From e45a92748f0a46142d2b5a936eb56c4d74d6c884 Mon Sep 17 00:00:00 2001
From: Rossi Sun <zanmato1984@gmail.com>
Date: Thu, 26 Mar 2026 14:28:07 -0700
Subject: [PATCH 1/2] GH-49392: [C++][Compute] Fix fixed-width gather byte
 offset overflow

---
 cpp/src/arrow/compute/kernels/gather_internal.h    | 14 ++++++++++----
 .../arrow/compute/kernels/vector_selection_test.cc | 13 +++++++++++++
 2 files changed, 23 insertions(+), 4 deletions(-)

diff --git a/cpp/src/arrow/compute/kernels/gather_internal.h b/cpp/src/arrow/compute/kernels/gather_internal.h
index 4c161533a727..de3872bf7c3f 100644
--- a/cpp/src/arrow/compute/kernels/gather_internal.h
+++ b/cpp/src/arrow/compute/kernels/gather_internal.h
@@ -35,6 +35,11 @@
 
 namespace arrow::internal {
 
+template <typename IndexType>
+constexpr int64_t FixedWidthByteOffset(IndexType index, int64_t value_width) {
+  return static_cast<int64_t>(index) * value_width;
+}
+
 // CRTP [1] base class for Gather that provides a gathering loop in terms of
 // Write*() methods that must be implemented by the derived class.
 //
@@ -185,8 +190,8 @@ class Gather : public GatherBaseCRTP<Gather<kValueWidthInBits, IndexCType, kWith
       memcpy(out_ + position * scaled_factor, src_ + idx_[position] * scaled_factor,
              scaled_factor);
     } else {
-      memcpy(out_ + position * kValueWidth, src_ + idx_[position] * kValueWidth,
-             kValueWidth);
+      memcpy(out_ + FixedWidthByteOffset(position, kValueWidth),
+             src_ + FixedWidthByteOffset(idx_[position], kValueWidth), kValueWidth);
     }
   }
 
@@ -195,7 +200,7 @@ class Gather : public GatherBaseCRTP<Gather<kValueWidthInBits, IndexCType, kWith
       const int64_t scaled_factor = kValueWidth * factor_;
       memset(out_ + position * scaled_factor, 0, scaled_factor);
     } else {
-      memset(out_ + position * kValueWidth, 0, kValueWidth);
+      memset(out_ + FixedWidthByteOffset(position, kValueWidth), 0, kValueWidth);
     }
   }
 
@@ -204,7 +209,8 @@ class Gather : public GatherBaseCRTP<Gather<kValueWidthInBits, IndexCType, kWith
       const int64_t scaled_factor = kValueWidth * factor_;
       memset(out_ + position * scaled_factor, 0, length * scaled_factor);
     } else {
-      memset(out_ + position * kValueWidth, 0, length * kValueWidth);
+      memset(out_ + FixedWidthByteOffset(position, kValueWidth), 0,
+             FixedWidthByteOffset(length, kValueWidth));
     }
   }
 
diff --git a/cpp/src/arrow/compute/kernels/vector_selection_test.cc b/cpp/src/arrow/compute/kernels/vector_selection_test.cc
index 5fa2d6824dc1..edac6e431da9 100644
--- a/cpp/src/arrow/compute/kernels/vector_selection_test.cc
+++ b/cpp/src/arrow/compute/kernels/vector_selection_test.cc
@@ -27,6 +27,7 @@
 #include "arrow/array/concatenate.h"
 #include "arrow/chunked_array.h"
 #include "arrow/compute/api.h"
+#include "arrow/compute/kernels/gather_internal.h"
 #include "arrow/compute/kernels/test_util_internal.h"
 #include "arrow/scalar.h"
 #include "arrow/table.h"
@@ -104,6 +105,18 @@ void CheckTakeIndicesCase(const std::string& filter_json, const std::string& ind
 
 }  // namespace
 
+// GH-49392: Fixed-width gather byte offsets must be computed in int64_t.
+TEST(GatherInternal, FixedWidthByteOffsetUses64BitArithmetic) {
+  constexpr int64_t kByteWidth = 8;
+  constexpr uint32_t kIndex = std::numeric_limits<uint32_t>::max() / kByteWidth + 1;
+  constexpr int64_t kExpectedByteOffset =
+      static_cast<int64_t>(std::numeric_limits<uint32_t>::max()) + 1;
+
+  const int64_t byte_offset = ::arrow::internal::FixedWidthByteOffset(kIndex, kByteWidth);
+  ASSERT_EQ(byte_offset, kExpectedByteOffset);
+  ASSERT_GT(byte_offset, std::numeric_limits<uint32_t>::max());
+}
+
 // ----------------------------------------------------------------------
 
 TEST(GetTakeIndices, Basics) {

From c2ab6771adb403233a6e59bbd0dc54a5cbf6c816 Mon Sep 17 00:00:00 2001
From: Rossi Sun <zanmato1984@gmail.com>
Date: Mon, 30 Mar 2026 11:27:54 -0700
Subject: [PATCH 2/2] GH-49392: Remove helper-level gather test

---
 .../arrow/compute/kernels/vector_selection_test.cc  | 13 -------------
 1 file changed, 13 deletions(-)

diff --git a/cpp/src/arrow/compute/kernels/vector_selection_test.cc b/cpp/src/arrow/compute/kernels/vector_selection_test.cc
index edac6e431da9..5fa2d6824dc1 100644
--- a/cpp/src/arrow/compute/kernels/vector_selection_test.cc
+++ b/cpp/src/arrow/compute/kernels/vector_selection_test.cc
@@ -27,7 +27,6 @@
 #include "arrow/array/concatenate.h"
 #include "arrow/chunked_array.h"
 #include "arrow/compute/api.h"
-#include "arrow/compute/kernels/gather_internal.h"
 #include "arrow/compute/kernels/test_util_internal.h"
 #include "arrow/scalar.h"
 #include "arrow/table.h"
@@ -105,18 +104,6 @@ void CheckTakeIndicesCase(const std::string& filter_json, const std::string& ind
 
 }  // namespace
 
-// GH-49392: Fixed-width gather byte offsets must be computed in int64_t.
-TEST(GatherInternal, FixedWidthByteOffsetUses64BitArithmetic) {
-  constexpr int64_t kByteWidth = 8;
-  constexpr uint32_t kIndex = std::numeric_limits<uint32_t>::max() / kByteWidth + 1;
-  constexpr int64_t kExpectedByteOffset =
-      static_cast<int64_t>(std::numeric_limits<uint32_t>::max()) + 1;
-
-  const int64_t byte_offset = ::arrow::internal::FixedWidthByteOffset(kIndex, kByteWidth);
-  ASSERT_EQ(byte_offset, kExpectedByteOffset);
-  ASSERT_GT(byte_offset, std::numeric_limits<uint32_t>::max());
-}
-
 // ----------------------------------------------------------------------
 
 TEST(GetTakeIndices, Basics) {