"""One system policy must reach planning, native calls and finalization.""" import io import ast import json import re import sys import threading import unittest from pathlib import Path from unittest.mock import patch sys.path.insert(0, str(Path(__file__).resolve().parents[1] / 'scripts')) class RuntimePolicyTest(unittest.TestCase): def test_production_model_callers_have_no_independent_literal_budget(self): root = Path(__file__).resolve().parents[1] for path in (root/'scripts').glob('nao*.py'): for node in ast.walk(ast.parse(path.read_text())): if isinstance(node,ast.Dict): for key,value in zip(node.keys,node.values): if isinstance(key,ast.Constant) and key.value in ('max_completion_tokens','max_tokens'): self.assertIsInstance(value,ast.Name,str(path)) self.assertEqual(value.id,'MAX_OUTPUT_TOKENS',str(path)) for path in (root/'backend/src/main/java/cn/yunsheng/quality/analysis').glob('*.java'): for value in re.findall(r'"max_(?:completion_)?tokens",\s*([^,\n]+)',path.read_text()): self.assertEqual(value,'AnalysisRuntimePolicy.MAX_OUTPUT_TOKENS',str(path)) def test_planner_uses_global_budget(self): from nao_agent_task import _model_json from nao_quality_runtime import native_output_budget bodies = [] def provider(request, timeout): bodies.append(json.loads(request.data)) return io.BytesIO(b'{"choices":[{"finish_reason":"stop","message":{"content":"{}"}}]}') _model_json(base_url='http://localhost/v1', api_key='fixture', model='fixture', system='fixture', payload={}, cancel_event=threading.Event(), opener=provider) self.assertEqual(bodies[0]['max_completion_tokens'], native_output_budget()) def test_stale_environment_cannot_create_another_budget(self): from nao_quality_runtime import native_output_budget baseline = native_output_budget() with patch.dict('os.environ', {'QUALITY_ANALYSIS_MAX_OUTPUT_TOKENS': '2048'}): self.assertEqual(native_output_budget(), baseline) def test_gate_replaces_both_legacy_caps_even_when_thinking_off(self): from nao_model_gate import ModelGate from nao_quality_runtime import native_output_budget gate = ModelGate(upstream_base='http://localhost/v1', upstream_key='fixture', model='fixture', port=0, client_key='fixture', authorize=lambda: None, thinking_options={'enabled':False, 'effort':'auto', 'protocol':'thinking'}) try: body = gate.apply_thinking({'max_tokens':1024, 'max_completion_tokens':4096}) self.assertNotIn('max_tokens', body) self.assertEqual(body['max_completion_tokens'], native_output_budget()) finally: gate.close() def test_finalization_time_is_reserved_and_deadline_is_shared(self): from nao_runtime_policy import turn_scope, remaining_seconds, request_timeout, native_timeout with patch('nao_runtime_policy.time.monotonic', return_value=100): with turn_scope(): total = remaining_seconds() self.assertLess(native_timeout(), total) with patch('nao_runtime_policy.time.monotonic', return_value=100+total-40): self.assertLessEqual(request_timeout('final'), 35) with self.assertRaises(TimeoutError): request_timeout('planning') if __name__ == '__main__': unittest.main()