iceberg::PositionDeleteIndex Class Reference

Tracks deleted row positions using a bitmap. More...

#include <position_delete_index.h>

Public Member Functions
	PositionDeleteIndex (std::shared_ptr< DataFile > delete_file)
	PositionDeleteIndex (std::vector< std::shared_ptr< DataFile > > delete_files)
void	Delete (int64_t pos)
	Mark a position as deleted.
void	Delete (int64_t pos_start, int64_t pos_end)
	Mark a range of positions as deleted [pos_start, pos_end).
bool	IsDeleted (int64_t pos) const
	Check if a position is deleted.
bool	IsEmpty () const
	Check if the index is empty (no positions deleted).
int64_t	Cardinality () const
	Get the number of deleted positions.
void	Merge (const PositionDeleteIndex &other)
	Merge another index into this one.
const std::vector< std::shared_ptr< DataFile > > &	delete_files () const
	The delete files whose positions were merged into this index.
Result< std::vector< uint8_t > >	Serialize ()
	Serialize the index into a deletion-vector-v1 blob.

Static Public Member Functions
static Result< PositionDeleteIndex >	Deserialize (std::span< const uint8_t > blob, std::shared_ptr< DataFile > delete_file)
	Deserialize a deletion-vector-v1 blob into an index.

Friends
void ICEBERG_EXPORT	ForEachPositionDelete (std::span< const int64_t > positions, PositionDeleteIndex &target, std::vector< uint32_t > &scratch)
	Apply positions to target as deletes; semantically equivalent to calling target.Delete(pos) for each entry. Out-of-range positions are silently ignored. Sorted, mostly-contiguous input is fastest.

Detailed Description

Tracks deleted row positions using a bitmap.

This class provides a domain-specific API for position deletes in Iceberg MOR (merge-on-read) tables. Positions are 0-based row indices within a data file.

Member Function Documentation

Delete() [1/2]

void iceberg::PositionDeleteIndex::Delete

(

int64_t

pos

)

Mark a position as deleted.

Parameters

pos	The 0-based row position to delete

Delete() [2/2]

void iceberg::PositionDeleteIndex::Delete	(	int64_t	pos_start,
		int64_t	pos_end
	)

Mark a range of positions as deleted [pos_start, pos_end).

Parameters

pos_start	Start position (inclusive)
pos_end	End position (exclusive)

delete_files()

const std::vector< std::shared_ptr< DataFile > > & iceberg::PositionDeleteIndex::delete_files ( ) const

inline

The delete files whose positions were merged into this index.

Populated by constructors and Deserialize, and preserved across Merge. Callers use these to report the delete files that were rewritten when replacing them with a new deletion vector.

Deserialize()

static Result< PositionDeleteIndex > iceberg::PositionDeleteIndex::Deserialize ( std::span< const uint8_t >  blob, std::shared_ptr< DataFile >  delete_file  )

static

Deserialize a deletion-vector-v1 blob into an index.

Validates the blob framing (length prefix, magic sequence, CRC-32) and, against the source delete file, that the blob length matches content_size_in_bytes and the bitmap cardinality matches record_count. The source delete file is retained and exposed via delete_files().

IsDeleted()

bool iceberg::PositionDeleteIndex::IsDeleted

(

int64_t

pos

)

const

Check if a position is deleted.

Parameters

pos	The 0-based row position to check

Returns

true if the position is deleted, false otherwise

Merge()

void iceberg::PositionDeleteIndex::Merge

(

const PositionDeleteIndex &

other

)

Merge another index into this one.

Parameters

other

The index to merge (union operation)

Serialize()

Result< std::vector< uint8_t > > iceberg::PositionDeleteIndex::Serialize

(

)

Serialize the index into a deletion-vector-v1 blob.

The positions are run-length encoded, then framed per the Puffin spec: https://iceberg.apache.org/puffin-spec/#deletion-vector-v1-blob-type

Friends And Related Symbol Documentation

ForEachPositionDelete

void ICEBERG_EXPORT ForEachPositionDelete ( std::span< const int64_t >  positions, PositionDeleteIndex &  target, std::vector< uint32_t > &  scratch  )

friend

Apply positions to target as deletes; semantically equivalent to calling target.Delete(pos) for each entry. Out-of-range positions are silently ignored. Sorted, mostly-contiguous input is fastest.

Parameters

scratch

Caller-owned reusable buffer for the bulk dispatch path. Cleared and reused per key group; retain across calls to amortize allocations. Pass a distinct scratch per thread when calling concurrently with disjoint target.

---

The documentation for this class was generated from the following file:

• iceberg/deletes/position_delete_index.h