temp/features/steps/plan_cli_coverage_boost_r2_steps.py

"""Step definitions for plan_cli_coverage_boost_r2.feature.

Targets remaining uncovered lines in cleveragents/cli/commands/plan.py:
- Line 153: _plan_spec_dict execution_env_priority fallback to FALLBACK.value
- Lines 785-786: apply command v3 lifecycle path (plan_id given)
- Lines 876-940: _apply_with_id full body + error handlers
- Line 1249: _get_plan_executor fallback when lifecycle_service is None
- Lines 1838-1839: execute_plan plan not found
- Lines 1898-1899: execute_plan PreflightRejection handler
- Lines 1910-1911: execute_plan ValueError handler
- Lines 2904, 2909, 2914: resume_plan rich output else "" branches
- Line 3299: tree_decisions_cmd TABLE format show_superseded branch
- Line 3331: tree_decisions_cmd TABLE format long chosen_option truncation
"""

from __future__ import annotations

from datetime import datetime
from unittest.mock import MagicMock, patch

from behave import given, then, when
from behave.runner import Context
from typer.testing import CliRunner

from cleveragents.application.services.plan_lifecycle_service import (
    InvalidPhaseTransitionError,
    PlanNotReadyError,
)
from cleveragents.application.services.plan_preflight_guardrail import (
    PreflightCheckName,
    PreflightRejection,
)
from cleveragents.cli.commands.plan import (
    _plan_spec_dict,
)
from cleveragents.cli.commands.plan import (
    app as plan_app,
)
from cleveragents.core.exceptions import CleverAgentsError
from cleveragents.domain.models.core.decision import Decision, DecisionType
from cleveragents.domain.models.core.plan import (
    ExecutionEnvPriority,
    NamespacedName,
    Plan,
    PlanIdentity,
    PlanPhase,
    PlanTimestamps,
    ProcessingState,
)
from cleveragents.domain.models.core.resume import ResumeSummary

_ULID_A = "01ARZ3NDEKTSV4RRFFQ69G5FAV"
_ULID_B = "01ARZ3NDEKTSV4RRFFQ69G5FBV"

_PATCH_CONTAINER = "cleveragents.application.container.get_container"
_PATCH_GET_LIFECYCLE = "cleveragents.cli.commands.plan._get_lifecycle_service"
_PATCH_GET_EXECUTOR = "cleveragents.cli.commands.plan._get_plan_executor"
_PATCH_GET_PROJECT = "cleveragents.cli.commands.plan._get_current_project"


def _make_lifecycle_plan(
    *,
    plan_id: str = _ULID_A,
    name: str = "local/test-plan",
    phase: PlanPhase = PlanPhase.EXECUTE,
    processing_state: ProcessingState = ProcessingState.COMPLETE,
    read_only: bool = False,
    execution_environment: str | None = None,
    execution_env_priority: ExecutionEnvPriority | None = None,
) -> Plan:
    """Build a real Plan object for testing."""
    plan = Plan(
        identity=PlanIdentity(plan_id=plan_id),
        namespaced_name=NamespacedName.parse(name),
        action_name="local/test-action",
        description="Coverage test plan",
        phase=phase,
        processing_state=processing_state,
        project_links=[],
        timestamps=PlanTimestamps(
            created_at=datetime(2025, 6, 15, 10, 0, 0),
            updated_at=datetime(2025, 6, 15, 11, 0, 0),
        ),
        reusable=True,
        read_only=read_only,
    )
    if execution_environment is not None:
        plan.execution_environment = execution_environment
    if execution_env_priority is not None:
        plan.execution_env_priority = execution_env_priority
    return plan


def _cleanup_patches(context: Context) -> None:
    """Stop all patches registered on context."""
    for stop_fn in getattr(context, "_r2boost_cleanups", []):
        stop_fn()
    context._r2boost_cleanups = []


# ======================================================================
# Given — CLI runner
# ======================================================================


@given("a r2boost CLI runner")
def step_r2boost_cli_runner(context: Context) -> None:
    context.r2boost_runner = CliRunner()
    context.r2boost_result = None
    if not hasattr(context, "_r2boost_cleanups"):
        context._r2boost_cleanups = []
    context.add_cleanup(lambda: _cleanup_patches(context))


# ======================================================================
# _plan_spec_dict: execution_env_priority fallback (line 153)
# ======================================================================


@given('a v3 plan with execution_environment "{env}" and execution_env_priority None')
def step_plan_with_exec_env_no_priority(context: Context, env: str) -> None:
    context.r2boost_plan = _make_lifecycle_plan(
        execution_environment=env,
        execution_env_priority=None,
    )


@when("I call _plan_spec_dict for r2 coverage")
def step_call_plan_spec_dict_r2(context: Context) -> None:
    context.r2boost_spec_dict = _plan_spec_dict(context.r2boost_plan)


@then('the r2 spec dict should contain "{key}" with value "{value}"')
def step_r2_spec_dict_contains(context: Context, key: str, value: str) -> None:
    assert key in context.r2boost_spec_dict, (
        f"Key '{key}' not in spec dict: {list(context.r2boost_spec_dict.keys())}"
    )
    assert str(context.r2boost_spec_dict[key]) == value, (
        f"Expected '{value}', got '{context.r2boost_spec_dict[key]}'"
    )


# ======================================================================
# apply command: v3 lifecycle happy path (lines 785-786, 876-925)
# ======================================================================


def _make_apply_happy_service() -> MagicMock:
    """Create a lifecycle service mock that walks through the full apply flow."""
    service = MagicMock()

    # Phase 1: pre_plan — Execute/complete
    pre_plan = _make_lifecycle_plan(
        phase=PlanPhase.EXECUTE,
        processing_state=ProcessingState.COMPLETE,
    )

    # Phase 2: after apply_plan -> Apply/queued
    apply_queued = _make_lifecycle_plan(
        phase=PlanPhase.APPLY,
        processing_state=ProcessingState.QUEUED,
    )

    # Phase 3: after start_apply -> Apply/processing
    apply_processing = _make_lifecycle_plan(
        phase=PlanPhase.APPLY,
        processing_state=ProcessingState.PROCESSING,
    )

    # Phase 4: after complete_apply -> Apply/applied
    apply_applied = _make_lifecycle_plan(
        phase=PlanPhase.APPLY,
        processing_state=ProcessingState.APPLIED,
    )

    # get_plan returns different results as apply progresses
    service.get_plan.side_effect = [
        pre_plan,  # First call: pre-flight check
        apply_queued,  # After apply_plan: Apply/queued
        apply_processing,  # After start_apply: Apply/processing
        apply_applied,  # Final fetch after complete_apply
    ]
    service.apply_plan.return_value = None
    service.start_apply.return_value = None
    service.complete_apply.return_value = None
    service._complete_apply_if_queued.return_value = apply_applied
    return service


@given("a mocked lifecycle service for r2boost apply happy path")
def step_mock_apply_happy(context: Context) -> None:
    service = _make_apply_happy_service()
    p = patch(_PATCH_GET_LIFECYCLE, return_value=service)
    p.start()
    context._r2boost_cleanups.append(p.stop)
    # Mock _apply_sandbox_changes to avoid hitting the DI container
    p2 = patch("cleveragents.cli.commands.plan._apply_sandbox_changes")
    p2.start()
    context._r2boost_cleanups.append(p2.stop)
    context.r2boost_service = service


@when("I invoke r2boost apply with plan_id in rich format")
def step_invoke_apply_rich(context: Context) -> None:
    context.r2boost_result = context.r2boost_runner.invoke(
        plan_app, ["apply", "--yes", _ULID_A]
    )


@when("I invoke r2boost apply with plan_id and format json")
def step_invoke_apply_json(context: Context) -> None:
    context.r2boost_result = context.r2boost_runner.invoke(
        plan_app, ["apply", "--yes", _ULID_A, "--format", "json"]
    )


# ======================================================================
# _apply_with_id: plan not found (lines 886-888)
# ======================================================================


@given("a mocked lifecycle service for r2boost apply where plan is not found")
def step_mock_apply_not_found(context: Context) -> None:
    service = MagicMock()
    service.get_plan.return_value = None
    p = patch(_PATCH_GET_LIFECYCLE, return_value=service)
    p.start()
    context._r2boost_cleanups.append(p.stop)


# ======================================================================
# _apply_with_id: plan is read-only (lines 889-893)
# ======================================================================


@given("a mocked lifecycle service for r2boost apply where plan is read-only")
def step_mock_apply_read_only(context: Context) -> None:
    service = MagicMock()
    read_only_plan = _make_lifecycle_plan(read_only=True)
    service.get_plan.return_value = read_only_plan
    p = patch(_PATCH_GET_LIFECYCLE, return_value=service)
    p.start()
    context._r2boost_cleanups.append(p.stop)


# ======================================================================
# _apply_with_id: error handlers (lines 927-940)
# ======================================================================


@given(
    "a mocked lifecycle service for r2boost apply that raises InvalidPhaseTransitionError"
)
def step_mock_apply_invalid_transition(context: Context) -> None:
    service = MagicMock()
    plan = _make_lifecycle_plan(
        phase=PlanPhase.EXECUTE,
        processing_state=ProcessingState.COMPLETE,
    )
    service.get_plan.return_value = plan
    service.apply_plan.side_effect = InvalidPhaseTransitionError(
        from_phase=PlanPhase.EXECUTE,
        to_phase=PlanPhase.APPLY,
    )
    p = patch(_PATCH_GET_LIFECYCLE, return_value=service)
    p.start()
    context._r2boost_cleanups.append(p.stop)


@given("a mocked lifecycle service for r2boost apply that raises PlanNotReadyError")
def step_mock_apply_not_ready(context: Context) -> None:
    service = MagicMock()
    plan = _make_lifecycle_plan(
        phase=PlanPhase.EXECUTE,
        processing_state=ProcessingState.COMPLETE,
    )
    service.get_plan.return_value = plan
    service.apply_plan.side_effect = PlanNotReadyError(
        plan_id=_ULID_A,
        phase=PlanPhase.EXECUTE,
        state=ProcessingState.PROCESSING,
    )
    p = patch(_PATCH_GET_LIFECYCLE, return_value=service)
    p.start()
    context._r2boost_cleanups.append(p.stop)


@given("a mocked lifecycle service for r2boost apply that raises ValueError")
def step_mock_apply_value_error(context: Context) -> None:
    service = MagicMock()
    plan = _make_lifecycle_plan(
        phase=PlanPhase.EXECUTE,
        processing_state=ProcessingState.COMPLETE,
    )
    service.get_plan.return_value = plan
    service.apply_plan.side_effect = ValueError("missing API key for provider")
    p = patch(_PATCH_GET_LIFECYCLE, return_value=service)
    p.start()
    context._r2boost_cleanups.append(p.stop)


@given("a mocked lifecycle service for r2boost apply that raises CleverAgentsError")
def step_mock_apply_clever_agents_error(context: Context) -> None:
    service = MagicMock()
    plan = _make_lifecycle_plan(
        phase=PlanPhase.EXECUTE,
        processing_state=ProcessingState.COMPLETE,
    )
    service.get_plan.return_value = plan
    service.apply_plan.side_effect = CleverAgentsError("general apply failure")
    p = patch(_PATCH_GET_LIFECYCLE, return_value=service)
    p.start()
    context._r2boost_cleanups.append(p.stop)


# ======================================================================
# _get_plan_executor: lifecycle_service=None fallback (line 1249)
# ======================================================================


@given("a mocked container for r2boost plan executor")
def step_mock_container_for_executor(context: Context) -> None:
    if not hasattr(context, "_r2boost_cleanups"):
        context._r2boost_cleanups = []
    mock_container = MagicMock()
    mock_container.provider_registry.return_value = MagicMock()
    mock_lifecycle_svc = MagicMock()
    mock_container.plan_lifecycle_service.return_value = mock_lifecycle_svc

    p_container = patch(_PATCH_CONTAINER, return_value=mock_container)
    p_container.start()
    context._r2boost_cleanups.append(p_container.stop)
    context.r2boost_mock_container = mock_container
    context.r2boost_mock_lifecycle_svc = mock_lifecycle_svc

    context.add_cleanup(lambda: _cleanup_patches(context))


@when("I call _get_plan_executor without lifecycle_service")
def step_call_get_plan_executor_no_svc(context: Context) -> None:
    from cleveragents.cli.commands.plan import _get_plan_executor

    # lifecycle_service is None, so it should call _get_lifecycle_service()
    context.r2boost_executor = _get_plan_executor(lifecycle_service=None)


@then("the r2boost plan executor should be returned")
def step_verify_executor_returned(context: Context) -> None:
    assert context.r2boost_executor is not None
    # Verify that plan_lifecycle_service was called on the container
    context.r2boost_mock_container.plan_lifecycle_service.assert_called_once()


# ======================================================================
# execute_plan: plan not found (lines 1838-1839)
# ======================================================================


@given("a mocked lifecycle service for r2boost execute where plan vanishes")
def step_mock_execute_plan_vanishes(context: Context) -> None:
    service = MagicMock()
    # First get_plan for read_only check returns a plan
    non_read_only_plan = _make_lifecycle_plan(
        phase=PlanPhase.STRATEGIZE,
        processing_state=ProcessingState.QUEUED,
        read_only=False,
    )
    # Second get_plan (after phase check) returns None = plan vanished
    service.get_plan.side_effect = [non_read_only_plan, None]
    service.list_plans.return_value = []

    mock_executor = MagicMock()

    p_svc = patch(_PATCH_GET_LIFECYCLE, return_value=service)
    p_exec = patch(_PATCH_GET_EXECUTOR, return_value=mock_executor)
    p_svc.start()
    p_exec.start()
    context._r2boost_cleanups.extend([p_svc.stop, p_exec.stop])


@when("I invoke r2boost execute with plan_id")
def step_invoke_execute_with_plan_id(context: Context) -> None:
    context.r2boost_result = context.r2boost_runner.invoke(
        plan_app, ["execute", _ULID_A]
    )


# ======================================================================
# execute_plan: PreflightRejection handler (lines 1898-1899)
# ======================================================================


@given("a mocked lifecycle service for r2boost execute that raises PreflightRejection")
def step_mock_execute_preflight_rejection(context: Context) -> None:
    service = MagicMock()
    plan = _make_lifecycle_plan(
        phase=PlanPhase.STRATEGIZE,
        processing_state=ProcessingState.QUEUED,
    )
    service.get_plan.return_value = plan
    service.list_plans.return_value = []

    mock_executor = MagicMock()
    mock_executor.run_strategize.side_effect = PreflightRejection(
        PreflightCheckName.AUTOMATION_POLICY, "budget limit exceeded"
    )

    p_svc = patch(_PATCH_GET_LIFECYCLE, return_value=service)
    p_exec = patch(_PATCH_GET_EXECUTOR, return_value=mock_executor)
    p_svc.start()
    p_exec.start()
    context._r2boost_cleanups.extend([p_svc.stop, p_exec.stop])


# ======================================================================
# execute_plan: ValueError handler (lines 1910-1911)
# ======================================================================


@given("a mocked lifecycle service for r2boost execute that raises ValueError")
def step_mock_execute_value_error(context: Context) -> None:
    service = MagicMock()
    plan = _make_lifecycle_plan(
        phase=PlanPhase.STRATEGIZE,
        processing_state=ProcessingState.QUEUED,
    )
    service.get_plan.return_value = plan
    service.list_plans.return_value = []

    mock_executor = MagicMock()
    mock_executor.run_strategize.side_effect = ValueError(
        "missing OPENAI_API_KEY in environment"
    )

    p_svc = patch(_PATCH_GET_LIFECYCLE, return_value=service)
    p_exec = patch(_PATCH_GET_EXECUTOR, return_value=mock_executor)
    p_svc.start()
    p_exec.start()
    context._r2boost_cleanups.extend([p_svc.stop, p_exec.stop])


# ======================================================================
# resume_plan: rich output with no optional fields (lines 2904, 2909, 2914)
# ======================================================================


@given("a mocked resume service for r2boost with no optional fields")
def step_mock_resume_no_optional(context: Context) -> None:
    summary = ResumeSummary(
        plan_id=_ULID_A,
        phase="execute",
        processing_state="processing",
        last_completed_step=2,
        next_step_index=3,
        total_steps=5,
        decision_id=None,
        last_checkpoint_id=None,
        sandbox_ref=None,
    )
    mock_resume_svc = MagicMock()
    mock_resume_svc.resume_plan.return_value = summary

    mock_lifecycle_svc = MagicMock()
    p_lifecycle = patch(_PATCH_GET_LIFECYCLE, return_value=mock_lifecycle_svc)
    p_resume = patch(
        "cleveragents.application.services.plan_resume_service.PlanResumeService",
        return_value=mock_resume_svc,
    )

    p_lifecycle.start()
    p_resume.start()
    context._r2boost_cleanups.extend([p_lifecycle.stop, p_resume.stop])


@when("I invoke r2boost plan resume in rich format")
def step_invoke_resume_rich(context: Context) -> None:
    context.r2boost_result = context.r2boost_runner.invoke(
        plan_app, ["resume", _ULID_A]
    )


# ======================================================================
# tree_decisions_cmd: TABLE format with show_superseded (line 3299)
# ======================================================================


def _make_decision(
    *,
    decision_id: str,
    plan_id: str = _ULID_A,
    sequence_number: int = 0,
    question: str = "What approach to use?",
    chosen_option: str = "Option A",
    parent_decision_id: str | None = None,
    superseded_by: str | None = None,
) -> Decision:
    """Create a Decision for testing."""
    return Decision(
        decision_id=decision_id,
        plan_id=plan_id,
        sequence_number=sequence_number,
        decision_type=DecisionType.STRATEGY_CHOICE,
        question=question,
        chosen_option=chosen_option,
        parent_decision_id=parent_decision_id,
        superseded_by=superseded_by,
    )


@given("a mocked decision service for r2boost tree with superseded decisions")
def step_mock_tree_superseded(context: Context) -> None:
    mock_container = MagicMock()
    mock_decision_svc = MagicMock()

    decisions = [
        _make_decision(
            decision_id="01ARZ3NDEKTSV4RRFFQ69G5FA1",
            sequence_number=0,
            question="Root question",
            chosen_option="Root choice",
        ),
        _make_decision(
            decision_id="01ARZ3NDEKTSV4RRFFQ69G5FA2",
            sequence_number=1,
            question="Superseded question",
            chosen_option="Old choice",
            parent_decision_id="01ARZ3NDEKTSV4RRFFQ69G5FA1",
            superseded_by="01ARZ3NDEKTSV4RRFFQ69G5FA3",
        ),
        _make_decision(
            decision_id="01ARZ3NDEKTSV4RRFFQ69G5FA3",
            sequence_number=2,
            question="Replacement question",
            chosen_option="New choice",
            parent_decision_id="01ARZ3NDEKTSV4RRFFQ69G5FA1",
        ),
    ]
    mock_decision_svc.list_decisions.return_value = decisions
    mock_container.decision_service.return_value = mock_decision_svc

    p = patch(_PATCH_CONTAINER, return_value=mock_container)
    p.start()
    context._r2boost_cleanups.append(p.stop)


@when("I invoke r2boost tree in table format with show-superseded")
def step_invoke_tree_table_superseded(context: Context) -> None:
    context.r2boost_result = context.r2boost_runner.invoke(
        plan_app,
        ["tree", _ULID_A, "--format", "table", "--show-superseded"],
    )


# ======================================================================
# tree_decisions_cmd: TABLE format long chosen_option (line 3331)
# ======================================================================


@given("a mocked decision service for r2boost tree with long chosen option")
def step_mock_tree_long_option(context: Context) -> None:
    mock_container = MagicMock()
    mock_decision_svc = MagicMock()

    # Create a decision with chosen_option longer than 30 chars
    long_option = "This is a very long chosen option that exceeds thirty characters for truncation testing"
    decisions = [
        _make_decision(
            decision_id="01ARZ3NDEKTSV4RRFFQ69G5FA1",
            sequence_number=0,
            question="A question that is quite long and should be truncated at forty characters boundary",
            chosen_option=long_option,
        ),
    ]
    mock_decision_svc.list_decisions.return_value = decisions
    mock_container.decision_service.return_value = mock_decision_svc

    p = patch(_PATCH_CONTAINER, return_value=mock_container)
    p.start()
    context._r2boost_cleanups.append(p.stop)


@when("I invoke r2boost tree in table format")
def step_invoke_tree_table(context: Context) -> None:
    context.r2boost_result = context.r2boost_runner.invoke(
        plan_app, ["tree", _ULID_A, "--format", "table"]
    )


# ======================================================================
# Common then steps
# ======================================================================


@then("the r2boost command should exit normally")
def step_r2boost_exit_normal(context: Context) -> None:
    result = context.r2boost_result
    assert result is not None, "No result captured"
    assert result.exit_code == 0, (
        f"Expected exit code 0, got {result.exit_code}.\n"
        f"Output: {result.output}\n"
        f"Exception: {result.exception}"
    )


@then("the r2boost command should abort")
def step_r2boost_abort(context: Context) -> None:
    result = context.r2boost_result
    assert result is not None, "No result captured"
    assert result.exit_code != 0, (
        f"Expected non-zero exit code, got {result.exit_code}.\nOutput: {result.output}"
    )


@then('the r2boost output should contain "{text}"')
def step_r2boost_output_contains(context: Context, text: str) -> None:
    output = context.r2boost_result.output if context.r2boost_result else ""
    assert text in output, f"Expected '{text}' in output:\n{output}"


@then('the r2boost output should not contain "{text}"')
def step_r2boost_output_not_contains(context: Context, text: str) -> None:
    output = context.r2boost_result.output if context.r2boost_result else ""
    assert text not in output, f"Did not expect '{text}' in output:\n{output}"