commit ab084b507a70d4da16ef1dc7cfd0fec186083761
author Yuriy Chernyshov <thegeorg@yandex-team.com> Thu Jan 16 15:55:45 2025 +0800
committer Gang Wu <ustcwg@gmail.com> Thu Jan 16 15:55:45 2025 +0800
tree f0c20ac288a311e5691facce55473f9f97d8e1c3
parent 43ebc98b0fd8653754e60e93276ad9ccc6a93c36

ORC-1834: [C++] Fix undefined behavior

### What changes were proposed in this pull request?
Unaligned reads are UB in C++, memcpy-ing zero bytes is UB either.

### How was this patch tested?
Internal UBsan report was used to detect and fix this bug.

Closes #2112 from georgthegreat/patch-3.

Authored-by: Yuriy Chernyshov <thegeorg@yandex-team.com>
Signed-off-by: Gang Wu <ustcwg@gmail.com>

c++/src/ColumnReader.cc

diff --git a/c++/src/ColumnReader.cc b/c++/src/ColumnReader.cc
index e70f916..af434c3 100644
--- a/c++/src/ColumnReader.cc
+++ b/c++/src/ColumnReader.cc
@@ -395,7 +395,7 @@
       int64_t bits = 0;
       if (bufferEnd_ - bufferPointer_ >= 8) {
         if (isLittleEndian) {
-          bits = *(reinterpret_cast<const int64_t*>(bufferPointer_));
+          memcpy(&bits, bufferPointer_, sizeof(bits));
         } else {
           bits = static_cast<int64_t>(static_cast<unsigned char>(bufferPointer_[0]));
           bits |= static_cast<int64_t>(static_cast<unsigned char>(bufferPointer_[1])) << 8;
@@ -509,8 +509,10 @@
           bufferNum = std::min(numValues,
                                static_cast<size_t>(bufferEnd_ - bufferPointer_) / bytesPerValue_);
           uint64_t bufferBytes = bufferNum * bytesPerValue_;
-          memcpy(outArray, bufferPointer_, bufferBytes);
-          bufferPointer_ += bufferBytes;
+          if (bufferBytes > 0) {
+            memcpy(outArray, bufferPointer_, bufferBytes);
+            bufferPointer_ += bufferBytes;
+          }
         }
         for (size_t i = bufferNum; i < numValues; ++i) {
           outArray[i] = readDouble<ValueType>();