[ARROW-12676] RecordBatchBuilder with uint dictionary creates signed int Batch

• Type: Bug
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Affects Version/s: 3.0.0
• Component/s: C++
• External issue URL: https://github.com/apache/arrow/issues/28424
• Assignee: Unassigned
• Reporter: Kyle Kavanagh
• Created: 07/May/21 03:11
• Updated: 11/Jan/23 08:28

Description

When a RecordBatchBuilder with a dictionary type w/ a uint32 index is flushed to a batch, the resulting batch contains a int32 index:

BatchBuilder schema after flush: 
Symbol: dictionary<values=string, indices=int16, ordered=0>
Status: dictionary<values=string, indices=uint32, ordered=0>

Batch schema after flush:
Symbol: dictionary<values=string, indices=int16, ordered=0>
Status: dictionary<values=string, indices=int32, ordered=0>

from:

std::shared_ptr<arrow::RecordBatch> batch;  
auto status = batchBuilder_>Flush(&batch);  
std::cout<<"BatchBuilder schema after flush: "<<batchBuilder_->schema()->ToString()<<std::endl;  
std::cout<<"Batch schema after flush: "<<batch->schema()->ToString()<<std::endl;  

if(!status.ok()) {    throw Exception("Arrow batch flush failed: {}", status);  }

This results in a failure to write: "Invalid: Tried to write record batch with different schema"

I believe this is related to https://issues.apache.org/jira/browse/ARROW-9969 and in particular, this bit: https://github.com/apache/arrow/blob/master/cpp/src/arrow/table_builder.cc#L72

Is the dictionary->Equals comparison checking the signed-ness of the indices?