Package org.apache.lucene.util.packed

Packed integer arrays and streams.

The packed package provides

  • sequential and random access capable arrays of positive longs,
  • routines for efficient serialization and deserialization of streams of packed integers.
The implementations provide different trade-offs between memory usage and access speed. The standard usage scenario is replacing large int or long arrays in order to reduce the memory footprint.

The main access point is the PackedInts factory.

In-memory structures

  • PackedInts.Mutable
    • Only supports positive longs.
    • Requires the number of bits per value to be known in advance.
    • Random-access for both writing and reading.
  • GrowableWriter
    • Same as PackedInts.Mutable but grows the number of bits per values when needed.
    • Useful to build a PackedInts.Mutable from a read-once stream of longs.
  • PagedGrowableWriter
    • Slices data into fixed-size blocks stored in GrowableWriters.
    • Supports more than 2B values.
    • You should use Appending(Delta)PackedLongBuffer instead if you don't need random write access.
  • AppendingDeltaPackedLongBuffer
    • Can store any sequence of longs.
    • Compression is good when values are close to each other.
    • Supports random reads, but only sequential writes.
    • Can address up to 2^42 values.
  • AppendingPackedLongBuffer
    • Same as AppendingDeltaPackedLongBuffer but assumes values are 0-based.
  • MonotonicAppendingLongBuffer
    • Same as AppendingDeltaPackedLongBuffer except that compression is good when the stream is a succession of affine functions.

Disk-based structures