From 6dc74fb5ccf0c6ef6538e60835667b57b6d4f060 Mon Sep 17 00:00:00 2001
From: ltrk2 <107155950+ltrk2@users.noreply.github.com>
Date: Thu, 3 Aug 2023 12:38:43 -0700
Subject: [PATCH] Implement big-endian support for transform

---
 .../Serializations/SerializationArray.cpp          |  4 ++--
 .../Serializations/SerializationNumber.cpp         |  8 ++++----
 src/Functions/transform.cpp                        | 14 ++++++++------
 3 files changed, 14 insertions(+), 12 deletions(-)

diff --git a/src/DataTypes/Serializations/SerializationArray.cpp b/src/DataTypes/Serializations/SerializationArray.cpp
index cedcca870dd..e01c1aea0e9 100644
--- a/src/DataTypes/Serializations/SerializationArray.cpp
+++ b/src/DataTypes/Serializations/SerializationArray.cpp
@@ -129,7 +129,7 @@ namespace
         for (size_t i = offset; i < end; ++i)
         {
             ColumnArray::Offset current_offset = offset_values[i];
-            writeIntBinary(current_offset - prev_offset, ostr);
+            writeBinaryLittleEndian(current_offset - prev_offset, ostr);
             prev_offset = current_offset;
         }
     }
@@ -145,7 +145,7 @@ namespace
         while (i < initial_size + limit && !istr.eof())
         {
             ColumnArray::Offset current_size = 0;
-            readIntBinary(current_size, istr);
+            readBinaryLittleEndian(current_size, istr);
 
             if (unlikely(current_size > MAX_ARRAY_SIZE))
                 throw Exception(ErrorCodes::TOO_LARGE_ARRAY_SIZE, "Array size is too large: {}", current_size);
diff --git a/src/DataTypes/Serializations/SerializationNumber.cpp b/src/DataTypes/Serializations/SerializationNumber.cpp
index 8cabaec753d..0294a1c8a67 100644
--- a/src/DataTypes/Serializations/SerializationNumber.cpp
+++ b/src/DataTypes/Serializations/SerializationNumber.cpp
@@ -106,28 +106,28 @@ void SerializationNumber<T>::serializeBinary(const Field & field, WriteBuffer &
 {
     /// ColumnVector<T>::ValueType is a narrower type. For example, UInt8, when the Field type is UInt64
     typename ColumnVector<T>::ValueType x = static_cast<typename ColumnVector<T>::ValueType>(field.get<FieldType>());
-    writeBinary(x, ostr);
+    writeBinaryLittleEndian(x, ostr);
 }
 
 template <typename T>
 void SerializationNumber<T>::deserializeBinary(Field & field, ReadBuffer & istr, const FormatSettings &) const
 {
     typename ColumnVector<T>::ValueType x;
-    readBinary(x, istr);
+    readBinaryLittleEndian(x, istr);
     field = NearestFieldType<FieldType>(x);
 }
 
 template <typename T>
 void SerializationNumber<T>::serializeBinary(const IColumn & column, size_t row_num, WriteBuffer & ostr, const FormatSettings &) const
 {
-    writeBinary(assert_cast<const ColumnVector<T> &>(column).getData()[row_num], ostr);
+    writeBinaryLittleEndian(assert_cast<const ColumnVector<T> &>(column).getData()[row_num], ostr);
 }
 
 template <typename T>
 void SerializationNumber<T>::deserializeBinary(IColumn & column, ReadBuffer & istr, const FormatSettings &) const
 {
     typename ColumnVector<T>::ValueType x;
-    readBinary(x, istr);
+    readBinaryLittleEndian(x, istr);
     assert_cast<ColumnVector<T> &>(column).getData().push_back(x);
 }
 
diff --git a/src/Functions/transform.cpp b/src/Functions/transform.cpp
index e03701327b1..16326dd5a44 100644
--- a/src/Functions/transform.cpp
+++ b/src/Functions/transform.cpp
@@ -764,9 +764,8 @@ namespace
             }
 
             /// Note: Doesn't check the duplicates in the `from` array.
-
-            WhichDataType which(from_type);
-            if (isNativeNumber(which) || which.isDecimal32() || which.isDecimal64())
+            /// Field may be of Float type, but for the purpose of bitwise equality we can treat them as UInt64
+            if (WhichDataType which(from_type); isNativeNumber(which) || which.isDecimal32() || which.isDecimal64())
             {
                 cache.table_num_to_idx = std::make_unique<Cache::NumToIdx>();
                 auto & table = *cache.table_num_to_idx;
@@ -774,10 +773,13 @@ namespace
                 {
                     if (applyVisitor(FieldVisitorAccurateEquals(), (*cache.from_column)[i], (*from_column_uncasted)[i]))
                     {
-                        /// Field may be of Float type, but for the purpose of bitwise equality we can treat them as UInt64
-                        StringRef ref = cache.from_column->getDataAt(i);
                         UInt64 key = 0;
-                        memcpy(&key, ref.data, ref.size);
+                        auto * dst = reinterpret_cast<char *>(&key);
+                        const auto ref = cache.from_column->getDataAt(i);
+                        if constexpr (std::endian::native == std::endian::big)
+                            dst += sizeof(key) - ref.size;
+
+                        memcpy(dst, ref.data, ref.size);
                         table[key] = i;
                     }
                 }