Module position_delete 

Shared Iceberg position-delete (Parquet and Puffin deletion vector) helpers.

Structs

CachedPuffinReader 🔒

PositionDeleteFileNameGenerators

File-name generators shared by all Iceberg position-delete writers.

PositionDeleteReader

Shared decoder with at most one cached Puffin reader/footer. Callers may reuse it across sequential tasks for the same table/FileIO. Decoded bitmaps are never cached by path. Puffin paths must identify immutable files; drop the reader to release its cache.

PositionDeleteValidation 🔒

Checks required at the source boundary, but not by trusted sink read-modify-write calls.

Constants

DELETION_VECTOR_PROPERTY_CARDINALITY 🔒

Puffin blob property for deletion vector cardinality.

DELETION_VECTOR_PROPERTY_REFERENCED_DATA_FILE 🔒

Puffin blob property for referenced data file path.

POSITION_DELETE_WRITE_CHUNK_SIZE 🔒

How many positions to buffer before flushing one (file_path, pos) batch to the writer.

Functions

read_dv_positions_from_data_file

Reads the deletion-vector positions of a single Puffin DV DataFile.

read_parquet_position_deletes_from_file

Reads the positions stored in a V2 Parquet position-delete file into a [DeleteVector].

read_position_deletes_from_file

Reads the deleted positions of a single position-delete DataFile regardless of on-disk format,

validate_position_delete_descriptor 🔒

Metadata-only source checks, shared by planning (including compaction) and replay reads. Returns the validated blob range for Puffin, or None for Parquet.

write_dv_puffin_file

Writes delete_vector as a single Puffin deletion-vector blob referencing data_file_path, and returns its [DataFile] metadata (content PositionDeletes, format Puffin) with referenced_data_file set.

write_parquet_position_delete_file

Writes delete_vector as a single file-scoped Parquet position-delete file referencing data_file_path, and returns its [DataFile] metadata (content PositionDeletes, format Parquet) with referenced_data_file set.

write_position_delete_chunk 🔒

Writes one chunk of positions as a (file_path, pos) batch into writer. Every row shares data_file_path because the delete file is file-scoped.

write_position_delete_file

Write one file-scoped position-delete artifact using the table’s configured on-disk format.