add test and fix 9797

Author: elianddb

enginetest/queries/script_queries.go

@@ -11780,6 +11780,26 @@ select * from t1 except (
 			},
 		},
 	},
+	{
+		// https://github.com/dolthub/dolt/issues/9797
+		Name: "EXISTS subquery duplicate rows issue",
+		SetUpScript: []string{
+			"CREATE TABLE t(c0 INT, c1 INT, PRIMARY KEY(c0, c1));",
+			"INSERT INTO t VALUES (1, 1);",
+			"INSERT INTO t VALUES (2, 2);",
+			"INSERT INTO t VALUES (2, 3);",
+		},
+		Assertions: []ScriptTestAssertion{
+			{
+				Query: "SELECT * FROM t WHERE EXISTS (SELECT 1 FROM t AS x WHERE x.c0 = t.c0);",
+				Expected: []sql.Row{
+					{1, 1},
+					{2, 2},
+					{2, 3},
+				},
+			},
+		},
+	},
 }
 
 var SpatialScriptTests = []ScriptTest{

sql/analyzer/indexed_joins.go

@@ -467,12 +467,11 @@ func convertSemiToInnerJoin(m *memo.Memo) error {
 
 		// project is a new group
 		rightGrp := m.MemoizeProject(nil, semi.Right, projectExpressions)
-		if _, ok := semi.Right.First.(*memo.Distinct); !ok {
-			rightGrp.RelProps.Distinct = memo.HashDistinctOp
-		}
 
 		// join and its commute are a new group
 		joinGrp := m.MemoizeInnerJoin(nil, semi.Left, rightGrp, plan.JoinTypeInner, semi.Filter)
+		// Ensure DISTINCT operation to prevent duplicate rows in SemiJoin to InnerJoin conversion
+		joinGrp.RelProps.Distinct = memo.HashDistinctOp
 		// TODO: can't commute if right SubqueryAlias references outside scope (OuterScopeVisibility/IsLateral)
 		m.MemoizeInnerJoin(joinGrp, rightGrp, semi.Left, plan.JoinTypeInner, semi.Filter)
 

sql/analyzer/unnest_exists_subqueries.go

@@ -193,8 +193,12 @@ func unnestExistSubqueries(ctx *sql.Context, scope *plan.Scope, a *Analyzer, fil
 				ret = plan.NewAntiJoinIncludingNulls(ret, s.inner, cond).WithComment(comment)
 				qFlags.Set(sql.QFlagInnerJoin)
 			case plan.JoinTypeSemi:
-				ret = plan.NewCrossJoin(ret, s.inner).WithComment(comment)
-				qFlags.Set(sql.QFlagCrossJoin)
+				// Use SemiJoin with TRUE condition instead of CrossJoin to preserve EXISTS semantics
+				// CrossJoin would emit each outer row once per inner row, but EXISTS should emit
+				// each outer row at most once when there's at least one inner row
+				cond := expression.NewLiteral(true, types.Boolean)
+				ret = plan.NewSemiJoin(ret, s.inner, cond).WithComment(comment)
+				qFlags.Set(sql.QFlagInnerJoin)
 			default:
 				return filter, transform.SameTree, fmt.Errorf("hoistSelectExists failed on unexpected join type")
 			}

sql/rowexec/join_iters.go

@@ -185,6 +185,20 @@ func (i *joinIter) Next(ctx *sql.Context) (sql.Row, error) {
 		}
 
 		i.foundMatch = true
+		
+		// For semi joins, close secondary iterator and move to next primary row after first match
+		// This ensures each primary row is emitted at most once (semi join semantics)
+		if i.joinType.IsSemi() {
+			err = i.secondary.Close(ctx)
+			i.secondary = nil
+			if err != nil {
+				return nil, err
+			}
+			i.loadPrimaryRow = true
+			// For semi joins, return only the primary row, not the combined row
+			return i.removeParentRow(i.buildRow(primary, nil)), nil
+		}
+		
 		return i.removeParentRow(row), nil
 	}
 }