cherry pick pingcap#19620 to release-3.0

Signed-off-by: ti-srebot <ti-srebot@pingcap.com>

cmd/explaintest/r/explain_easy.result

@@ -233,6 +233,7 @@ StreamAgg_12	1.00	root	funcs:sum(col_0)
     └─IndexReader_17	10000.00	root	index:IndexScan_16
       └─IndexScan_16	10000.00	cop	table:t2, index:c1, range:[NULL,+inf], keep order:false, stats:pseudo
 explain select 1 in (select c2 from t2) from t1;
+<<<<<<< HEAD
 id	count	task	operator info
 Projection_6	10000.00	root	5_aux_0
 └─HashLeftJoin_7	10000.00	root	CARTESIAN left outer semi join, inner:TableReader_12
@@ -251,6 +252,23 @@ StreamAgg_12	1.00	root	funcs:sum(col_0)
     └─TableReader_18	10.00	root	data:Selection_17
       └─Selection_17	10.00	cop	eq(6, test.t2.c2)
         └─TableScan_16	10000.00	cop	table:t2, range:[-inf,+inf], keep order:false, stats:pseudo
+=======
+id	estRows	task	access object	operator info
+HashJoin_7	10000.00	root		CARTESIAN left outer semi join, other cond:eq(1, test.t2.c2)
+├─TableReader_13(Build)	10000.00	root		data:TableFullScan_12
+│ └─TableFullScan_12	10000.00	cop[tikv]	table:t2	keep order:false, stats:pseudo
+└─TableReader_9(Probe)	10000.00	root		data:TableFullScan_8
+  └─TableFullScan_8	10000.00	cop[tikv]	table:t1	keep order:false, stats:pseudo
+explain select sum(6 in (select c2 from t2)) from t1;
+id	estRows	task	access object	operator info
+StreamAgg_12	1.00	root		funcs:sum(Column#10)->Column#8
+└─Projection_21	10000.00	root		cast(Column#7, decimal(65,0) BINARY)->Column#10
+  └─HashJoin_20	10000.00	root		CARTESIAN left outer semi join, other cond:eq(6, test.t2.c2)
+    ├─TableReader_19(Build)	10000.00	root		data:TableFullScan_18
+    │ └─TableFullScan_18	10000.00	cop[tikv]	table:t2	keep order:false, stats:pseudo
+    └─TableReader_15(Probe)	10000.00	root		data:TableFullScan_14
+      └─TableFullScan_14	10000.00	cop[tikv]	table:t1	keep order:false, stats:pseudo
+>>>>>>> 3ef3e54b5... planner: don't push down null sensitive join conditions (#19620)
 explain format="dot" select sum(t1.c1 in (select c1 from t2)) from t1;
 dot contents
 
@@ -288,16 +306,22 @@ subgraph cluster6{
 node [style=filled, color=lightgrey]
 color=black
 label = "root"
+<<<<<<< HEAD
 "Projection_6" -> "HashLeftJoin_7"
 "HashLeftJoin_7" -> "IndexReader_9"
 "HashLeftJoin_7" -> "TableReader_12"
+=======
+"HashJoin_7" -> "TableReader_9"
+"HashJoin_7" -> "TableReader_13"
+>>>>>>> 3ef3e54b5... planner: don't push down null sensitive join conditions (#19620)
 }
 subgraph cluster8{
 node [style=filled, color=lightgrey]
 color=black
 label = "cop"
 "IndexScan_8"
 }
+<<<<<<< HEAD
 subgraph cluster11{
 node [style=filled, color=lightgrey]
 color=black
@@ -306,6 +330,16 @@ label = "cop"
 }
 "IndexReader_9" -> "IndexScan_8"
 "TableReader_12" -> "Selection_11"
+=======
+subgraph cluster12{
+node [style=filled, color=lightgrey]
+color=black
+label = "cop"
+"TableFullScan_12"
+}
+"TableReader_9" -> "TableFullScan_8"
+"TableReader_13" -> "TableFullScan_12"
+>>>>>>> 3ef3e54b5... planner: don't push down null sensitive join conditions (#19620)
 }
 
 drop table if exists t1, t2, t3, t4;

cmd/explaintest/r/explain_easy_stats.result

@@ -126,6 +126,7 @@ Limit_10	1.00	root	offset:0, count:1
     └─TableScan_18	1.00	cop	table:t1, range:[-inf,+inf], keep order:true, desc
 set @@session.tidb_opt_insubq_to_join_and_agg=0;
 explain select 1 in (select c2 from t2) from t1;
+<<<<<<< HEAD
 id	count	task	operator info
 Projection_6	1999.00	root	5_aux_0
 └─HashLeftJoin_7	1999.00	root	CARTESIAN left outer semi join, inner:TableReader_12
@@ -134,6 +135,14 @@ Projection_6	1999.00	root	5_aux_0
   └─TableReader_12	0.00	root	data:Selection_11
     └─Selection_11	0.00	cop	eq(1, test.t2.c2)
       └─TableScan_10	1985.00	cop	table:t2, range:[-inf,+inf], keep order:false
+=======
+id	estRows	task	access object	operator info
+HashJoin_7	1999.00	root		CARTESIAN left outer semi join, other cond:eq(1, test.t2.c2)
+├─TableReader_13(Build)	1985.00	root		data:TableFullScan_12
+│ └─TableFullScan_12	1985.00	cop[tikv]	table:t2	keep order:false
+└─TableReader_9(Probe)	1999.00	root		data:TableFullScan_8
+  └─TableFullScan_8	1999.00	cop[tikv]	table:t1	keep order:false
+>>>>>>> 3ef3e54b5... planner: don't push down null sensitive join conditions (#19620)
 explain format="dot" select 1 in (select c2 from t2) from t1;
 dot contents
 
@@ -142,16 +151,22 @@ subgraph cluster6{
 node [style=filled, color=lightgrey]
 color=black
 label = "root"
+<<<<<<< HEAD
 "Projection_6" -> "HashLeftJoin_7"
 "HashLeftJoin_7" -> "IndexReader_9"
 "HashLeftJoin_7" -> "TableReader_12"
+=======
+"HashJoin_7" -> "TableReader_9"
+"HashJoin_7" -> "TableReader_13"
+>>>>>>> 3ef3e54b5... planner: don't push down null sensitive join conditions (#19620)
 }
 subgraph cluster8{
 node [style=filled, color=lightgrey]
 color=black
 label = "cop"
 "IndexScan_8"
 }
+<<<<<<< HEAD
 subgraph cluster11{
 node [style=filled, color=lightgrey]
 color=black
@@ -160,6 +175,16 @@ label = "cop"
 }
 "IndexReader_9" -> "IndexScan_8"
 "TableReader_12" -> "Selection_11"
+=======
+subgraph cluster12{
+node [style=filled, color=lightgrey]
+color=black
+label = "cop"
+"TableFullScan_12"
+}
+"TableReader_9" -> "TableFullScan_8"
+"TableReader_13" -> "TableFullScan_12"
+>>>>>>> 3ef3e54b5... planner: don't push down null sensitive join conditions (#19620)
 }
 
 explain select * from index_prune WHERE a = 1010010404050976781 AND b = 26467085526790 LIMIT 1;

executor/executor_test.go

@@ -2096,6 +2096,32 @@ func (s *testSuite) TestRow(c *C) {
 	result.Check(testkit.Rows("0"))
 	result = tk.MustQuery("select (select 1)")
 	result.Check(testkit.Rows("1"))
+
+	tk.MustExec("drop table if exists t1")
+	tk.MustExec("create table t1 (a int, b int)")
+	tk.MustExec("insert t1 values (1,2),(1,null)")
+	tk.MustExec("drop table if exists t2")
+	tk.MustExec("create table t2 (c int, d int)")
+	tk.MustExec("insert t2 values (0,0)")
+
+	tk.MustQuery("select * from t2 where (1,2) in (select * from t1)").Check(testkit.Rows("0 0"))
+	tk.MustQuery("select * from t2 where (1,2) not in (select * from t1)").Check(testkit.Rows())
+	tk.MustQuery("select * from t2 where (1,1) not in (select * from t1)").Check(testkit.Rows())
+	tk.MustQuery("select * from t2 where (1,null) in (select * from t1)").Check(testkit.Rows())
+	tk.MustQuery("select * from t2 where (null,null) in (select * from t1)").Check(testkit.Rows())
+
+	tk.MustExec("delete from t1 where a=1 and b=2")
+	tk.MustQuery("select (1,1) in (select * from t2) from t1").Check(testkit.Rows("0"))
+	tk.MustQuery("select (1,1) not in (select * from t2) from t1").Check(testkit.Rows("1"))
+	tk.MustQuery("select (1,1) in (select 1,1 from t2) from t1").Check(testkit.Rows("1"))
+	tk.MustQuery("select (1,1) not in (select 1,1 from t2) from t1").Check(testkit.Rows("0"))
+
+	// MySQL 5.7 returns 1 for these 2 queries, which is wrong.
+	tk.MustQuery("select (1,null) not in (select 1,1 from t2) from t1").Check(testkit.Rows("<nil>"))
+	tk.MustQuery("select (t1.a,null) not in (select 1,1 from t2) from t1").Check(testkit.Rows("<nil>"))
+
+	tk.MustQuery("select (1,null) in (select * from t1)").Check(testkit.Rows("<nil>"))
+	tk.MustQuery("select (1,null) not in (select * from t1)").Check(testkit.Rows("<nil>"))
 }
 
 func (s *testSuite) TestColumnName(c *C) {

executor/join_test.go

@@ -677,6 +677,65 @@ func (s *testSuite2) TestSubquery(c *C) {
 	tk.MustExec("insert into t2 values(1)")
 	tk.MustQuery("select * from t1 where a in (select a from t2)").Check(testkit.Rows("1"))
 
+	tk.MustExec("insert into t2 value(null)")
+	tk.MustQuery("select * from t1 where 1 in (select b from t2)").Check(testkit.Rows("1"))
+	tk.MustQuery("select * from t1 where 1 not in (select b from t2)").Check(testkit.Rows())
+	tk.MustQuery("select * from t1 where 2 not in (select b from t2)").Check(testkit.Rows())
+	tk.MustQuery("select * from t1 where 2 in (select b from t2)").Check(testkit.Rows())
+	tk.MustQuery("select 1 in (select b from t2) from t1").Check(testkit.Rows("1"))
+	tk.MustQuery("select 1 in (select 1 from t2) from t1").Check(testkit.Rows("1"))
+	tk.MustQuery("select 1 not in (select b from t2) from t1").Check(testkit.Rows("0"))
+	tk.MustQuery("select 1 not in (select 1 from t2) from t1").Check(testkit.Rows("0"))
+
+	tk.MustExec("delete from t2 where b=1")
+	tk.MustQuery("select 1 in (select b from t2) from t1").Check(testkit.Rows("<nil>"))
+	tk.MustQuery("select 1 not in (select b from t2) from t1").Check(testkit.Rows("<nil>"))
+	tk.MustQuery("select 1 not in (select 1 from t2) from t1").Check(testkit.Rows("0"))
+	tk.MustQuery("select 1 in (select 1 from t2) from t1").Check(testkit.Rows("1"))
+	tk.MustQuery("select 1 not in (select null from t1) from t2").Check(testkit.Rows("<nil>"))
+	tk.MustQuery("select 1 in (select null from t1) from t2").Check(testkit.Rows("<nil>"))
+
+	tk.MustExec("drop table if exists s")
+	tk.MustExec("create table s(a int not null, b int)")
+	tk.MustExec("set sql_mode = ''")
+	tk.MustQuery("select (2,0) in (select s.a, min(s.b) from s) as f").Check(testkit.Rows("<nil>"))
+	tk.MustQuery("select (2,0) not in (select s.a, min(s.b) from s) as f").Check(testkit.Rows("<nil>"))
+	tk.MustQuery("select (2,0) = any (select s.a, min(s.b) from s) as f").Check(testkit.Rows("<nil>"))
+	tk.MustQuery("select (2,0) != all (select s.a, min(s.b) from s) as f").Check(testkit.Rows("<nil>"))
+	tk.MustQuery("select (2,0) in (select s.b, min(s.b) from s) as f").Check(testkit.Rows("<nil>"))
+	tk.MustQuery("select (2,0) not in (select s.b, min(s.b) from s) as f").Check(testkit.Rows("<nil>"))
+	tk.MustQuery("select (2,0) = any (select s.b, min(s.b) from s) as f").Check(testkit.Rows("<nil>"))
+	tk.MustQuery("select (2,0) != all (select s.b, min(s.b) from s) as f").Check(testkit.Rows("<nil>"))
+	tk.MustExec("insert into s values(1,null)")
+	tk.MustQuery("select 1 in (select b from s)").Check(testkit.Rows("<nil>"))
+
+	tk.MustExec("drop table if exists t")
+	tk.MustExec("create table t(a int)")
+	tk.MustExec("insert into t values(1),(null)")
+	tk.MustQuery("select a not in (select 1) from t").Sort().Check(testkit.Rows(
+		"0",
+		"<nil>",
+	))
+	tk.MustQuery("select 1 not in (select null from t t1) from t").Check(testkit.Rows(
+		"<nil>",
+		"<nil>",
+	))
+	tk.MustQuery("select 1 in (select null from t t1) from t").Check(testkit.Rows(
+		"<nil>",
+		"<nil>",
+	))
+	tk.MustQuery("select a in (select 0) xx from (select null as a) x").Check(testkit.Rows("<nil>"))
+
+	tk.MustExec("drop table t")
+	tk.MustExec("create table t(a int, b int)")
+	tk.MustExec("insert into t values(1,null),(null, null),(null, 2)")
+	tk.MustQuery("select * from t t1 where (2 in (select a from t t2 where (t2.b=t1.b) is null))").Check(testkit.Rows())
+	tk.MustQuery("select (t2.a in (select t1.a from t t1)) is true from t t2").Sort().Check(testkit.Rows(
+		"0",
+		"0",
+		"1",
+	))
+
 	tk.MustExec("set @@tidb_hash_join_concurrency=5")
 }
 

expression/aggregation/base_func_test.go

@@ -39,14 +39,3 @@ func (s *testBaseFuncSuite) TestClone(c *check.C) {
 	c.Assert(desc.Args[0], check.Equals, col)
 	c.Assert(desc.equal(s.ctx, cloned), check.IsFalse)
 }
-
-func (s *testBaseFuncSuite) TestMaxMin(c *check.C) {
-	col := &expression.Column{
-		UniqueID: 0,
-		RetType:  types.NewFieldType(mysql.TypeLonglong),
-	}
-	col.RetType.Flag |= mysql.NotNullFlag
-	desc, err := newBaseFuncDesc(s.ctx, ast.AggFuncMax, []expression.Expression{col})
-	c.Assert(err, check.IsNil)
-	c.Assert(mysql.HasNotNullFlag(desc.RetTp.Flag), check.IsFalse)
-}

expression/aggregation/descriptor.go

@@ -18,6 +18,7 @@ import (
 	"math"
 	"strconv"
 
+	"github.com/pingcap/errors"
 	"github.com/pingcap/parser/ast"
 	"github.com/pingcap/parser/model"
 	"github.com/pingcap/parser/mysql"
@@ -251,3 +252,43 @@ func (a *AggFuncDesc) evalNullValueInOuterJoin4BitOr(ctx sessionctx.Context, sch
 	}
 	return con.Value, true
 }
+
+// UpdateNotNullFlag4RetType checks if we should remove the NotNull flag for the return type of the agg.
+func (a *AggFuncDesc) UpdateNotNullFlag4RetType(hasGroupBy, allAggsFirstRow bool) error {
+	var removeNotNull bool
+	switch a.Name {
+	case ast.AggFuncCount, ast.AggFuncApproxCountDistinct, ast.AggFuncApproxPercentile,
+		ast.AggFuncBitAnd, ast.AggFuncBitOr, ast.AggFuncBitXor,
+		ast.WindowFuncFirstValue, ast.WindowFuncLastValue, ast.WindowFuncNthValue, ast.WindowFuncRowNumber,
+		ast.WindowFuncRank, ast.WindowFuncDenseRank, ast.WindowFuncCumeDist, ast.WindowFuncNtile, ast.WindowFuncPercentRank,
+		ast.WindowFuncLead, ast.WindowFuncLag, ast.AggFuncJsonObjectAgg,
+		ast.AggFuncVarSamp, ast.AggFuncVarPop, ast.AggFuncStddevPop, ast.AggFuncStddevSamp:
+		removeNotNull = false
+	case ast.AggFuncSum, ast.AggFuncAvg, ast.AggFuncGroupConcat:
+		if !hasGroupBy {
+			removeNotNull = true
+		}
+	// `select max(a) from empty_tbl` returns `null`, while `select max(a) from empty_tbl group by b` returns empty.
+	case ast.AggFuncMax, ast.AggFuncMin:
+		if !hasGroupBy && a.RetTp.Tp != mysql.TypeBit {
+			removeNotNull = true
+		}
+	// `select distinct a from empty_tbl` returns empty
+	// `select a from empty_tbl group by b` returns empty
+	// `select a, max(a) from empty_tbl` returns `(null, null)`
+	// `select a, max(a) from empty_tbl group by b` returns empty
+	// `select a, count(a) from empty_tbl` returns `(null, 0)`
+	// `select a, count(a) from empty_tbl group by b` returns empty
+	case ast.AggFuncFirstRow:
+		if !allAggsFirstRow && !hasGroupBy {
+			removeNotNull = true
+		}
+	default:
+		return errors.Errorf("unsupported agg function: %s", a.Name)
+	}
+	if removeNotNull {
+		a.RetTp = a.RetTp.Clone()
+		a.RetTp.Flag &^= mysql.NotNullFlag
+	}
+	return nil
+}

expression/expression.go

@@ -142,6 +142,30 @@ func IsEQCondFromIn(expr Expression) bool {
 	return len(cols) > 0
 }
 
+<<<<<<< HEAD
+=======
+// ExprNotNull checks if an expression is possible to be null.
+func ExprNotNull(expr Expression) bool {
+	if c, ok := expr.(*Constant); ok {
+		return !c.Value.IsNull()
+	}
+	// For ScalarFunction, the result would not be correct until we support maintaining
+	// NotNull flag for it.
+	return mysql.HasNotNullFlag(expr.GetType().Flag)
+}
+
+// HandleOverflowOnSelection handles Overflow errors when evaluating selection filters.
+// We should ignore overflow errors when evaluating selection conditions:
+//		INSERT INTO t VALUES ("999999999999999999");
+//		SELECT * FROM t WHERE v;
+func HandleOverflowOnSelection(sc *stmtctx.StatementContext, val int64, err error) (int64, error) {
+	if sc.InSelectStmt && err != nil && types.ErrOverflow.Equal(err) {
+		return -1, nil
+	}
+	return val, err
+}
+
+>>>>>>> 3ef3e54b5... planner: don't push down null sensitive join conditions (#19620)
 // EvalBool evaluates expression list to a boolean value. The first returned value
 // indicates bool result of the expression list, the second returned value indicates
 // whether the result of the expression list is null, it can only be true when the