[ARROW-14805] [C++][Dataset] Support Count function without projections in ORC to avoid loading all columns

• Type: Sub-task
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Component/s: C++
• Labels: ORC c++ dataset good-second-issue
• External issue URL: https://github.com/apache/arrow/issues/30340
• Assignee: Unassigned
• Reporter: zhixingheyi-tian
• Created: 24/Nov/21 08:31
• Updated: 11/Jan/23 08:42

Description

For ORC support in dataset, when execute count query without projections, just like "select count from table", it will load all columns. Because orc lib code is that https://github.com/apache/orc/blob/22828f79a526069d9629719c9476b7addad91ae6/c%2B%2B/src/Reader.cc#L120-L144.

 

Arrow side can improve it like parquet in dataset.