[C++][Parquet] Fix undefined behavior in memcpy with nullptr for empty ByteArray
- Dominant language
- C++
- Stars
- 17.1k
- Forks
- 4.3k
- Avg merge
- 3d 13h
- Merged PRs (30d)
- 88
Description
### Describe the enhancement requested
As noted by @wgtmac in https://github.com/apache/arrow/pull/48717#discussion_r2665550214:
> Since it allows `ptr` to be nullptr for empty string, following `Copy` may be a UB if min is an empty string with nullptr.
>
> ```cpp
> template <>
> inline void TypedStatisticsImpl::Copy(const ByteArray& src, ByteArray* dst,
> ResizableBuffer* buffer) {
> if (dst->ptr == src.ptr) return;
> PARQUET_THROW_NOT_OK(buffer->Resize(src.len, false));
> std::memcpy(buffer->mutable_data(), src.ptr, src.len);
> *dst = ByteArray(src.len, buffer->data());
> }
> ```
When `ByteArray` has `len=0` and `ptr=nullptr` (the default-constructed state per `types.h:649`), calling `std::memcpy` with a nullptr source is undefined behavior according to the C++ standard, even when the size is 0.
This issue also exists in other places:
- `DictDecoderImpl::SetDict` in `decoder.cc:1061`
- Test utilities in `statistics_test.cc` and `column_writer_test.cc`
### Component(s)
C++, Parquet
Contributor guide
Assessment
This issue has not been assessed yet.