Table of Contents

Public namespace Rowles.LeanCorpus.Codecs.Postings

Classes

Internal classInternal BlockPostingsWriter

Writes postings in packed block format (v3). Doc IDs and frequencies are written in 128-int delta-encoded packed blocks, with VarInt encoding for the tail (remaining < 128 values). Skip data is emitted after every block for efficient Advance().

Public class PackedIntCodec

Frame-of-Reference bit-packing codec for blocks of 128 integers. Packs values using the minimum number of bits needed for the largest value in the block.

Output format: [numBits : 1 byte][packed data : numBits × 16 bytes]. When numBits is 0 (all values are zero) the output is a single byte.

Internal classInternal PostingsReader

Reads delta-encoded postings lists written by PostingsWriter.

Internal classInternal PostingsWriter

Writes delta-encoded postings lists for a given term. Deltas are encoded as variable-length integers (VarInt/LEB128) for compactness.

Internal classInternal StreamingPostingsMerger

Streaming k-way merge of per-segment postings into a single merged segment. Iterates terms in sorted order across all source segments without ever materialising the full term-doc map in memory. Per-term position data is the only buffered state, bounded by the document frequency of one term.

Internal classInternal StreamingPostingsMerger.Source

One source segment for the merge. The DocIdMap maps source local doc IDs to merged doc IDs; entries containing -1 are dropped (deleted).

Structs

Public struct BlockPostingsEnum

Block-at-a-time postings iterator (v3 format). Reads packed blocks of 128 doc IDs written by BlockPostingsWriter. Only the current block is decoded, keeping memory at a constant ~1 KB (2 × 128 ints) regardless of postings list length.

Public struct PostingsEnum

Forward-only cursor over a postings list. Decodes doc IDs and frequencies once into ArrayPool-rented buffers, then yields (DocId, Freq) pairs via MoveNext(). Optionally decodes positions when created via CreateWithPositions(IndexInput, long, byte).

Lifetime contract: When using the lazy position path, this struct holds a raw byte* pointer into a memory-mapped IndexInput. The source input (_sourceInput) must remain open and un-disposed for the entire lifetime of this PostingsEnum. Callers must not dispose the IndexInput while any PostingsEnum referencing it is still alive.

Internal structInternal SkipEntry
Internal structInternal StreamingPostingsMerger.Result

Result of a streaming merge: the sorted term list and the per-term .pos offsets needed to write the .dic file.

Internal structInternal TermPostingMetadata

Metadata returned by FinishTerm() for storage in the term dictionary.