On Calc optimization of calcite

background

Recently, I asked the community CSE Related work , Community feedback can be used Calc. I don't know much about this at present , So learn .

Code

• RelNode
  Calc It's an interface , One implementation is LogicalCalc. The relevant optimization rules are mentioned in the comments .
  
• Rule

Calc Relevant rules are collected in CALC_RULES in .

• Program
  calc There is a special HepProgram
  
  The program Also encapsulated in standard Of program in

standard Will be in Prepare Called , It means Calcite perform query Unit tests of can be used standard Of program, That is to say, it may trigger Calc Related optimizations .

test

stay JdbcTest Write the following test cases in Calcite adopt RexProgram Support CSE

  @Test void testCommonSubExpression() throws SQLException {
    
    CalciteAssert.hr()
        .query("select \"salary\","
            + "12 * \"salary\", "
            + "3 * \"salary\", "
            + "\"salary\" * 12 + \"salary\" * 3,"
            + "\"salary\" * 3 + \"salary\" * 12 from \"hr\".\"emps\"")
        .returns("");
  }

As shown in the figure below ：salary * 12 + salary * 3 and salary * 3 + salary * 12 All point to $t13, so Calcite extract CSE Support simple commutative law

Notice the conditionExpr by null, combination LogicalCalc It's not difficult to guess Filter It is also possible to trigger this optimization , And build new program.

in addition , Finished SQL and no CSE The optimization of the , That is to say, it cannot be reused directly by other engines .