Skip to main content

Crate purrdf_columnar

Crate purrdf_columnar 

Source
Expand description

PurRDF’s first-party, byte-deterministic Parquet bridge.

The codec projects an RDF 1.2 purrdf_core::DatasetView plus a purrdf_core::ContentStore into the fixed five-table schema exposed by schema. Its deliberately narrow Parquet subset keeps the implementation wasm-clean and auditable: INT64/BYTE_ARRAY columns, flat OPTIONAL fields, PLAIN values, RLE definition levels, Data Page V2, and UNCOMPRESSED or ZSTD page bodies.

No vocabulary is built into this crate. Every IRI is carried from the caller’s dataset verbatim.

Re-exports§

pub use error::CanonicalReason;
pub use error::ColumnarError;
pub use error::ColumnarSinkError;
pub use error::ColumnarWriteError;
pub use error::CompactReason;
pub use error::ConstructionReason;
pub use error::EncodingReason;
pub use error::MetadataReason;
pub use error::PageBodyReason;
pub use error::ProjectionReason;
pub use error::ReaderReason;
pub use error::ValueBodyReason;
pub use error::ZstdReason;
pub use schema::ColumnSchema;
pub use schema::PhysicalType;
pub use schema::Repetition;
pub use schema::Table;
pub use schema::TableSchema;

Modules§

error
Typed failures from the narrow columnar codec.
schema
The closed five-table columnar schema.

Structs§

BloomOptions
Standard split-block Bloom filter sizing for each admitted column chunk. Filters are opt-in and never affect values or exact native reconstruction.
ColumnOptions
An explicit physical override for one schema column.
ColumnarOrdinalRead
Requested provenance bound to the actual dataset of a complete strict read. No public constructor accepts caller-assembled rows or a replacement dataset.
ColumnarProjection
A validated canonical five-table projection with its own dense term ids.
ColumnarRead
The result of a complete Parquet-to-RDF conversion.
ColumnarWrite
The result of a complete RDF-to-Parquet conversion.
ParquetFiles
Exactly one Parquet file for each member of Table::ALL.
ProjectionRows
A borrowed traversal of a canonical table’s rows.
ProjectionTermGraph
An exact rejected term graph retained without expanding shared children.
TableWriteReport
Physical output counts for a table in crate::Table::ALL order.
WriteOptions
Runtime physical layout choices; these do not change the RDF projection.
WriteReport
Result of a complete five-table sink export.

Enums§

ColumnarOrdinal
One original zero-based row in a Columnar RDF table.
ColumnarOrdinalValue
Original row fields translated by the strict reader’s dictionary-to-kernel map. The IDs belong only to the dataset held by the accompanying read evidence.
Compression
Page compression selected for one deterministic five-table write.
EncodingPolicy
Physical encoding candidates admitted for one column chunk.
ProjectionCell
A cell in the native v1 five-table projection.

Traits§

BlobSource
A verified content identity with a known-length, replayable payload source. Source reads must remain stable until the export finishes.

Functions§

project
Project an RDF view into the authoritative five-table schema.
read
Read PurRDF’s exact five-table Parquet profile back into RDF and blob values.
read_with_ordinals
Complete the same strict read while retaining only requested original rows.
source_manifest
Describe all five actual files in canonical schema order.
write
Project any RDF 1.2 DatasetView and ContentStore into five Parquet files.
write_to
Emit each of the five files in Table::ALL order without retaining a complete output file. Each callback must accept its whole chunk or fail. For ordinary Rust writers, the callback can select the table’s writer and use std::io::Write::write_all, or the existing Core WriterDrain adapter.
write_with_blob_sources_to
Write native uncompressed files with window-bounded blob payload storage.
write_with_options
Export into resident file buffers through the same incremental writer as write_to. The complete returned files are additional output storage.