feat: Allow duplicate output names in root ProjectNode

velox/core/PlanConsistencyChecker.cpp

@@ -29,7 +29,9 @@
 
 class Checker : public PlanNodeVisitor {
  public:
+  explicit Checker(PlanNodeId rootId) : rootId_{std::move(rootId)} {}
+
   void visit(const AggregationNode& node, PlanNodeVisitorContext& ctx)
       const override {
     const auto& rowType = node.sources().at(0)->outputType();
     for (const auto& expr : node.groupingKeys()) {
@@ -200,7 +202,11 @@
       checkInputs(expr, rowType);
     }
 
-    verifyOutputNames(node);
+    // The root ProjectNode may have empty or duplicate output names when used
+    // to apply user-specified column aliases (e.g. SELECT 1 as x, 2 as x).
+    if (node.id() != rootId_) {
+      verifyOutputNames(node);
+    }
 
     visitSources(&node, ctx);
   }
@@ -333,13 +339,18 @@
       checkInputs(input, rowType);
     }
   }
+
+  // ID of the root node. Used to skip output name validation on the root
+  // ProjectNode, which may have user-specified aliases that are empty or
+  // duplicated.
+  const PlanNodeId rootId_;
 };
 } // namespace
 
 void PlanConsistencyChecker::check(const core::PlanNodePtr& plan) {
   ExceptionContextSetter exceptionContext({planNodeMessage, (void*)plan.get()});
   PlanNodeVisitorContext ctx;
-  Checker checker;
+  Checker checker{plan->id()};
   plan->accept(checker, ctx);
 }
 }; // namespace facebook::velox::core

velox/core/tests/PlanConsistencyCheckerTest.cpp

@@ -102,32 +102,49 @@
   auto valuesNode =
       std::make_shared<ValuesNode>(nextId(), std::vector<RowVectorPtr>{});
 
-  auto projectNode = std::make_shared<ProjectNode>(
-      nextId(),
-      std::vector<std::string>{"a", "b", "c"},
-      std::vector<TypedExprPtr>{Lit(true), Lit(1), Lit(0.1)},
-      valuesNode);
-  ASSERT_NO_THROW(PlanConsistencyChecker::check(projectNode));
+  {
+    auto projectNode = std::make_shared<ProjectNode>(
+        nextId(),
+        std::vector<std::string>{"a", "b", "c"},
+        std::vector<TypedExprPtr>{Lit(true), Lit(1), Lit(0.1)},
+        valuesNode);
+    ASSERT_NO_THROW(PlanConsistencyChecker::check(projectNode));
+  }
 
-  // Duplicate output name.
-  projectNode = std::make_shared<ProjectNode>(
-      nextId(),
-      std::vector<std::string>{"a", "a", "c"},
-      std::vector<TypedExprPtr>{Lit(true), Lit(1), Lit(0.1)},
-      valuesNode);
+  {
+    // Duplicate output name in the root ProjectNode is allowed. This is used to
+    // apply user-specified column aliases (e.g. SELECT 1 AS x, 2 AS x).
+    auto projectNode = std::make_shared<ProjectNode>(
+        nextId(),
+        std::vector<std::string>{"a", "a", "c"},
+        std::vector<TypedExprPtr>{Lit(true), Lit(1), Lit(0.1)},
+        valuesNode);
+    ASSERT_NO_THROW(PlanConsistencyChecker::check(projectNode));
 
-  VELOX_ASSERT_THROW(
-      PlanConsistencyChecker::check(projectNode), "Duplicate output column: a");
+    // Duplicate output name in a non-root ProjectNode is not allowed.
+    auto outputProject = std::make_shared<ProjectNode>(
+        nextId(),
+        std::vector<std::string>{"x", "y", "z"},
+        std::vector<TypedExprPtr>{
+            Col(BOOLEAN(), "a"), Col(BOOLEAN(), "a"), Col(DOUBLE(), "c")},
+        projectNode);
+
+    VELOX_ASSERT_THROW(
+        PlanConsistencyChecker::check(outputProject),
+        "Duplicate output column: a");
+  }
 
   // Wrong column name.
-  projectNode = std::make_shared<ProjectNode>(
-      nextId(),
-      std::vector<std::string>{"a", "a", "c"},
-      std::vector<TypedExprPtr>{Lit(true), Col(REAL(), "x"), Lit(0.1)},
-      valuesNode);
+  {
+    auto projectNode = std::make_shared<ProjectNode>(
+        nextId(),
+        std::vector<std::string>{"a", "a", "c"},
+        std::vector<TypedExprPtr>{Lit(true), Col(REAL(), "x"), Lit(0.1)},
+        valuesNode);
 
-  VELOX_ASSERT_THROW(
-      PlanConsistencyChecker::check(projectNode), "Field not found: x");
+    VELOX_ASSERT_THROW(
+        PlanConsistencyChecker::check(projectNode), "Field not found: x");
+  }
 }
 
 TEST_F(PlanConsistencyCheckerTest, aggregation) {