Julian Hyde created CALCITE-4723: ------------------------------------ Summary: Check whether JDBC adapter generates "GROUP BY ()" against Oracle, DB2, MSSQL Key: CALCITE-4723 URL: https://issues.apache.org/jira/browse/CALCITE-4723 Project: Calcite Issue Type: Bug Reporter: Julian Hyde
Oracle, DB2 and MSSQL have non-standard semantics for "GROUP BY ()". Standard behavior is to always return one "grand total" row, but [Oracle, DB2 and MSSQL return no rows if the input is empty|https://blog.jooq.org/2018/05/25/how-to-group-by-nothing-in-sql/]. Calcite's semantics is that "GROUP BY ()" always returns one row, and the JDBC adapter currently assumes that all back ends have the same semantics. On back ends that have different semantics, some queries might be giving incorrect results. I suggest the following remedy: * Add a {{SqlDialect}} method {{boolean omitGrandTotalOnEmptyInput()}} * Run the test suite, and see whether we ever generate "GROUP BY ()" on one of the affected dialects. Try to write a test case where we do this. * Modify the dialects to generate safe SQL in these cases (possibly "GROUP BY ()", or possibly something else). As the above article notes, it is particularly difficult to find SQL that works for MSSQL, because it bumps into the no-constants rule (see CALCITE-4702) -- This message was sent by Atlassian Jira (v8.3.4#803005)