| 1 | """Tests for main turn-loop orchestration.""" |
| 2 | |
| 3 | from __future__ import annotations |
| 4 | |
| 5 | from pathlib import Path |
| 6 | |
| 7 | import pytest |
| 8 | |
| 9 | from loader.agent.loop import Agent, AgentConfig |
| 10 | from loader.runtime.conversation import ConversationRuntime |
| 11 | from loader.runtime.turn_iteration import TurnIterationAction, TurnIterationDecision |
| 12 | from loader.runtime.turn_preamble import TurnPreludeDecision |
| 13 | from tests.helpers.runtime_harness import ScriptedBackend |
| 14 | |
| 15 | |
| 16 | def non_streaming_config() -> AgentConfig: |
| 17 | """Shared config for direct turn-loop tests.""" |
| 18 | |
| 19 | return AgentConfig(auto_context=False, stream=False, max_iterations=8) |
| 20 | |
| 21 | |
| 22 | async def _prepare_runtime( |
| 23 | runtime: ConversationRuntime, |
| 24 | *, |
| 25 | task: str, |
| 26 | ) -> tuple: |
| 27 | events = [] |
| 28 | |
| 29 | async def capture(event) -> None: |
| 30 | events.append(event) |
| 31 | |
| 32 | prepared = await runtime.turn_preparation.prepare( |
| 33 | task=task, |
| 34 | emit=capture, |
| 35 | requested_mode="execute", |
| 36 | original_task=None, |
| 37 | on_user_question=None, |
| 38 | ) |
| 39 | return prepared, events, capture |
| 40 | |
| 41 | |
| 42 | @pytest.mark.asyncio |
| 43 | async def test_turn_loop_retries_after_preamble_continue( |
| 44 | temp_dir: Path, |
| 45 | ) -> None: |
| 46 | backend = ScriptedBackend() |
| 47 | agent = Agent( |
| 48 | backend=backend, |
| 49 | config=non_streaming_config(), |
| 50 | project_root=temp_dir, |
| 51 | ) |
| 52 | runtime = ConversationRuntime(agent) |
| 53 | prepared, _, capture = await _prepare_runtime( |
| 54 | runtime, |
| 55 | task="Explain the runtime loop shape.", |
| 56 | ) |
| 57 | prelude_calls: list[int] = [] |
| 58 | iteration_calls: list[int] = [] |
| 59 | |
| 60 | async def fake_preamble(**kwargs): |
| 61 | prelude_calls.append(kwargs["iterations"]) |
| 62 | kwargs["summary"].iterations = kwargs["iterations"] |
| 63 | return TurnPreludeDecision(should_continue=kwargs["iterations"] == 1) |
| 64 | |
| 65 | async def fake_iteration(**kwargs): |
| 66 | iteration_calls.append(kwargs["iterations"]) |
| 67 | return TurnIterationDecision( |
| 68 | action=TurnIterationAction.COMPLETE, |
| 69 | continuation_count=0, |
| 70 | empty_retry_count=0, |
| 71 | extracted_iterations=0, |
| 72 | consecutive_errors=0, |
| 73 | ) |
| 74 | |
| 75 | runtime.turn_preamble.prepare_iteration = fake_preamble |
| 76 | runtime.turn_iteration.run_iteration = fake_iteration |
| 77 | |
| 78 | loop_exit = await runtime.turn_loop.run_loop( |
| 79 | task=prepared.task, |
| 80 | effective_task=prepared.effective_task, |
| 81 | original_task=None, |
| 82 | effective_max_tokens=prepared.effective_max_tokens, |
| 83 | dod=prepared.definition_of_done, |
| 84 | emit=capture, |
| 85 | summary=prepared.summary, |
| 86 | executor=prepared.executor, |
| 87 | rollback_plan=prepared.rollback_plan, |
| 88 | on_confirmation=None, |
| 89 | on_user_question=None, |
| 90 | emit_confirmation=runtime._emit_confirmation(capture), |
| 91 | ) |
| 92 | |
| 93 | assert prelude_calls == [1, 2] |
| 94 | assert iteration_calls == [2] |
| 95 | assert prepared.summary.iterations == 2 |
| 96 | assert loop_exit.reason_code == "turn_complete" |
| 97 | assert loop_exit.reason_summary == "Finalizing completed turn" |
| 98 | |
| 99 | |
| 100 | @pytest.mark.asyncio |
| 101 | async def test_turn_loop_carries_iteration_state_between_attempts( |
| 102 | temp_dir: Path, |
| 103 | ) -> None: |
| 104 | backend = ScriptedBackend() |
| 105 | agent = Agent( |
| 106 | backend=backend, |
| 107 | config=non_streaming_config(), |
| 108 | project_root=temp_dir, |
| 109 | ) |
| 110 | runtime = ConversationRuntime(agent) |
| 111 | prepared, _, capture = await _prepare_runtime( |
| 112 | runtime, |
| 113 | task="Inspect the loop bookkeeping.", |
| 114 | ) |
| 115 | iteration_inputs: list[tuple[int, int, int, int, int, list[str]]] = [] |
| 116 | |
| 117 | async def fake_preamble(**kwargs): |
| 118 | kwargs["summary"].iterations = kwargs["iterations"] |
| 119 | return TurnPreludeDecision() |
| 120 | |
| 121 | async def fake_iteration(**kwargs): |
| 122 | iteration_inputs.append( |
| 123 | ( |
| 124 | kwargs["iterations"], |
| 125 | kwargs["continuation_count"], |
| 126 | kwargs["empty_retry_count"], |
| 127 | kwargs["extracted_iterations"], |
| 128 | kwargs["consecutive_errors"], |
| 129 | list(kwargs["actions_taken"]), |
| 130 | ) |
| 131 | ) |
| 132 | if len(iteration_inputs) == 1: |
| 133 | return TurnIterationDecision( |
| 134 | action=TurnIterationAction.CONTINUE, |
| 135 | continuation_count=1, |
| 136 | empty_retry_count=2, |
| 137 | extracted_iterations=1, |
| 138 | consecutive_errors=1, |
| 139 | new_actions_taken=["read: {'file_path': 'README.md'}"], |
| 140 | ) |
| 141 | return TurnIterationDecision( |
| 142 | action=TurnIterationAction.FINALIZE, |
| 143 | continuation_count=1, |
| 144 | empty_retry_count=2, |
| 145 | extracted_iterations=1, |
| 146 | consecutive_errors=0, |
| 147 | finalize_reason_code="tool_batch_halted", |
| 148 | finalize_reason_summary="Finalizing after halted tool batch", |
| 149 | ) |
| 150 | |
| 151 | runtime.turn_preamble.prepare_iteration = fake_preamble |
| 152 | runtime.turn_iteration.run_iteration = fake_iteration |
| 153 | |
| 154 | loop_exit = await runtime.turn_loop.run_loop( |
| 155 | task=prepared.task, |
| 156 | effective_task=prepared.effective_task, |
| 157 | original_task=None, |
| 158 | effective_max_tokens=prepared.effective_max_tokens, |
| 159 | dod=prepared.definition_of_done, |
| 160 | emit=capture, |
| 161 | summary=prepared.summary, |
| 162 | executor=prepared.executor, |
| 163 | rollback_plan=prepared.rollback_plan, |
| 164 | on_confirmation=None, |
| 165 | on_user_question=None, |
| 166 | emit_confirmation=runtime._emit_confirmation(capture), |
| 167 | ) |
| 168 | |
| 169 | assert iteration_inputs == [ |
| 170 | (1, 0, 0, 0, 0, []), |
| 171 | (2, 1, 2, 1, 1, ["read: {'file_path': 'README.md'}"]), |
| 172 | ] |
| 173 | assert loop_exit.reason_code == "tool_batch_halted" |
| 174 | assert loop_exit.reason_summary == "Finalizing after halted tool batch" |