Implementation of a simple compressed sorted string dictionary based on rear-coding and fixed-length blocks.
It supports the select (aka access) operation, which returns the i-th lexicographically-ranked string for a given integer i, and the rank (aka lexicographic search) operation, which counts the number of strings that are lexicographically smaller than or equal to a given string pattern.

Rear-coding compresses the sorted strings in the dictionary by removing the longest common prefix, of length lcp, between any two consecutive strings s, t, and by encoding the length of the suffix of s to be removed from it to get the value lcp. This length is encoded via the variable-byte code.

To provide fast query operations, a bucketing approach is used: when a given amount of bytes (block_bytes) is written, the compression is “restarted” by creating a block in which the next string to be encoded (referred to as header) is stored explicitly. The cumulative counts of the strings in the blocks are also stored. Then, select (resp. rank) is implemented via a binary search on these counts (resp. headers) to find the appropriate block containing the answer, followed by a sequential decompression of the block.

Two implementations are provided: one that stores the headers at the start of the blocks, the other that stores the headers separately in a contiguous area.


This is a header-only library. To compile the example, use the following commands:

git clone
cd RearCodedArray
cmake . -DCMAKE_BUILD_TYPE=Release


  1. Paolo Ferragina, Roberto Grossi, Ankur Gupta, Rahul Shah, Jeffrey Scott Vitter. On searching compressed string collections cache-obliviously. PODS 2008: 181-190.
  2. Miguel A. Martínez-Prieto, Nieves R. Brisaboa, Rodrigo Cánovas, Francisco Claude, Gonzalo Navarro. Practical compressed string dictionaries. Inf. Syst. 56: 73-108 (2016).


View Github