("PORT",)),
        ) as commit,
    ):
        result = await runtime.apply_admin_config({"PORT": "9090"})

    commit.assert_called_once_with(prepared)
    assert manager.current_generation_id == 1
    assert len(factory.runtimes) == 1
    assert result["restart"] == {
        "required": True,
        "automatic": True,
        "admin_url": "http://127.0.0.1:9090/admin",
        "fields": ["PORT"],
        "instance_id": instance_id,
    }
    restart.assert_called_once_with()
    await manager.close()


@pytest.mark.asyncio
@pytest.mark.parametrize("pending", [(), ("PORT",)])
@pytest.mark.parametrize("status", list(CredentialStatus))
async def test_credential_checks_gate_both_apply_paths(tmp_path, pending, status):
    factory = TrackingFactory()
    manager = ProviderRuntimeManager(
        _settings("nvidia_nim/old"), runtime_factory=factory
    )
    runtime = ApplicationRuntime(
        manager, configuration=AsyncMock(spec=ConfigurationService), transcriber=None
    )
    prepared = replace(
        _prepared(_settings("nvidia_nim/new"), tmp_path, pending_fields=pending),
        changed_keys=("GROQ_API_KEY",),
        credential_submissions=(
            CredentialSubmission("GROQ_API_KEY", normalize_api_keys("new")),
        ),
    )
    checks = (
        CredentialCheck("GROQ_API_KEY", status, "Safe result"),
        CredentialCheck("OPENROUTER_API_KEY", CredentialStatus.VERIFIED, "Accepted"),
    )
    with (
        patch.object(
            runtime._configuration,
            "prepare",
            return_value=prepared,
        ),
        patch(
            "free_claude_code.runtime.application.check_credentials",
            AsyncMock(return_value=checks),
        ) as check,
        patch.object(
            runtime._configuration,
            "commit",
            return_value=_applied_response(pending),
        ) as commit,
    ):
        result = await runtime.apply_admin_config({"GROQ_API_KEY": "new"})
    check.assert_awaited_once_with(prepared.settings, prepared.credential_submissions)
    assert result["credential_checks"] == [
        {
            "key": check.key,
            "index": check.index,
            "status": check.status.value,
            "message": check.message,
        }
        for check in checks
    ]
    if status == CredentialStatus.REJECTED:
        commit.assert_not_called()
        assert result["applied"] is False
        assert manager.current_generation_id == 1
    else:
        commit.assert_called_once_with(prepared)
        assert result["applied"] is True
        assert manager.current_generation_id == (1 if pending else 2)
    await manager.close()


@pytest.mark.asyncio
async def test_cancelled_credential_check_never_commits(tmp_path):
    manager = ProviderRuntimeManager(_settings("nvidia_nim/old"))
    runtime = ApplicationRuntime(
        manager, configuration=AsyncMock(spec=ConfigurationService), transcriber=None
    )
    prepared = replace(
        _prepared(_settings("nvidia_nim/new"), tmp_path), changed_keys=("GROQ_API_KEY",)
    )
    with (
        patch.object(
            runtime._configuration,
            "prepare",
            return_value=prepared,
        ),
        patch(
            "free_claude_code.runtime.application.check_credentials",
            AsyncMock(side_effect=asyncio.CancelledError),
        ),
        patch.object(runtime._configuration, "commit") as commit,
        pytest.raises(asyncio.CancelledError),
    ):
        await runtime.apply_admin_config({"GROQ_API_KEY": "new"})
    commit.assert_not_called()
    assert manager.current_generation_id == 1
    await manager.close()


@pytest.mark.asyncio
async def test_close_drains_messaging_before_transcriber_and_is_idempotent() -> None:
    events: list[str] = []
    manager = ProviderRuntimeManager(_settings("nvidia_nim/model"))
    transcriber = TrackingTranscriber(events)
    runtime = ApplicationRuntime(
        manager,
        configuration=AsyncMock(spec=ConfigurationService),
        transcriber=transcriber,
    )
    runtime._messaging._messaging_runtime = TrackingMessagingRuntime(events)
    workflow = MagicMock()
    workflow.close = AsyncMock(side_effect=lambda: events.append("workflow.close"))
    runtime._messaging._messaging_workflow = workflow
    runtime._messaging._cli_manager = MagicMock()

    assert runtime.is_closed is False
    assert await runtime.close() is True
    assert await runtime.close() is True

    assert events == [
        "messaging.quiesce",
        "workflow.close",
        "messaging.close",
        "transcriber.close",
    ]
    assert transcriber.close_calls == 1
    assert runtime._messaging._transcriber is None
    assert runtime._messaging._messaging_runtime is None
    assert runtime._messaging._messaging_workflow is None
    assert runtime.is_closed is True


@pytest.mark.asyncio
async def test_close_retains_transcriber_ownership_when_close_fails() -> None:
    events: list[str] = []
    manager = ProviderRuntimeManager(_settings("nvidia_nim/model"))
    transcriber = FailingTranscriber(events)
    runtime = ApplicationRuntime(
        manager,
        configuration=AsyncMock(spec=ConfigurationService),
        transcriber=transcriber,
    )
    runtime._messaging._messaging_runtime = TrackingMessagingRuntime(events)

    assert await runtime.close() is False

    assert events == [
        "messaging.quiesce",
        "messaging.close",
        "transcriber.close",
    ]
    assert transcriber.close_calls == 1
    assert runtime._messaging._transcriber is transcriber
    assert runtime._closed is False
    await manager.close()


@pytest.mark.asyncio
async def test_close_retries_runtime_before_closing_later_resources() -> None:
    events: list[str] = []
    manager = ProviderRuntimeManager(_settings("nvidia_nim/model"))
    transcriber = TrackingTranscriber(events)
    runtime = ApplicationRuntime(
        manager,
        configuration=AsyncMock(spec=ConfigurationService),
        transcriber=transcriber,
    )
    messaging = TrackingMessagingRuntime(events, fail_close_once=True)
    runtime._messaging._messaging_runtime = messaging

    assert await runtime.close() is False

    assert events == ["messaging.quiesce", "messaging.close"]
    assert runtime._messaging._messaging_runtime is messaging
    assert runtime._messaging._transcriber is transcriber
    assert runtime._closed is False

    assert await runtime.close() is True

    assert events == [
        "messaging.quiesce",
        "messaging.close",
        "messaging.quiesce",
        "messaging.close",
        "transcriber.close",
    ]
    assert runtime._messaging._messaging_runtime is None
    assert runtime._messaging._transcriber is None
    assert runtime._closed is True


@pytest.mark.asyncio
async def test_close_retries_connected_account_without_reclosing_providers() -> None:
    factory = TrackingFactory()
    manager = ProviderRuntimeManager(
        _settings("nvidia_nim/model"),
        runtime_factory=factory,
    )
    account = MagicMock()
    account.status.return_value = MagicMock(revision=0)
    account.close = AsyncMock(side_effect=[RuntimeError("auth cleanup failed"), None])
    runtime = ApplicationRuntime(
        manager,
        configuration=AsyncMock(spec=ConfigurationService),
        transcriber=None,
        connected_accounts={"openai": account},
    )

    assert await runtime.close() is False
    assert manager._closed is True
    assert factory.runtimes[0].cleanup_calls == 1

    assert await runtime.close() is True
    assert account.close.await_count == 2
    assert factory.runtimes[0].cleanup_calls == 1
    assert runtime.is_closed is True


@pytest.mark.asyncio
async def test_connected_account_status_reports_cached_model_count() -> None:
    account = MagicMock()
    account.is_connected.return_value = True
    account.status.return_value = ConnectedAccountStatus(
        provider_id="openai",
        state=ConnectedAccountState.CONNECTED,
        connected=True,
        revision=1,
        email="[email protected]",
    )
    account.close = AsyncMock()
    manager = ProviderRuntimeManager(
        _settings("nvidia_nim/model"),
        connected_provider_ids=lambda: ("openai",),
    )
    manager.cache_model_infos(
        "openai",
        {
            ProviderModelInfo(model_id="gpt-one"),
            ProviderModelInfo(model_id="gpt-two"),
        },
    )
    runtime = ApplicationRuntime(
        manager,
        configuration=AsyncMock(spec=ConfigurationService),
        transcriber=None,
        connected_accounts={"openai": account},
    )

    status = await runtime.connected_account_status("openai")

    assert status.email == "[email protected]"
    assert status.model_count == 2
    assert status.as_dict()["model_count"] == 2
    assert await runtime.close() is True


@pytest.mark.asyncio
async def test_close_retries_workflow_close_before_closing_delivery() -> None:
    events: list[str] = []
    manager = ProviderRuntimeManager(_settings("nvidia_nim/model"))
    runtime = ApplicationRuntime(
        manager, configuration=AsyncMock(spec=ConfigurationService), transcriber=None
    )
    messaging = TrackingMessagingRuntime(events)
    workflow = MagicMock()
    close_calls = 0

    async def close_workflow() -> None:
        nonlocal close_calls
        close_calls += 1
        if close_calls == 1:
            raise RuntimeError("drain failed")
        events.append("workflow.close")

    workflow.close = AsyncMock(side_effect=close_workflow)
    runtime._messaging._messaging_runtime = messaging
    runtime._messaging._messaging_workflow = workflow

    assert await runtime.close() is False

    assert events == ["messaging.quiesce"]
    assert runtime._messaging._messaging_runtime is messaging
    assert runtime._messaging._messaging_workflow is workflow
    assert runtime._closed is False

    assert await runtime.close() is True

    assert events == [
        "messaging.quiesce",
        "messaging.quiesce",
        "workflow.close",
        "messaging.close",
    ]
    assert runtime._closed is True


@pytest.mark.asyncio
async def test_close_does_not_drain_workflow_until_ingress_is_quiescent() -> None:
    events: list[str] = []
    manager = ProviderRuntimeManager(_settings("nvidia_nim/model"))
    runtime = ApplicationRuntime(
        manager, configuration=AsyncMock(spec=ConfigurationService), transcriber=None
    )
    messaging = TrackingMessagingRuntime(events, fail_quiesce_once=True)
    workflow = MagicMock()
    workflow.close = AsyncMock(side_effect=lambda: events.append("workflow.close"))
    runtime._messaging._messaging_runtime = messaging
    runtime._messaging._messaging_workflow = workflow

    assert await runtime.close() is False

    workflow.close.assert_not_awaited()
    assert runtime._messaging._messaging_runtime is messaging
    assert runtime._messaging._messaging_workflow is workflow

    assert await runtime.close() is True

    assert events == [
        "messaging.quiesce",
        "messaging.quiesce",
        "workflow.close",
        "messaging.close",
    ]
    assert runtime._closed is True


@pytest.mark.asyncio
async def test_close_retries_failed_persistence_before_closing_delivery() -> None:
    events: list[str] = []
    manager = ProviderRuntimeManager(_settings("nvidia_nim/model"))
    runtime = ApplicationRuntime(
        manager, configuration=AsyncMock(spec=ConfigurationService), transcriber=None
    )
    messaging = TrackingMessagingRuntime(events)
    workflow = MagicMock()
    close_calls = 0

    async def close_workflow() -> None:
        nonlocal close_calls
        close_calls += 1
        if close_calls == 1:
            raise RuntimeError("flush failed")
        events.append("workflow.close")

    workflow.close = AsyncMock(side_effect=close_workflow)
    runtime._messaging._messaging_runtime = messaging
    runtime._messaging._messaging_workflow = workflow

    await runtime.close()

    assert runtime._messaging._messaging_workflow is workflow
    assert runtime._messaging._messaging_runtime is messaging
    assert "messaging.close" not in events

    await runtime.close()

    assert events == [
        "messaging.quiesce",
        "messaging.quiesce",
        "workflow.close",
        "messaging.close",
    ]
    assert runtime._closed is True


@pytest.mark.asyncio
async def test_cancelled_transcriber_close_retains_ownership() -> None:
    events: list[str] = []
    manager = ProviderRuntimeManager(_settings("nvidia_nim/model"))
    transcriber = CancelledTranscriber(events)
    runtime = ApplicationRuntime(
        manager,
        configuration=AsyncMock(spec=ConfigurationService),
        transcriber=transcriber,
    )

    with pytest.raises(asyncio.CancelledError):
        await runtime._messaging.close_transcriber()

    assert transcriber.close_calls == 1
    assert runtime._messaging._transcriber is transcriber
    await manager.close()


@pytest.mark.asyncio
async def test_cancelled_application_close_remains_retryable() -> None:
    events: list[str] = []
    manager = ProviderRuntimeManager(_settings("nvidia_nim/model"))
    transcriber = CancellingOnceTranscriber(events)
    runtime = ApplicationRuntime(
        manager,
        configuration=AsyncMock(spec=ConfigurationService),
        transcriber=transcriber,
    )

    with pytest.raises(asyncio.CancelledError):
        await runtime.close()

    assert runtime._closed is False
    assert runtime._messaging._transcriber is transcriber

    await runtime.close()

    assert transcriber.close_calls == 2
    assert runtime._messaging._transcriber is None
    assert runtime._closed is True


@pytest.mark.asyncio
async def test_startup_failure_closes_owned_transcriber() -> None:
    events: list[str] = []
    manager = ProviderRuntimeManager(_settings("nvidia_nim/model"))
    transcriber = TrackingTranscriber(events)
    runtime = ApplicationRuntime(
        manager,
        configuration=AsyncMock(spec=ConfigurationService),
        transcriber=transcriber,
    )

    with (
        patch.object(
            runtime._configuration,
            "initialize",
            AsyncMock(side_effect=RuntimeError("startup failed")),
        ),
        pytest.raises(RuntimeError, match="startup failed"),
    ):
        await runtime.start()

    assert transcriber.close_calls == 1


@pytest.mark.asyncio
async def test_startup_cancellation_cleans_partial_messaging_and_reraises() -> None:
    events: list[str] = []
    manager = ProviderRuntimeManager(_settings("nvidia_nim/model"))
    transcriber = TrackingTranscriber(events)
    runtime = ApplicationRuntime(
        manager,
        configuration=AsyncMock(spec=ConfigurationService),
        transcriber=transcriber,
    )
    messaging = TrackingMessagingRuntime(events)
    entered = asyncio.Event()

    async def start_messaging() -> None:
        runtime._messaging._messaging_runtime = messaging
        entered.set()
        await asyncio.Event().wait()

    with patch.object(
        runtime._messaging,
        "_start_messaging_if_configured",
        side_effect=start_messaging,
    ):
        await runtime.start()
        await entered.wait()
        assert await runtime.close()

    assert events == [
        "messaging.quiesce",
        "messaging.close",
        "transcriber.close",
    ]
    assert runtime._closed is True
    assert runtime._messaging._messaging_runtime is None
    assert runtime._messaging._transcriber is None


@pytest.mark.asyncio
async def test_public_start_retries_transient_partial_messaging_cleanup() -> None:
    events: list[str] = []
    manager = ProviderRuntimeManager(
        _settings("nvidia_nim/model").model_copy(
            update={"messaging_platform": "telegram"}
        )
    )
    runtime = ApplicationRuntime(
        manager, configuration=AsyncMock(spec=ConfigurationService), transcriber=None
    )
    messaging = TrackingMessagingRuntime(events, fail_quiesce_once=True)
    workflow = MagicMock()
    workflow.close = AsyncMock(side_effect=lambda: events.append("workflow.close"))
    cli_manager = MagicMock()
    components = MessagingPlatformComponents(
        name="tracking",
        runtime=messaging,
        outbound=MagicMock(),
    )
    startup_failure = RuntimeError("partial messaging startup failed")

    async def fail_after_publication(
        published: MessagingPlatformComponents,
    ) -> None:
        assert published is components
        runtime._messaging._messaging_runtime = messaging
        runtime._messaging._messaging_workflow = workflow
        runtime._messaging._cli_manager = cli_manager
        raise startup_failure

    with (
        patch.object(manager, "start_model_list_refresh"),
        patch(
            "free_claude_code.runtime.messaging_service.messaging_platform_factory.create_messaging_components",
            return_value=components,
        ),
        patch.object(
            runtime._messaging,
            "_start_messaging_workflow",
            side_effect=fail_after_publication,
        ),
    ):
        await runtime.start()
        with pytest.raises(RuntimeError, match="cleanup incomplete") as raised:
            await asyncio.gather(*runtime._messaging._startup_tasks)
        assert runtime._messaging._messaging_state == "failed"
        assert await runtime.close() is True

    assert raised.value.__cause__ is startup_failure
    assert events == [
        "messaging.quiesce",
        "messaging.quiesce",
        "workflow.close",
        "messaging.close",
    ]
    workflow.close.assert_awaited_once()
    assert runtime._messaging._messaging_runtime is None
    assert runtime._messaging._messaging_workflow is None
    assert runtime._messaging._cli_manager is None
    assert runtime.is_closed is True


@pytest.mark.asyncio
async def test_public_start_retains_persistently_unclean_partial_messaging_graph() -> (
    None
):
    events: list[str] = []
    manager = ProviderRuntimeManager(
        _settings("nvidia_nim/model").model_copy(
            update={"messaging_platform": "telegram"}
        )
    )
    transcriber = TrackingTranscriber(events)
    runtime = ApplicationRuntime(
        manager,
        configuration=AsyncMock(spec=ConfigurationService),
        transcriber=transcriber,
    )
    messaging = PersistentlyFailingMessagingRuntime(events)
    workflow = MagicMock()
    workflow.close = AsyncMock(side_effect=lambda: events.append("workflow.close"))
    cli_manager = MagicMock()
    components = MessagingPlatformComponents(
        name="tracking",
        runtime=messaging,
        outbound=MagicMock(),
    )
    startup_failure = RuntimeError("partial messaging startup failed")

    async def fail_after_publication(
        published: MessagingPlatformComponents,
    ) -> None:
        assert published is components
        runtime._messaging._messaging_runtime = messaging
        runtime._messaging._messaging_workflow = workflow
        runtime._messaging._cli_manager = cli_manager
        raise startup_failure

    with (
        patch.object(manager, "start_model_list_refresh"),
        patch(
            "free_claude_code.runtime.messaging_service.messaging_platform_factory.create_messaging_components",
            return_value=components,
        ),
        patch.object(
            runtime._messaging,
            "_start_messaging_workflow",
            side_effect=fail_after_publication,
        ),
    ):
        await runtime.start()
        with pytest.raises(RuntimeError, match="cleanup incomplete") as raised:
            await asyncio.gather(*runtime._messaging._startup_tasks)
        assert runtime._messaging._messaging_state == "failed"
        assert await runtime.close() is False

    assert raised.value.__cause__ is startup_failure
    assert events == ["messaging.quiesce", "messaging.quiesce"]
    workflow.close.assert_not_awaited()
    assert transcriber.close_calls == 0
    assert runtime._messaging._messaging_runtime is messaging
    assert runtime._messaging._messaging_workflow is workflow
    assert runtime._messaging._cli_manager is cli_manager
    assert runtime._messaging._transcriber is transcriber
    assert runtime._provider_manager_closed is False
    assert runtime.is_closed is False

    messaging.fail_quiesce = False
    assert await runtime.close() is True

    assert events == [
        "messaging.quiesce",
        "messaging.quiesce",
        "messaging.quiesce",
        "workflow.close",
        "messaging.close",
        "transcriber.close",
    ]
    assert runtime.is_closed is True


@pytest.mark.asyncio
async def test_messaging_start_failure_is_nonfatal_after_complete_cleanup() -> None:
    manager = ProviderRuntimeManager(_settings("nvidia_nim/model"))
    runtime = ApplicationRuntime(
        manager, configuration=AsyncMock(spec=ConfigurationService), transcriber=None
    )

    with (
        patch(
            "free_claude_code.runtime.messaging_service.messaging_platform_factory.create_messaging_components",
            side_effect=RuntimeError("messaging unavailable"),
        ),
        patch.object(
            runtime._messaging, "close_delivery", AsyncMock(return_value=True)
        ),
    ):
        await runtime._messaging._start_messaging_if_configured()

    await manager.close()


@pytest.mark.asyncio
async def test_messaging_start_failure_fails_closed_when_cleanup_is_incomplete() -> (
    None
):
    manager = ProviderRuntimeManager(_settings("nvidia_nim/model"))
    runtime = ApplicationRuntime(
        manager, configuration=AsyncMock(spec=ConfigurationService), transcriber=None
    )

    with (
        patch(
            "free_claude_code.runtime.messaging_service.messaging_platform_factory.create_messaging_components",
            side_effect=RuntimeError("messaging unavailable"),
        ),
        patch.object(
            runtime._messaging, "close_delivery", AsyncMock(return_value=False)
        ),
        pytest.raises(RuntimeError, match="cleanup incomplete") as exc_info,
    ):
        await runtime._messaging._start_messaging_if_configured()

    assert isinstance(exc_info.value.__cause__, RuntimeError)
    await manager.close()


@pytest.mark.asyncio
async def test_composition_records_runtime_before_workspace_setup() -> None:
    events: list[str] = []
    manager = ProviderRuntimeManager(_settings("nvidia_nim/model"))
    runtime = ApplicationRuntime(
        manager, configuration=AsyncMock(spec=ConfigurationService), transcriber=None
    )
    messaging = TrackingMessagingRuntime(events)
    components = MessagingPlatformComponents(
        name="tracking",
        runtime=messaging,
        outbound=MagicMock(),
    )

    with (
        patch(
            "free_claude_code.runtime.messaging_service.os.makedirs",
            side_effect=OSError("workspace failed"),
        ),
        pytest.raises(OSError, match="workspace failed"),
    ):
        await runtime._messaging._start_messaging_workflow(components)

    assert runtime._messaging._messaging_runtime is messaging

    await runtime.close()

    assert events == ["messaging.quiesce", "messaging.close"]
    assert runtime._closed is True


@pytest.mark.asyncio
async def test_composition_publishes_startup_notice_after_runtime_and_repair() -> None:
    events: list[str] = []
    manager = ProviderRuntimeManager(_settings("nvidia_nim/model"))
    runtime = ApplicationRuntime(
        manager, configuration=AsyncMock(spec=ConfigurationService), transcriber=None
    )
    messaging = TrackingMessagingRuntime(events)
    notice = MessagingStartupNotice(
        chat_id="chat",
        transport_label="test transport",
    )
    components = MessagingPlatformComponents(
        name="tracking",
        runtime=messaging,
        outbound=MagicMock(),
        startup_notice=notice,
    )
    workflow = MagicMock()
    workflow.handle_message = AsyncMock()
    workflow.restore = AsyncMock(side_effect=lambda: events.append("workflow.restore"))
    workflow.repair_restored_statuses = AsyncMock(
        side_effect=lambda: events.append("workflow.repair")
    )
    workflow.publish_startup_notice = AsyncMock(
        side_effect=lambda published: events.append("workflow.notice")
    )
    workflow.close = AsyncMock()
    cli_manager = MagicMock()

    with (
        patch(
            "free_claude_code.cli.managed.ManagedClaudeSessionManager",
            return_value=cli_manager,
        ) as manager_constructor,
        patch.object(runtime._messaging, "_messaging_store", AsyncMock()),
        patch(
            "free_claude_code.messaging.workflow.MessagingWorkflow",
            return_value=workflow,
        ),
    ):
        runtime.http_started()
        await runtime._messaging._start_messaging_workflow(components)

    assert events == [
        "workflow.restore",
        "messaging.start",
        "workflow.repair",
        "workflow.notice",
    ]
    workflow.publish_startup_notice.assert_awaited_once_with(notice)
    assert manager_constructor.call_args.kwargs["proxy_root_url"] == (
        "http://127.0.0.1:8082"
    )
    assert "api_url" not in manager_constructor.call_args.kwargs
    assert "plans_directory" not in manager_constructor.call_args.kwargs

    assert await runtime.close() is True


@pytest.mark.asyncio
async def test_folder_picker_is_stopped_before_http_shutdown_drains(monkeypatch):
    import uvicorn

    from free_claude_code.cli.uvicorn_server import RuntimeServer

    runtime, _manager = _runtime_with_admin_provider(AdminModelProvider())
    started = asyncio.Event()
    signals = []

    async def select(_initial, stop):
        signals.append(stop)
        started.set()
        await stop.wait()
        return None

    monkeypatch.setattr(runtime._folder_picker, "_select", select)
    request = asyncio.create_task(runtime.pick_folder(None))
    await started.wait()

    async def drain(_server, sockets=None):
        assert signals[0].is_set()
        assert await request is None

    monkeypatch.setattr(uvicorn.Server, "shutdown", drain)
    server = RuntimeServer(
        uvicorn.Config("unused:app"),
        begin_shutdown=runtime.begin_shutdown,
        on_started=runtime.http_started,
        close_runtime=runtime.close,
    )
    await server.shutdown()
    assert await runtime.close() is True