#coding:utf-8 """ ID: issue-2028 ISSUE: 2028 TITLE: Correlated subquery is optimized badly if it depends on the union stream DESCRIPTION: JIRA: CORE-1607 FBTEST: bugs.core_1607 """ import pytest from firebird.qa import * db = db_factory() test_script = """SET PLANONLY ON; select 1 from ( select rdb$relation_name, ( select 1 from rdb$database ) as c from rdb$relations ) r where exists ( select * from rdb$relation_fields f where f.rdb$relation_name = r.rdb$relation_name ); select 1 from ( select * from rdb$relations union all select * from rdb$relations ) r where exists ( select * from rdb$relation_fields f where f.rdb$relation_name = r.rdb$relation_name ); select ( select first 1 r.rdb$relation_name from rdb$relations r where r.rdb$relation_id = d.rdb$relation_id - 1 ) from ( select * from rdb$database union all select * from rdb$database ) d; """ act = isql_act('db', test_script) expected_stdout = """ PLAN (R RDB$DATABASE NATURAL) PLAN (F INDEX (RDB$INDEX_4)) PLAN (R RDB$RELATIONS NATURAL) PLAN (F INDEX (RDB$INDEX_4)) PLAN (R RDB$RELATIONS NATURAL, R RDB$RELATIONS NATURAL) PLAN (R INDEX (RDB$INDEX_1)) PLAN (D RDB$DATABASE NATURAL, D RDB$DATABASE NATURAL) """ @pytest.mark.version('>=3.0') def test_1(act: Action): act.expected_stdout = expected_stdout act.execute() assert act.clean_stdout == act.clean_expected_stdout