Fix find_first Presto function on empty arrays and invalid start index (

#8030)

Summary:
Pull Request resolved: #8030

find_first(<empty array>,  0, <lambda>) should throw.

Fixes #8020

Reviewed By: xiaoxmeng

Differential Revision: D52140846

fbshipit-source-id: 8305ea28e1c874d07daf458c98b9855f87dd42cf

velox/functions/prestosql/FindFirst.cpp

@@ -20,6 +20,14 @@
 namespace facebook::velox::functions {
 namespace {
 
+void recordInvalidStartIndex(vector_size_t row, exec::EvalCtx& context) {
+  try {
+    VELOX_USER_FAIL("SQL array indices start at 1. Got 0.");
+  } catch (const VeloxUserError& exception) {
+    context.setVeloxExceptionError(row, std::current_exception());
+  }
+}
+
 class FindFirstFunctionBase : public exec::VectorFunction {
  public:
   bool isDefaultNullBehavior() const override {
@@ -175,11 +183,7 @@ class FindFirstFunctionBase : public exec::VectorFunction {
             rawAdjustedOffsets[row] = 0;
             rawAdjustedSizes[row] = 0;
 
-            try {
-              VELOX_USER_FAIL("SQL array indices start at 1. Got 0.");
-            } catch (const VeloxUserError& exception) {
-              context.setVeloxExceptionError(row, std::current_exception());
-            }
+            recordInvalidStartIndex(row, context);
           } else if (start > 0) {
             rawAdjustedOffsets[row] = offset + (start - 1);
             rawAdjustedSizes[row] = size - (start - 1);
@@ -271,17 +275,34 @@ class FindFirstFunction : public FindFirstFunctionBase {
       exec::EvalCtx& context,
       VectorPtr& result) const override {
     auto flatArray = prepareInputArray(args[0], rows, context);
+    auto startIndexVector = (args.size() == 3 ? args[1] : nullptr);
 
     if (flatArray->elements()->size() == 0) {
       // All arrays are NULL or empty.
+
+      if (startIndexVector != nullptr) {
+        // Check start indices for zeros.
+
+        exec::LocalDecodedVector startIndexDecoder(
+            context, *startIndexVector, rows);
+        rows.applyToSelected([&](auto row) {
+          if (flatArray->isNullAt(row) || startIndexDecoder->isNullAt(row)) {
+            return;
+          }
+          const auto start = startIndexDecoder->valueAt<int64_t>(row);
+          if (start == 0) {
+            recordInvalidStartIndex(row, context);
+          }
+        });
+      }
+
       auto localResult = BaseVector::createNullConstant(
           outputType, rows.end(), context.pool());
       context.moveOrCopyResult(localResult, rows, result);
       return;
     }
 
     auto* predicateVector = args.back()->asUnchecked<FunctionVector>();
-    auto startIndexVector = (args.size() == 3 ? args[1] : nullptr);
 
     // Collect indices of the first matching elements or NULLs if no match or
     // error.

velox/functions/prestosql/tests/FindFirstTest.cpp

@@ -194,6 +194,27 @@ TEST_F(FindFirstTest, invalidIndex) {
   auto expected = makeNullableFlatVector<int32_t>(
       {2, std::nullopt, std::nullopt, std::nullopt});
   verify("find_first(c0, c1, x -> (x > 0))", data, expected);
+
+  // All null or empty arrays.
+  data = makeRowVector({
+      makeArrayVectorFromJson<int32_t>({
+          "[]",
+          "null",
+          "[]",
+          "null",
+      }),
+      makeFlatVector<int32_t>({2, 0, 0, 1}),
+  });
+
+  // Index 0 is not valid. Expect an error in the 3rd row.
+  VELOX_ASSERT_THROW(
+      evaluate("find_first(c0, c1, x -> (x > 0))", data),
+      "SQL array indices start at 1. Got 0.");
+
+  // Mark 3rd row null. Expect no error.
+  data->setNull(2, true);
+  expected = makeAllNullFlatVector<int32_t>(4);
+  verify("find_first(c0, c1, x -> (x > 0))", data, expected);
 }
 
 // Verify that null arrays with non-zero offsets/sizes are processed correctly.