felipecrv commented on code in PR #35345:
URL: https://github.com/apache/arrow/pull/35345#discussion_r1389942849
##########
cpp/src/arrow/ipc/writer.cc:
##########
@@ -350,6 +350,67 @@ class RecordBatchSerializer {
return Status::OK();
}
+ template <typename ArrayType, typename offset_type = typename
ArrayType::offset_type>
+ Status GetZeroBasedListViewOffsets(const ArrayType& array,
+ std::shared_ptr<Buffer>*
out_value_offsets,
+ offset_type* out_min_offset,
+ offset_type* out_max_end) {
+ auto offsets = array.value_offsets();
+ auto sizes = array.value_sizes();
+
+ const int64_t required_bytes = sizeof(offset_type) * array.length();
+ if (array.offset() != 0) {
+ // If we have a non-zero offset, it's likely that the smallest offset is
+ // not zero. We must a) create a new offsets array with shifted offsets
and
+ // b) slice the values array accordingly.
+
+ ARROW_ASSIGN_OR_RAISE(auto shifted_offsets,
+ AllocateBuffer(required_bytes,
options_.memory_pool));
+ offset_type min_offset = 0;
+ offset_type max_end = 0;
+ if (array.length() > 0) {
+ min_offset = std::numeric_limits<offset_type>::max();
+ for (int i = 0; i < array.length(); ++i) {
+ min_offset = std::min(min_offset, array.value_offset(i));
+ max_end = std::max(max_end, array.value_offset(i) +
array.value_length(i));
+ }
+ }
+
+ auto* dest_offsets = shifted_offsets->mutable_data_as<offset_type>();
+
+ for (int i = 0; i < array.length(); ++i) {
+ dest_offsets[i] = array.value_offset(i) - min_offset;
+ }
+ *out_min_offset = min_offset;
+ *out_max_end = max_end;
+ offsets = std::move(shifted_offsets);
+ } else {
+ // ARROW-6046: Slice offsets to used extent, in case we have a truncated
+ // slice
+ if (offsets != nullptr && offsets->size() > required_bytes) {
+ offsets = SliceBuffer(offsets, 0, required_bytes);
+ }
+ *out_min_offset = 0;
+ *out_max_end = static_cast<offset_type>(array.values()->length());
+ }
+ *out_value_offsets = std::move(offsets);
+ return Status::OK();
+ }
+
+ template <typename ArrayType, typename offset_type = typename
ArrayType::offset_type>
+ Status GetListViewSizes(const ArrayType& array,
+ std::shared_ptr<Buffer>* out_value_sizes) {
+ const int64_t required_bytes = sizeof(offset_type) * array.length();
+ auto sizes = array.value_sizes();
+ if (sizes != nullptr && (array.offset() != 0 || sizes->size() >
required_bytes)) {
Review Comment:
Remember that in the beginning I was open to list-views having null buffers
just like empty lists can have empty buffers? This is the reason. And I think
keeping that check would be wise in case we revert that decision in the future.
This extra condition avoids an extra `shared_ptr` allocation. Cheap, but
something worth avoiding I guess. This was based on the code from
`GetZeroBasedValueOffsets` introduced in 7f6e6ab2.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
