Merge pull request #3135 from dolthub/angela/antijoin

include null values when evaluating anti joins created from `WHERE NOT EXISTS`

Author: angelamayxie

enginetest/join_op_tests.go

@@ -1921,6 +1921,18 @@ SELECT SUM(x) FROM xy WHERE x IN (
 			},
 		},
 	},
+	{
+		name: "where not exists",
+		setup: [][]string{
+			setup.XyData[0],
+		},
+		tests: []JoinOpTests{
+			{
+				Query:    `select * from xy_hasnull x where not exists(select 1 from ab_hasnull a where a.b = x.y)`,
+				Expected: []sql.Row{{1, 0}, {3, nil}},
+			},
+		},
+	},
 	{
 		name: "multi-column merge join",
 		setup: [][]string{

enginetest/queries/imdb_plans.go

@@ -561,7 +561,11 @@ func convertAntiToLeftJoin(m *memo.Memo) error {
 		rightGrp := m.MemoizeProject(nil, anti.Right, projectExpressions)
 
 		// join is a new group
-		joinGrp := m.MemoizeLeftJoin(nil, anti.Left, rightGrp, plan.JoinTypeLeftOuterExcludeNulls, anti.Filter)
+		joinType := plan.JoinTypeLeftOuter
+		if anti.Op.IsExcludeNulls() {
+			joinType = plan.JoinTypeLeftOuterExcludeNulls
+		}
+		joinGrp := m.MemoizeLeftJoin(nil, anti.Left, rightGrp, joinType, anti.Filter)
 
 		// drop null projected columns on right table
 		nullFilters := make([]sql.Expression, len(nullify))

enginetest/queries/integration_plans.go

@@ -190,7 +190,7 @@ func unnestExistSubqueries(ctx *sql.Context, scope *plan.Scope, a *Analyzer, fil
 			switch joinType {
 			case plan.JoinTypeAnti:
 				cond := expression.NewLiteral(true, types.Boolean)
-				ret = plan.NewAntiJoin(ret, s.inner, cond).WithComment(comment)
+				ret = plan.NewAntiJoinIncludingNulls(ret, s.inner, cond).WithComment(comment)
 				qFlags.Set(sql.QFlagInnerJoin)
 			case plan.JoinTypeSemi:
 				ret = plan.NewCrossJoin(ret, s.inner).WithComment(comment)
@@ -209,7 +209,7 @@ func unnestExistSubqueries(ctx *sql.Context, scope *plan.Scope, a *Analyzer, fil
 
 		switch joinType {
 		case plan.JoinTypeAnti:
-			ret = plan.NewAntiJoin(ret, s.inner, expression.JoinAnd(outerFilters...)).WithComment(comment)
+			ret = plan.NewAntiJoinIncludingNulls(ret, s.inner, expression.JoinAnd(outerFilters...)).WithComment(comment)
 			qFlags.Set(sql.QFlagInnerJoin)
 		case plan.JoinTypeSemi:
 			ret = plan.NewSemiJoin(ret, s.inner, expression.JoinAnd(outerFilters...)).WithComment(comment)

enginetest/queries/query_plans.go

@@ -811,7 +811,7 @@ func (j *joinOrderBuilder) constructJoin(
 		rel = &LeftJoin{b}
 	case plan.JoinTypeSemi:
 		rel = &SemiJoin{b}
-	case plan.JoinTypeAnti:
+	case plan.JoinTypeAnti, plan.JoinTypeAntiIncludeNulls:
 		rel = &AntiJoin{b}
 	case plan.JoinTypeLateralInner, plan.JoinTypeLateralCross,
 		plan.JoinTypeLateralRight, plan.JoinTypeLateralLeft:
@@ -1428,7 +1428,7 @@ func getOpIdx(e *edge) int {
 		return 1
 	case plan.JoinTypeSemi:
 		return 2
-	case plan.JoinTypeAnti:
+	case plan.JoinTypeAnti, plan.JoinTypeAntiIncludeNulls:
 		return 3
 	case plan.JoinTypeLeftOuter:
 		return 4