[ORC-1143] [C++] Support reading the PRESENT stream without reading the column data

• Type: New Feature
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Component/s: C++
• Assignee: Unassigned
• Reporter: Quanlong Huang
• Created: 29/Mar/22 02:43
• Updated: 29/Mar/22 02:53

Description

Queries like "select count(a) from tbl" just requires checking whether the column value is not NULL. ORC files already have the PRESENT stream for each column (though it's optional). We can serve the request by just reading the PRESENT stream.

Currently, ReadIntent has two items:

enum ReadIntent {
  ReadIntent_ALL = 0,

  // Only read the offsets of selected type. Do not read the children types.
  ReadIntent_OFFSETS = 1
};

We can extend it to add an item like ReadIntent_PRESENT. The corresponding ColumnVectorBatch will only have valid notNull results.

This would help more on string columns. E.g. checking how many customers have email address

select count(email_address) from tpcds.customer 

Issue Links

• relates to: ORC-450 [C++] Support selecting list indices without materializing list items