import copy import json import sys import threading import unittest from pathlib import Path from unittest.mock import patch ROOT = Path(__file__).resolve().parents[1] sys.path[:0] = [str(ROOT/'scripts'), str(Path(__file__).parent)] from dsh_row_selection import selection_hint, select_rows, ranking_chart, validate_selection from dsh_model_input import canonical_proposal, planning_input, compact_tool_result from nao_agent_task import normalize_task, plan_task from nao_native_exploration import prepare_step from nao_quality_tools import QualityToolError, QualityTools, tool_definition from test_nao_agent_task import task_value from test_nao_quality_tools import RELEASE from test_native_exploration import root QUESTION = '帮我筛选,10月份合格率最低TOP20产品。且重量大于1000KG,并生成报告' def selection(limit=20, stage='any'): value = selection_hint(QUESTION) value['limit'] = limit value['weightFilter']['stage'] = stage return value def population(): rows = [{'key': f'P{i:03d}', 'label': f'产品{i:03d}', 'compositeRate': .5 + i/1000, 'black': {'rate': .99-i/10000, 'totalWeight': 1001 if i % 2 else 10}, 'bright': {'rate': .98, 'totalWeight': 1001 if not i % 2 else 10}} for i in range(150)] rows.insert(0, {'key': 'LOW_WEIGHT', 'compositeRate': .01, 'black': {'totalWeight': 1000}, 'bright': {'totalWeight': 1000}}) return {'kind': 'quality_metrics', 'releaseId': RELEASE, 'view': 'product', 'period': {'year': 2026, 'month': 10}, 'filters': {}, 'rows': rows, 'rowCount': len(rows), 'rowsReturned': len(rows), 'rowsTruncated': False} class AggregateSelectionTests(unittest.TestCase): def test_captured_failure_is_bound_as_aggregate_condition_not_source_filter(self): captured = task_value(period={'year':2026,'month':10}, fields=['rate'], stages=['black'], dimensions=['product'], filterUpdates={'extraFilters':[{'key':'totalWeight','operator':'gt','value':1000}]}) captured.update(needsClarification=True, clarification='请确认黑片、亮片还是任一阶段投入重量?') before = copy.deepcopy(captured) value = canonical_proposal(captured, question=QUESTION) task = normalize_task(value, QUESTION, {'_businessCatalog': {'years': [2026]}}) self.assertTrue(task['needsClarification']) self.assertEqual(task['aggregateSelection']['weightFilter']['value'], 1000) self.assertNotIn('extraFilters', task['filters']) self.assertEqual(captured, before) def test_clarification_can_resume_same_month_top_and_report_request(self): previous = root() previous.update(needsClarification=True, aggregateSelection=selection(stage=None), period={'year': 2026, 'month': 10}) context = {'_agentTask': previous} for answer, expected in [('黑片or亮片', 'any'), ('我的选择:任一阶段重量大于1000KG', 'any'), ('两阶段均满足', 'all'), ('黑片和亮片都要大于1000', 'all'), ('亮片', 'bright'), ('黑片', 'black')]: fixed = canonical_proposal(task_value(), question=answer, context=context) self.assertEqual(fixed['aggregateSelection']['weightFilter']['stage'], expected) self.assertEqual(fixed['aggregateSelection']['requestQuestion'], QUESTION) self.assertEqual(fixed['period'], {'year': 2026, 'month': 10}) self.assertFalse(fixed['needsClarification']) hint = planning_input({'question': QUESTION})['aggregateSelectionHint'] self.assertIsNone(hint['weightFilter']['stage']) def test_weight_only_model_alias_uses_existing_rank_contract_without_losing_conditions(self): question='2026年10月,按亮片投入重量由高到低列出产品Top100,不生成报告。' value=task_value(fields=['totalWeight'],stages=['bright'],dimensions=['product'], aggregateSelection={'dimension':'product','metric':'totalWeight','rateStage':'bright','direction':'desc','limit':100,'requestQuestion':question}) fixed=canonical_proposal(value,question=question) self.assertNotIn('aggregateSelection',fixed) self.assertEqual(fixed['fields'],['totalWeight']) self.assertEqual(fixed['stages'],['bright']) predicate=copy.deepcopy(value);predicate['aggregateSelection']['weightFilter']={'metric':'totalWeight','operator':'gt','value':1000,'stage':'bright'} self.assertIn('aggregateSelection',canonical_proposal(predicate,question=question)) def test_filter_precedes_sort_and_top_and_does_not_recalculate_composite(self): source = population(); before = copy.deepcopy(source) selected = select_rows(source, selection()) self.assertEqual(selected['eligibleRows'], 150) self.assertEqual([row['key'] for row in selected['rows']], [f'P{i:03d}' for i in range(20)]) self.assertEqual(selected['rows'][0]['selectedRate'], .5) self.assertEqual(source, before) all_stages = select_rows(source, selection(stage='all')) self.assertEqual(all_stages['returnedRows'], 0) def test_top100_and_chart_panels_keep_every_rank(self): source = population(); task = {'aggregateSelection': selection(100)} view = compact_tool_result(source, task=task) self.assertEqual(len(view['rows']), 100) self.assertEqual(view['rowCount'], 151) self.assertEqual(view['selectionDisplay']['eligibleRows'], 150) charts = ranking_chart({'result': source, 'evidenceRef': 'Q12345678'}, task, 'test-turn') self.assertEqual(len(charts), 4) self.assertEqual(sum(len(chart['categories']) for chart in charts), 100) self.assertTrue(all(chart['series'][0]['key'] == chart['metric'] for chart in charts)) self.assertEqual(charts[-1]['categories'][-1]['label'], '100. 产品099') def test_missing_invalid_rows_and_ties_do_not_become_zero_or_fake_rates(self): source = population() source['rows'][1]['compositeRate'] = None source['rows'][2]['compositeRate'] = float('nan') source['rows'][3]['compositeRate'] = source['rows'][4]['compositeRate'] selected = select_rows(source, selection()) self.assertEqual(selected['missingRateRows'], 2) self.assertEqual([row['key'] for row in selected['rows'][:2]], ['P002', 'P003']) source['rowsTruncated'] = True with self.assertRaises(ValueError): select_rows(source, selection()) def test_limit_bounds_and_batch_scope_keep_root_filters(self): for count in (0, 101): with self.assertRaises(ValueError): validate_selection(selection(count)) task = root(); task.update(aggregateSelection=selection(100), comparisonPeriod=None) args, proposal = prepare_step({'mode': 'query', 'view': 'product'}, task, {'releaseId': RELEASE, 'scopes': []}) self.assertEqual(proposal['node']['dimension'], 'product') self.assertEqual(args['limit'], 500) self.assertEqual(args['period'], task['period']) schema = tool_definition(native_protocol=True, task=task, release_id=RELEASE)['inputSchema'] self.assertIn('product', schema['properties']['view']['enum']) task.update(stages=['black'], fields=['totalWeight'], dimensions=['product']) args, proposal = prepare_step({'mode': 'rank', 'dimension': 'product', 'metric': 'totalWeight', 'stage': 'black', 'limit': 100}, task, {'releaseId': RELEASE, 'scopes': []}) self.assertEqual(args['limit'], 100) with self.assertRaises(QualityToolError): prepare_step({**args, 'limit': 101}, task, {'releaseId': RELEASE, 'scopes': []}) def test_model_clarification_cannot_enter_native_execution(self): value = task_value(executionMode='native_agent', needsClarification=True, clarification='重量筛选用黑片、亮片、任一阶段还是两阶段都满足?', clarificationRequest=['黑片','亮片','任一阶段','两阶段均满足'], period={'year': 2026, 'month': 10}, dimensions=['product']) context = {'_protocol': {'autonomousEnabled': True}, '_businessCatalog': {'years': [2026]}} with patch('nao_agent_task._model_json', return_value=value): task = plan_task(base_url='http://test', api_key='test', model='test', question=QUESTION, context=context, cancel_event=threading.Event(), checks_enabled=False) self.assertTrue(task['needsClarification']) self.assertNotEqual(task.get('executionMode'), 'native_agent') self.assertEqual(task['clarification'], value['clarification']) self.assertEqual([entry['value'] for entry in task['clarificationRequest']['options']], value['clarificationRequest']) def test_unrelated_new_tasks_and_additional_ambiguity_are_not_swallowed(self): previous = root(); previous.update(needsClarification=True, aggregateSelection=selection(stage=None)) context = {'_agentTask': previous} self.assertIsNone(selection_hint('新任务,查询综合率', context)) value = task_value(needsClarification=True, clarification='产品有多个候选,请选择具体规格。') fixed = canonical_proposal(value, question='亮片', context=context) self.assertTrue(fixed['needsClarification']) self.assertEqual(fixed['clarification'], value['clarification']) self.assertEqual(selection_hint('改成Top100,亮片', context)['limit'], 100) def test_unpaginated_metrics_do_not_require_pagination_only_rowcount(self): raw = {'summary': {'key':'overall','compositeRate': .95}, 'context': {}, 'rows':[{'key':'P1','compositeRate':.9}], 'coverage': {}, 'trend':[]} tools = object.__new__(QualityTools) with patch.object(tools, '_metrics', return_value=raw) as read: result = tools._query({'id': RELEASE}, {'year':2026,'month':10}, {}, 'product', 500) self.assertEqual(result['rowCount'], 1) self.assertFalse(result['rowsTruncated']) self.assertIsNone(read.call_args.kwargs['row_limit']) raw['context']['rowCount'] = 2 with patch.object(tools, '_metrics', return_value=raw): with self.assertRaises(QualityToolError): tools._query({'id':RELEASE}, {'year':2026,'month':10}, {}, 'product', 500) def test_product_batch_charts_accept_only_exact_signed_dimension_scope(self): from analysis_visual_evidence import protocol_visual_result value = population() value['protocolEvidence'] = {'nodeId':'products', 'scopeId':'scope-products','queryId':'q-products'} document = {'mode':'query','releaseId':RELEASE,'result':value,'evidenceRef':'Q12345678'} task = root(); task.update(period=value['period'], comparisonPeriod=None, comparisonRequired=False, aggregateSelection=selection(100), dimensions=['product'], fields=['rate','totalWeight'], stages=['black','bright']) scope = {'nodeId':'products','scopeId':'scope-products','purpose':'metrics','task':{**task,'kind':'metrics'}} result = protocol_visual_result([document],[scope],QUESTION,'test-turn',release_id=RELEASE,base_context={'_agentTask':task}) self.assertEqual(result['visualState'], 'ready') self.assertEqual(sum(len(chart['categories']) for chart in result['visuals']), 100) changed = copy.deepcopy(document); changed['result']['protocolEvidence']['scopeId'] = 'wrong-scope' denied = protocol_visual_result([changed],[scope],QUESTION,'test-turn',release_id=RELEASE,base_context={'_agentTask':task}) self.assertEqual(denied['visuals'], []) empty=copy.deepcopy(task);empty['aggregateSelection']['weightFilter']['value']=1e12 no_chart=protocol_visual_result([document],[scope],QUESTION,'test-turn',release_id=RELEASE,base_context={'_agentTask':empty}) self.assertEqual(no_chart['visualState'],'not_requested') def test_initial_top_comparison_read_is_never_a_single_period_weight_rank(self): import runpy initial = runpy.run_path(str(ROOT/'runtime/dsh/application/quality_native_run.py'))['initial_read_arguments'] task = root();task.update(freshQuery=True, dimensions=['product'], fields=['rate','totalWeight'], stages=['bright']) self.assertEqual(initial(task,'合格率下降最多Top100产品'), {'mode':'compare'}) task['aggregateSelection']=selection(100) self.assertEqual(initial(task,'合格率下降最多Top100产品'), {'mode':'compare'}) def test_two_period_rate_change_uses_existing_changes_contract(self): question='比较2026年9月和10月,列出亮片合格率下降最多的产品Top100' value=task_value(fields=['rate'],stages=['bright'],dimensions=['product'], aggregateSelection={'dimension':'product','metric':'rate','rateStage':'bright','direction':'asc','limit':100,'requestQuestion':question}) fixed=canonical_proposal(value,question=question) self.assertNotIn('aggregateSelection',fixed) self.assertEqual(fixed['period'],{'year':2026,'month':10}) self.assertEqual(fixed['comparisonPeriod'],{'year':2026,'month':9}) if __name__ == '__main__': unittest.main()