Compression::Util is a pure-Perl toolkit for experimenting with, composing, and implementing data-compression algorithms and formats. It exposes everything from individual bit/byte primitives and entropy coders to complete BWT, LZ77/LZSS, LZW, DEFLATE, Bzip2, GZIP, ZLIB, and LZ4 pipelines.
🧠 Design philosophy: build compression pipelines from small, reusable pieces. Use a ready-made compressor when you want convenience, or combine low-level transforms and coders when you want complete control.
- 🌀 Burrows–Wheeler, Move-to-Front, Run-Length and delta transforms
- 🌳 Huffman and arithmetic coding, including adaptive arithmetic coding
- 🔤 Fibonacci, Elias gamma/omega, ABC and OBH integer coding
- 📦 LZ77/LZSS/LZW compression and symbolic variants for arbitrary integer symbols
- 🧰 BWT-based compression, plus Bzip2, GZIP, ZLIB, DEFLATE and LZ4 support
- 🔬 Low-level bit/byte I/O, checksums, alphabet handling and format primitives
- 🧩 Pipelines are composable: plug different entropy coders or LZ parsers into higher-level algorithms
- 🐪 Pure Perl — useful for portability, experimentation, education and algorithm research
Install from CPAN with:
cpanm Compression::UtilOr install directly with your preferred CPAN client:
cpan Compression::Util- ⚡ Quick Start
- 🧩 Overview
- 🧭 Choosing the Right API
- ⚙️ Package Variables
- 🚀 High-Level Functions
- 🧱 Medium-Level Functions
- 🔬 Low-Level Functions
- 💡 Examples
- 📖 References
- 🔎 See Also
Version: 0.17
use 5.036;
use Getopt::Std qw(getopts);
use Compression::Util qw(:all);
use constant {CHUNK_SIZE => 1 << 17};
local $Compression::Util::VERBOSE = 0;
getopts('d', \my %opts);
sub compress ($fh, $out_fh) {
while (read($fh, (my $chunk), CHUNK_SIZE)) {
print $out_fh bwt_compress($chunk);
}
}
sub decompress ($fh, $out_fh) {
while (!eof($fh)) {
print $out_fh bwt_decompress($fh);
}
}
$opts{d} ? decompress(\*STDIN, \*STDOUT) : compress(\*STDIN, \*STDOUT);
A quicker taste, compressing and decompressing a string in memory with a real, industry-standard container format:
use 5.036;
use Compression::Util qw(gzip_compress gzip_decompress);
my $data = "the quick brown fox jumps over the lazy dog " x 20;
my $compressed = gzip_compress($data);
my $round_trip = gzip_decompress($compressed);
$round_trip eq $data or die "round-trip failed";
printf("%d -> %d bytes (%.1f%%)\n",
length($data), length($compressed), 100 * length($compressed) / length($data));
Compression::Util is a function-based module, implementing various techniques used in data compression, such as:
- Burrows-Wheeler transform
- Move-to-front transform
- Huffman Coding
- Arithmetic Coding (in fixed bits, static and adaptive)
- Run-length encoding
- Fibonacci coding
- Elias gamma/omega coding
- Delta coding
- BWT-based (de)compression
- LZ77/LZSS (de)compression
- LZW (de)compression
- Bzip2 (de)compression
- GZIP (de)compression
- ZLIB (de)compression
- LZ4 (de)compression
Nothing is exported by default: every function below is available on demand, either individually or via the :all tag (see "EXPORT"). This makes it straightforward to build compressors ranging from a couple of lines that reuse a full pipeline ("HIGH-LEVEL FUNCTIONS"), down to hand-rolled formats assembled from individual transforms and entropy coders ("LOW-LEVEL FUNCTIONS" and "MEDIUM-LEVEL FUNCTIONS").
The provided techniques can be easily combined in various ways to create powerful compressors, such as the Bzip2 compressor, which is a pipeline of the following methods:
1. Run-length encoding (RLE4)
2. Burrows-Wheeler transform (BWT)
3. Move-to-front transform (MTF)
4. Zero run-length encoding (ZRLE)
5. Huffman coding
A simple BWT-based compression method (similar to Bzip2) is provided by the function bwt_compress(), which can be explicitly implemented as:
use 5.036;
use Compression::Util qw(:all);
my $data = do { open my $fh, '<:raw', $^X; local $/; <$fh> };
my $rle4 = rle4_encode(string2symbols($data));
my ($bwt, $idx) = bwt_encode(symbols2string($rle4));
my ($mtf, $alphabet) = mtf_encode(string2symbols($bwt));
my $rle = zrle_encode($mtf);
my $enc = pack('N', $idx)
. encode_alphabet($alphabet)
. create_huffman_entry($rle);
say "Original size : ", length($data);
say "Compressed size: ", length($enc);
# Decompress the result
bwt_decompress($enc) eq $data or die "decompression error";
If you just want to shrink some data and don't care how, reach for a "HIGH-LEVEL FUNCTIONS" pair such as gzip_compress() / gzip_decompress() (a real, interoperable GZIP stream) or bwt_compress() / bwt_decompress() (a lighter, module-specific container). If you want to design your own container format out of individual building blocks -- for example, applying Huffman coding to the output of your own transform -- use the "MEDIUM-LEVEL FUNCTIONS". If you are implementing a specific bit-exact format (such as DEFLATE) or need direct access to bit/byte I/O, transforms, or entropy coders in isolation, use the "LOW-LEVEL FUNCTIONS".
A bit value is either 1 or 0.
A bitstring is a string containing only the characters 1 and 0 (not packed bits). Many low-level encoders return a bitstring; pass it through pack('B*', $bitstring) to get a packed binary string, and unpack('B*', $string) to go the other way.
A byte value is an integer between 0 and 255, inclusive.
A string means a binary (non-UTF*) string. Any file-handle used to read or write such a string must have its encoding layer set to :raw (see "filehandle" below).
An array of symbols means an array-ref of non-negative integer values. Symbols are not restricted to the [0, 255] byte range unless a specific function says otherwise (for example encode_alphabet_256()); this is what makes the "symbolic" variants of many functions (bwt_encode_symbolic(), lzss_encode_symbolic(), lz77_encode_symbolic(), etc.) useful for compressing sequences of large integers rather than bytes, such as intermediate output from another codec.
An alphabet is a sorted array-ref of the distinct symbols that occur in a sequence, with no repeats. Several functions (such as mtf_encode() and adaptive_ac_encode()) return or accept an alphabet alongside the encoded data.
A filehandle is denoted by $fh. The encoding of file-handles must be set to :raw (i.e. open(my $fh, '<:raw', $path)), so that bytes are read and written without any character-encoding or line-ending translation.
Many of the higher-level functions accept an optional code-ref parameter (commonly named $entropy_sub) that plugs in the final entropy-coding stage of a pipeline, letting you swap Huffman coding for Arithmetic Coding (or a custom coder) without re-implementing the surrounding pipeline. See "create_huffman_entry" and "create_ac_entry" for the two coders built into this module, and "Combining LZSS + MRL compression" for a usage example.
A few conventions apply consistently across the module and are worth knowing up front, rather than being repeated in every function's description:
-
String or symbols. Functions documented as accepting
\@symbolswill also accept a plain string in its place (it is converted internally viastring2symbols()). Functions with a_symboliccounterpart (e.g.bwt_encode()/bwt_encode_symbolic()) exist because the plain version is optimized for byte strings, while the_symbolicversion works with arbitrary non-negative integers and has no upper bound on symbol values. -
Filehandle or string, transparently. Any decoder that takes
$fhas its first argument will also accept a plain packed binary string; internally it opens an in-memory filehandle over the string (open($fh2, '<:raw', \$string)) and recurses. This meanslzss_decompress($string)andlzss_decompress($fh)both work, and the two forms can be mixed freely when chaining decoders. -
Streaming decompression. Because decoders read only as much as they need from
$fh, several compressed blocks can be concatenated back-to-back in a single file and decoded one at a time in a loop (see the "SYNOPSIS"), without needing to record each block's length up front. -
Package variables are dynamically scoped. The
$LZ_*tuning variables ("PACKAGE VARIABLES" below) are read at the time an LZ-family encoder runs. Uselocalto override them for the duration of a single call, so that concurrent or later calls elsewhere in your program are unaffected:{ local $Compression::Util::LZ_MAX_CHAIN_LEN = 128; my $enc = lzss_compress($data); # uses the higher chain length } # back to the default of 32 here
Compression::Util provides the following package variables:
$Compression::Util::VERBOSE = 0; # true to enable verbose/debug mode
$Compression::Util::LZ_MIN_LEN = 4; # minimum match length in LZ parsing
$Compression::Util::LZ_MAX_LEN = 1 << 15; # maximum match length in LZ parsing
$Compression::Util::LZ_MAX_DIST = ~0; # maximum back-reference distance allowed
$Compression::Util::LZ_MAX_CHAIN_LEN = 32; # how many recent positions to remember for each match in LZ parsing
$Compression::Util::LZ_MAX_CHAIN_WIDTH = 1024; # how many positions to store per matched prefix
These package variables can also be imported as:
use Compression::Util qw(
$LZ_MIN_LEN
$LZ_MAX_LEN
$LZ_MAX_DIST
$LZ_MAX_CHAIN_LEN
$LZ_MAX_CHAIN_WIDTH
);
When set to a true value, many functions print diagnostic information (block sizes, header fields, checksums, and so on) to STDERR. Useful when debugging a hand-assembled pipeline or inspecting third-party GZIP/ZLIB/LZ4 streams being decoded.
By default, $VERBOSE is set to 0.
Minimum length of a match in LZ parsing. The value must be an integer greater than or equal to 2. Larger values will result in faster parsing, but lower compression ratio.
By default, $LZ_MIN_LEN is set to 4.
NOTE: for lzss_encode_fast() is recommended to set $LZ_MIN_LEN = 5, which will result in slightly better compression ratio. lzss_encode_hash4() always uses a fixed minimum match length of 4, ignoring this variable.
Maximum length of a match in LZ parsing. The value must be an integer greater than or equal to 0.
By default, $LZ_MAX_LEN is set to 32768.
NOTE: the functions lz77_encode() and lzb_compress() will ignore this value and will always use unlimited match lengths.
Maximum back-reference distance allowed in LZ parsing. Smaller values will result in faster parsing, but lower compression ratio.
By default, the value is unlimited, meaning that arbitrarily large back-references will be generated.
NOTE: the function lzb_compress() will ignore this value and will always use the value 2**16 - 1 as the maximum back-reference distance.
The value of $LZ_MAX_CHAIN_LEN controls the amount of recent positions to remember for each matched prefix. A larger value results in better compression, finding longer matches, at the expense of speed.
By default, $LZ_MAX_CHAIN_LEN is set to 32.
NOTE: the function lzss_encode_fast() will ignore this value, always using a value of 1.
The value of $LZ_MAX_CHAIN_WIDTH controls the number of positions to store from each match. A larger value may result in better compression, finding longer matches, at the expense of speed.
By default, $LZ_MAX_CHAIN_WIDTH is set to 1024.
NOTE: the function lzss_encode_fast() will ignore this value, always using a value of 1.
The module ships three LZ-parsing engines with different speed/ratio trade-offs, all of which produce output consumable by the same lzss_decode() / lzss_decode_symbolic():
lzss_encode() # best compression; honors all $LZ_* tuning variables
lzss_encode_fast() # faster, chain length fixed to 1; good with $LZ_MIN_LEN = 5
lzss_encode_hash4() # fastest, LZ4-style O(1) hashing; bytes only, min match fixed to 4
Any of these can be plugged into the higher-level compressors (lz77_compress(), lzss_compress(), lzb_compress(), lz4_compress(), gzip_compress(), zlib_compress()) via their optional $lzss_encoding_sub argument, so the trade-off can be tuned per call without touching the rest of the pipeline.
Ready-made compressors and decompressors, each wrapping a full pipeline of transforms and an entropy coder. Start here unless you need to design your own format.
create_huffman_entry(\@symbols) # Create a Huffman Coding block
decode_huffman_entry($fh) # Decode a Huffman Coding block
create_ac_entry(\@symbols) # Create an Arithmetic Coding block
decode_ac_entry($fh) # Decode an Arithmetic Coding block
create_adaptive_ac_entry(\@symbols) # Create an Adaptive Arithmetic Coding block
decode_adaptive_ac_entry($fh) # Decode an Adaptive Arithmetic Coding block
mrl_compress($string) # MRL compression (MTF+ZRLE+RLE4+Huffman coding)
mrl_decompress($fh) # Inverse of the above method
mrl_compress_symbolic(\@symbols) # Symbolic MRL compression (MTF+ZRLE+RLE4+Huffman coding)
mrl_decompress_symbolic($fh) # Inverse of the above method
bwt_compress($string) # BWT-based compression (RLE4+BWT+MTF+ZRLE+Huffman coding)
bwt_decompress($fh) # Inverse of the above method
bwt_compress_symbolic(\@symbols) # Symbolic BWT-based compression (RLE4+sBWT+MTF+ZRLE+Huffman coding)
bwt_decompress_symbolic($fh) # Inverse of the above method
bzip2_compress($string) # Compress a given string using the Bzip2 format
bzip2_decompress($fh) # Inverse of the above method
gzip_compress($string) # Compress a given string using the GZIP format
gzip_decompress($fh) # Inverse of the above method
zlib_compress($string) # Compress a given string using the ZLIB format
zlib_decompress($fh) # Inverse of the above method
lzss_compress($string) # LZSS + DEFLATE-like encoding of lengths and distances
lzss_decompress($fh) # Inverse of the above method
lzss_compress_symbolic(\@symbols) # Symbolic LZSS + DEFLATE-like encoding of lengths and distances
lzss_decompress_symbolic($fh) # Inverse of the above method
lz77_compress($string) # LZ77 + Huffman coding of lengths and literals + OBH for distances
lz77_decompress($fh) # Inverse of the above method
lz77_compress_symbolic(\@symbols) # Symbolic LZ77 + Huffman coding of lengths and literals + OBH for distances
lz77_decompress_symbolic($fh) # Inverse of the above method
lzb_compress($string) # LZSS compression, using a byte-aligned encoding method, similar to LZ4
lzb_decompress($fh) # Inverse of the above method
lzw_compress($string) # LZW + abc_encode() compression
lzw_decompress($fh) # Inverse of the above method
lz4_compress($string) # Compress a given string using the LZ4 frame format
lz4_decompress($fh) # Inverse of the above method
Individual transforms and entropy coders, meant to be composed together into your own pipeline (as the "HIGH-LEVEL FUNCTIONS" do internally).
deltas(\@ints) # Computes the differences between integers
accumulate(\@deltas) # Inverse of the above method
delta_encode(\@ints) # Delta+RLE+Elias omega encoding of an array-ref of integers
delta_decode($fh) # Inverse of the above method
fibonacci_encode(\@symbols) # Fibonacci coding of an array-ref of symbols
fibonacci_decode($fh) # Inverse of the above method
elias_gamma_encode(\@symbols) # Elias Gamma coding method of an array-ref of symbols
elias_gamma_decode($fh) # Inverse of the above method
elias_omega_encode(\@symbols) # Elias Omega coding method of an array-ref of symbols
elias_omega_decode($fh) # Inverse of the above method
abc_encode(\@symbols) # Adaptive Binary Concatenation method of an array-ref of symbols
abc_decode($fh) # Inverse of the above method
obh_encode(\@symbols) # Offset bits + Huffman coding of an array-ref of symbols
obh_decode($fh) # Inverse of the above method
bwt_encode($string) # Burrows-Wheeler transform
bwt_decode($bwt, $idx) # Inverse of Burrows-Wheeler transform
bwt_encode_symbolic(\@symbols) # Burrows-Wheeler transform over an array-ref of symbols
bwt_decode_symbolic(\@bwt, $idx) # Inverse of symbolic Burrows-Wheeler transform
mtf_encode(\@symbols) # Move-to-front transform
mtf_decode(\@mtf, \@alphabet) # Inverse of the above method
encode_alphabet(\@alphabet) # Encode an alphabet of symbols into a binary string
decode_alphabet($fh) # Inverse of the above method
encode_alphabet_256(\@alphabet) # Encode an alphabet of symbols (limited to [0..255]) into a binary string
decode_alphabet_256($fh) # Inverse of the above method
frequencies(\@symbols) # Returns a dictionary with symbol frequencies
run_length(\@symbols, $max=undef) # Run-length encoding, returning a 2D array-ref
rle4_encode(\@symbols, $max=255) # Run-length encoding with 4 or more consecutive characters
rle4_decode(\@rle4) # Inverse of the above method
zrle_encode(\@symbols) # Run-length encoding of zeros
zrle_decode(\@zrle) # Inverse of the above method
ac_encode(\@symbols) # Arithmetic Coding applied on an array-ref of symbols
ac_decode($bitstring, \%freq) # Inverse of the above method
adaptive_ac_encode(\@symbols) # Adaptive Arithmetic Coding applied on an array-ref of symbols
adaptive_ac_decode($bitstring, \@alphabet) # Inverse of the above method
lzw_encode($string) # LZW encoding of a given string
lzw_decode(\@symbols) # Inverse of the above method
Bit/byte I/O primitives, raw LZ parsing, Huffman table construction, and DEFLATE block handling. These are what the medium- and high-level functions are built from; reach for them when implementing a specific bit-exact format or when you need finer control than the higher tiers expose.
crc32($string, $prev_crc = 0) # Compute the CRC32 value of a given string
adler32($string, $prev_adler = 1) # Compute the Adler32 value of a given string
read_bit($fh, \$buffer) # Read one bit from file-handle (MSB)
read_bit_lsb($fh, \$buffer) # Read one bit from file-handle (LSB)
read_bits($fh, $len) # Read `$len` bits from file-handle (MSB)
read_bits_lsb($fh, $len) # Read `$len` bits from file-handle (LSB)
int2bits($symbol, $size) # Convert an integer to bits of width `$size` (MSB)
int2bits_lsb($symbol, $size) # Convert an integer to bits of width `$size` (LSB)
bits2int($fh, $size, \$buffer) # Inverse of `int2bits()`
bits2int_lsb($fh, $size, \$buffer) # Inverse of `int2bits_lsb()`
bytes2int($fh, $n) # Read `$n` bytes from file-handle as an integer (MSB)
bytes2int_lsb($fh, $n) # Read `$n` bytes from file-handle as an integer (LSB)
int2bytes($symbol, $size) # Convert an integer into `$size` bytes. (MSB)
int2bytes_lsb($symbol, $size) # Convert an integer into `$size` bytes. (LSB)
string2symbols($string) # Returns an array-ref of code points
symbols2string(\@symbols) # Returns a string, given an array-ref of code points
read_null_terminated($fh) # Read a binary string that ends with NULL ("\0")
binary_vrl_encode($bitstring) # Binary variable run-length encoding
binary_vrl_decode($bitstring) # Binary variable run-length decoding
bwt_sort($string) # Burrows-Wheeler sorting
bwt_sort_symbolic(\@symbols) # Burrows-Wheeler sorting, applied on an array-ref of symbols
huffman_encode(\@symbols, \%dict) # Huffman encoding
huffman_decode($bitstring, \%dict) # Huffman decoding, given a string of bits
huffman_from_freq(\%freq) # Create Huffman dictionaries, given an hash-ref of frequencies
huffman_from_symbols(\@symbols) # Create Huffman dictionaries, given an array-ref of symbols
huffman_from_code_lengths(\@lens) # Create canonical Huffman codes, given an array-ref of code lengths
make_deflate_tables($max_dist, $max_len) # Returns the DEFLATE tables for distance and length symbols
find_deflate_index($value, \@table) # Returns the index in a DEFLATE table, given a numerical value
lzss_encode($string) # LZSS encoding into literals, distances and lengths
lzss_encode_symbolic(\@symbols) # LZSS encoding into literals, distances and lengths (symbolic)
lzss_encode_fast($string) # Fast-LZSS encoding into literals, distances and lengths
lzss_encode_fast_symbolic(\@symbols) # Fast-LZSS encoding into literals, distances and lengths (symbolic)
lzss_encode_hash4($string) # LZSS encoding via O(1) flat-array hashing, LZ4-style.
lzss_decode(\@lits, \@dist, \@lens) # Inverse of lzss_encode() and lzss_encode_fast()
lzss_decode_symbolic(\@lits, \@dist, \@lens) # Inverse of lzss_encode_symbolic() and lzss_encode_fast_symbolic()
lz77_encode($string) # LZ77 encoding into literals, distances, lengths and matches
lz77_encode_symbolic(\@symbols) # LZ77 encoding into literals, distances, lengths and matches (symbolic)
lz77_decode(\@lits, \@dist, \@lens, \@matches) # Inverse of lz77_encode()
lz77_decode_symbolic(\@lits, \@dist, \@lens, \@matches) # Inverse of lz77_encode_symbolic()
deflate_encode(\@lits, \@dist, \@lens) # DEFLATE-like encoding of values returned by lzss_encode()
deflate_decode($fh) # Inverse of the above method
my $string = create_huffman_entry(\@symbols);
High-level function that generates a Huffman coding block, given an array-ref of symbols.
Example:
my $block = create_huffman_entry([65, 65, 66, 65, 67]);
decode_huffman_entry($block); # [65, 65, 66, 65, 67]
my $symbols = decode_huffman_entry($fh);
my $symbols = decode_huffman_entry($string);
Inverse of create_huffman_entry().
my $string = create_ac_entry(\@symbols);
High-level function that generates an Arithmetic Coding block, given an array-ref of symbols.
my $symbols = decode_ac_entry($fh);
my $symbols = decode_ac_entry($string);
Inverse of create_ac_entry().
my $string = create_adaptive_ac_entry(\@symbols);
High-level function that generates an Adaptive Arithmetic Coding block, given an array-ref of symbols.
Unlike create_ac_entry(), no frequency table needs to be stored in the block: the model adapts to the data as it is decoded, which can save space on short or skewed inputs at the cost of somewhat slower encoding/decoding.
my $symbols = decode_adaptive_ac_entry($fh);
my $symbols = decode_adaptive_ac_entry($string);
Inverse of create_adaptive_ac_entry().
# With Huffman coding
my $string = lz77_compress($data);
my $string = lz77_compress(\@symbols);
# With Arithmetic Coding
my $string = lz77_compress($data, \&create_ac_entry);
# Using Fast-LZSS parsing + Huffman coding
my $string = lz77_compress($data, \&create_huffman_entry, \&lzss_encode_fast);
High-level function that performs LZ77 compression on the provided data, using the pipeline:
1. lz77_encode
2. create_huffman_entry(literals)
3. create_huffman_entry(lengths)
4. create_huffman_entry(matches)
5. obh_encode(distances)
The function accepts either a string or an array-ref of symbols as the first argument.
# With Huffman coding
my $data = lz77_decompress($fh);
my $data = lz77_decompress($string);
# With Arithemtic coding
my $data = lz77_decompress($fh, \&decode_ac_entry);
my $data = lz77_decompress($string, \&decode_ac_entry);
# Symbolic, with Huffman coding
my $symbols = lz77_decompress_symbolic($fh);
my $symbols = lz77_decompress_symbolic($string);
Inverse of lz77_compress() and lz77_compress_symbolic(), respectively.
# With Huffman coding
my $string = lzss_compress($data);
my $string = lzss_compress(\@symbols);
# With Arithmetic Coding
my $string = lzss_compress($data, \&create_ac_entry);
# Using Fast-LZSS parsing + Huffman coding
my $string = lzss_compress($data, \&create_huffman_entry, \&lzss_encode_fast);
High-level function that performs LZSS (Lempel-Ziv-Storer-Szymanski) compression on the provided data, using the pipeline:
1. lzss_encode
2. deflate_encode
The function accepts either a string or an array-ref of symbols as the first argument.
# With Huffman coding
my $data = lzss_decompress($fh);
my $data = lzss_decompress($string);
# With Arithmetic coding
my $data = lzss_decompress($fh, \&decode_ac_entry);
my $data = lzss_decompress($string, \&decode_ac_entry);
# Symbolic, with Huffman coding
my $symbols = lzss_decompress_symbolic($fh);
my $symbols = lzss_decompress_symbolic($string);
Inverse of lzss_compress() and lzss_compress_symbolic(), respectively.
my $string = lzb_compress($data);
my $string = lzb_compress($data, \&lzss_encode_fast); # with fast-LZ parsing
High-level function that performs byte-oriented LZSS compression, inspired by LZ4. Unlike lz4_compress(), this is a self-contained format specific to this module, not an interoperable LZ4 stream.
my $data = lzb_decompress($fh);
my $data = lzb_decompress($string);
Inverse of lzb_compress().
my $string = lz4_compress($fh);
my $string = lz4_compress($data);
my $string = lz4_compress($data, \&lzss_encode_fast); # with fast-LZ parsing
Valid LZ4 compressor, using the LZ4 Frame format, given either a string or an input file-handle.
The input data is split into chunks of length 2**17 and compressed into independent LZ4 blocks.
my $data = lz4_decompress($fh);
my $data = lz4_decompress($string);
Decompress LZ4 Frame data, given either a string or an input file-handle. Concatenated LZ4 Frames are also supported.
my $string = lzw_compress($data);
High-level function that performs LZW (Lempel-Ziv-Welch) compression on the provided data, using the pipeline:
1. lzw_encode
2. abc_encode
my $data = lzw_decompress($fh);
Performs Lempel-Ziv-Welch (LZW) decompression on the provided string or file-handle. Inverse of lzw_compress().
# Using Huffman Coding
my $string = bwt_compress($data);
# Using Arithmetic Coding
my $string = bwt_compress($data, \&create_ac_entry);
High-level function that performs BWT-based compression on the provided data, using the pipeline:
1. rle4_encode
2. bwt_encode
3. mtf_encode
4. zrle_encode
5. create_huffman_entry
# With Huffman coding
my $data = bwt_decompress($fh);
my $data = bwt_decompress($string);
# With Arithmetic coding
my $data = bwt_decompress($fh, \&decode_ac_entry);
my $data = bwt_decompress($string, \&decode_ac_entry);
Inverse of bwt_compress().
# Does Huffman coding
my $string = bwt_compress_symbolic(\@symbols);
# Does Arithmetic coding
my $string = bwt_compress_symbolic(\@symbols, \&create_ac_entry);
Similar to bwt_compress(), except that it accepts an arbitrary array-ref of non-negative integer values as input. It is also a bit slower on large inputs.
# Using Huffman coding
my $symbols = bwt_decompress_symbolic($fh);
my $symbols = bwt_decompress_symbolic($string);
# Using Arithmetic coding
my $symbols = bwt_decompress_symbolic($fh, \&decode_ac_entry);
my $symbols = bwt_decompress_symbolic($string, \&decode_ac_entry);
Inverse of bwt_compress_symbolic().
my $string = bzip2_compress($data);
my $string = bzip2_compress($fh);
Valid Bzip2 compressor, given a string or an input file-handle. The output can be decompressed by any standard bzip2 tool, not just by bzip2_decompress().
my $data = bzip2_decompress($string);
my $data = bzip2_decompress($fh);
Valid Bzip2 decompressor, given a string or an input file-handle. Understands streams produced by the standard bzip2 tool as well as by bzip2_compress().
my $string = gzip_compress($fh);
my $string = gzip_compress($data);
my $string = gzip_compress($data, \&lzss_encode_fast); # using fast LZ-parsing
Valid GZIP compressor (RFC 1952), given a string or an input file-handle. The output is a standard .gz stream, decodable with gzip -d, zcat, or any GZIP-compatible library.
use Compression::Util qw(gzip_compress);
open my $out_fh, '>:raw', 'file.txt.gz' or die $!;
print $out_fh gzip_compress("Hello, world!\n");
my $data = gzip_decompress($string);
my $data = gzip_decompress($fh);
Valid GZIP decompressor (RFC 1952), given a string or an input file-handle. Understands streams produced by the standard gzip tool as well as by gzip_compress(), including multiple concatenated members.
my $string = zlib_compress($fh);
my $string = zlib_compress($data);
my $string = zlib_compress($data, \&lzss_encode_fast); # using fast LZ-parsing
Valid ZLIB compressor (RFC 1950), given a string or an input file-handle.
my $data = zlib_decompress($string);
my $data = zlib_decompress($fh);
Valid ZLIB decompressor (RFC 1950), given a string or an input file-handle.
# Does Huffman coding
my $enc = mrl_compress($str);
my $enc = mrl_compress(\@symbols);
# Does Arithmetic coding
my $enc = mrl_compress($str, \&create_ac_entry);
my $enc = mrl_compress(\@symbols, \&create_ac_entry);
A fast compression method (no BWT stage, so it is cheaper but generally less effective than bwt_compress()), using the following pipeline:
1. mtf_encode
2. zrle_encode
3. rle4_encode
4. create_huffman_entry
It accepts either a string or an arbitrary array-ref of non-negative integer values as input; mrl_compress() and mrl_compress_symbolic() are, in fact, the same function under two names.
# With Huffman coding
my $data = mrl_decompress($fh);
my $data = mrl_decompress($string);
# Symbolic, with Huffman coding
my $symbols = mrl_decompress_symbolic($fh);
my $symbols = mrl_decompress_symbolic($string);
# Symbolic, with Arithmetic coding
my $symbols = mrl_decompress_symbolic($fh, \&decode_ac_entry);
my $symbols = mrl_decompress_symbolic($string, \&decode_ac_entry);
Inverse of mrl_compress() and mrl_compress_symbolic().
my $freq = frequencies(\@symbols);
Returns a hash-ref dictionary with frequencies, given an array-ref of symbols. For example:
frequencies([65, 65, 66, 65, 67]); # { 65 => 3, 66 => 1, 67 => 1 }
my $deltas = deltas(\@integers);
Computes the differences between consecutive integers, returning an array-ref, where the first delta is relative to 0. For example:
deltas([10, 12, 13, 11]); # [10, 2, 1, -2]
my $integers = accumulate(\@deltas);
Inverse of deltas(). For example:
accumulate([10, 2, 1, -2]); # [10, 12, 13, 11]
my $string = delta_encode(\@integers);
Encodes a sequence of integers (including negative integers) using Delta + Run-length + Elias omega coding, returning a binary string.
Delta encoding calculates the difference between consecutive integers in the sequence and encodes these differences using Elias omega coding. When it's beneficial, runs of identical symbols are collapsed with RLE.
This method supports both positive and negative integers, and is well suited to sequences that increase monotonically or vary smoothly, such as timestamps or sorted indices.
# Given a file-handle
my $integers = delta_decode($fh);
# Given a string
my $integers = delta_decode($string);
Inverse of delta_encode().
my $string = fibonacci_encode(\@symbols);
Encodes a sequence of non-negative integers using Fibonacci coding, returning a binary string. Fibonacci codes are self-synchronizing (a corrupted bit cannot propagate errors past the next codeword) and tend to compress small values well.
# Given a file-handle
my $symbols = fibonacci_decode($fh);
# Given a binary string
my $symbols = fibonacci_decode($string);
Inverse of fibonacci_encode().
my $string = elias_gamma_encode(\@symbols);
Encodes a sequence of non-negative integers using Elias Gamma coding, returning a binary string. Elias Gamma coding is simple and fast, and works best when small values dominate the sequence.
# Given a file-handle
my $symbols = elias_gamma_decode($fh);
# Given a binary string
my $symbols = elias_gamma_decode($string);
Inverse of elias_gamma_encode().
my $string = elias_omega_encode(\@symbols);
Encodes a sequence of non-negative integers using Elias Omega coding, returning a binary string. Compared to Elias Gamma, Omega coding scales better to large values, at a small cost for small ones.
# Given a file-handle
my $symbols = elias_omega_decode($fh);
# Given a binary string
my $symbols = elias_omega_decode($string);
Inverse of elias_omega_encode().
my $string = abc_encode(\@symbols);
Encodes a sequence of non-negative integers using the Adaptive Binary Concatenation encoding method.
This method is particularly effective in encoding a sequence of integers that are in ascending order or have roughly the same size in binary (this is why it is used as the default entropy coder for LZW codes in lzw_compress(), since LZW dictionary indices grow roughly monotonically).
# Given a file-handle
my $symbols = abc_decode($fh);
# Given a binary string
my $symbols = abc_decode($string);
Inverse of abc_encode().
# With Huffman Coding
my $string = obh_encode(\@symbols);
# With Arithmetic Coding
my $string = obh_encode(\@symbols, \&create_ac_entry);
Encodes a sequence of non-negative integers using offset bits and Huffman coding.
This method is particularly effective in encoding a sequence of moderately large random integers, such as the list of distances returned by lzss_encode() (which is exactly how lz77_compress() uses it internally).
# Given a file-handle
my $symbols = obh_decode($fh); # Huffman decoding
my $symbols = obh_decode($fh, \&decode_ac_entry); # Arithmetic decoding
# Given a binary string
my $symbols = obh_decode($string); # Huffman decoding
my $symbols = obh_decode($string, \&decode_ac_entry); # Arithmetic decoding
Inverse of obh_encode().
my ($bwt, $idx) = bwt_encode($string);
my ($bwt, $idx) = bwt_encode($string, $lookahead_len);
Applies the Burrows-Wheeler Transform (BWT) to a given string, returning the transformed string and the index of the original string among the sorted rotations.
By default, $lookahead_len is 128 (see bwt_sort()).
Example:
my ($bwt, $idx) = bwt_encode("banana");
bwt_decode($bwt, $idx); # "banana"
my $string = bwt_decode($bwt, $idx);
Reverses the Burrows-Wheeler Transform (BWT) applied to a string.
The function returns the original string.
my ($bwt_symbols, $idx) = bwt_encode_symbolic(\@symbols);
Applies the Burrows-Wheeler Transform (BWT) to a sequence of symbolic elements.
my $symbols = bwt_decode_symbolic(\@bwt_symbols, $idx);
Reverses the Burrows-Wheeler Transform (BWT) applied to a sequence of symbolic elements.
my $mtf = mtf_encode(\@symbols, \@alphabet);
my ($mtf, $alphabet) = mtf_encode(\@symbols);
Performs Move-To-Front (MTF) encoding on a sequence of symbols (also accepts a plain string in place of \@symbols).
The function returns the encoded MTF sequence and the sorted list of unique symbols in the input data, representing the alphabet.
Optionally, the alphabet can be provided as a second argument. When two arguments are provided, only the MTF sequence is returned.
Example:
my ($mtf, $alphabet) = mtf_encode([2, 2, 5, 2, 9]);
# $mtf = [0, 0, 1, 1, 2]
# $alphabet = [2, 5, 9]
mtf_decode($mtf, $alphabet); # [2, 2, 5, 2, 9]
my $symbols = mtf_decode(\@mtf, \@alphabet);
Inverse of mtf_encode().
my $string = encode_alphabet(\@alphabet); # supports arbitrarily large symbols
my $string = encode_alphabet_256(\@alphabet); # limited to symbols [0..255]
Encode a sorted alphabet of symbols into a binary string. Use encode_alphabet_256() when you know every symbol fits in a byte (for instance, the alphabet produced by mtf_encode() on ordinary text); it produces a more compact header than the general-purpose encode_alphabet().
my $alphabet = decode_alphabet($fh);
my $alphabet = decode_alphabet($string);
my $alphabet = decode_alphabet_256($fh);
my $alphabet = decode_alphabet_256($string);
Decodes an encoded alphabet, given a file-handle or a binary string, returning an array-ref of symbols. Inverse of encode_alphabet() and encode_alphabet_256(), respectively.
my $rl = run_length(\@symbols);
my $rl = run_length(\@symbols, $max_run);
Performs Run-Length Encoding (RLE) on a sequence of symbolic elements.
It takes two parameters: \@symbols, representing an array of symbols, and $max_run, indicating the maximum run length allowed.
The function returns a 2D-array, with pairs: [symbol, run_length], such that the following code reconstructs the \@symbols array:
my @symbols = map { ($_->[0]) x $_->[1] } @$rl;
By default, the maximum run-length is unlimited. Example:
run_length([1, 1, 1, 2, 2, 3]); # [[1, 3], [2, 2], [3, 1]]
my $rle4 = rle4_encode($string);
my $rle4 = rle4_encode(\@symbols);
my $rle4 = rle4_encode(\@symbols, $max_run);
Performs Run-Length Encoding (RLE) on a sequence of symbolic elements, specifically designed for runs of four or more consecutive symbols (shorter runs are left untouched, which is cheaper when most of the data doesn't repeat).
It takes two parameters: \@symbols, representing an array of symbols, and $max_run, indicating the maximum run length allowed during encoding.
The function returns the encoded RLE sequence as an array-ref of symbols.
By default, the maximum run-length is limited to 255.
my $symbols = rle4_decode(\@rle4);
my $symbols = rle4_decode($rle4_string);
Inverse of rle4_encode().
my $zrle = zrle_encode(\@symbols);
Performs Zero-Run-Length Encoding (ZRLE) on a sequence of symbolic elements, returning the encoded ZRLE sequence as an array-ref of symbols.
This function efficiently encodes runs of zeros, but also increments each symbol by 1 (so that 0 is freed up to serve purely as a run marker in the output alphabet -- useful right after mtf_encode(), whose output is dominated by zeros).
my $symbols = zrle_decode($zrle);
Inverse of zrle_encode().
my ($bitstring, $freq) = ac_encode(\@symbols);
Performs (static) Arithmetic Coding on the provided symbols.
It takes a single parameter, \@symbols, representing the symbols to be encoded.
The function returns two values: $bitstring, which is a string of 1s and 0s, and $freq, representing the frequency table used for encoding -- this table must be kept (or otherwise reconstructed) in order to decode the data with ac_decode(), since it is not embedded in $bitstring. (create_ac_entry() takes care of storing it for you.)
my $symbols = ac_decode($bits_fh, \%freq);
my $symbols = ac_decode($bitstring, \%freq);
Performs Arithmetic Coding decoding using the provided frequency table and a string of 1s and 0s. Inverse of ac_encode().
It takes two parameters: $bitstring, representing a string of 1s and 0s containing the arithmetic coded data, and \%freq, representing the frequency table used for encoding.
The function returns the decoded sequence of symbols.
my ($bitstring, $alphabet) = adaptive_ac_encode(\@symbols);
Performs Adaptive Arithmetic Coding on the provided symbols.
It takes a single parameter, \@symbols, representing the symbols to be encoded.
The function returns two values: $bitstring, which is a string of 1s and 0s, and $alphabet, which is an array-ref of distinct sorted symbols. Unlike ac_encode(), no per-symbol frequencies need to be stored -- only the alphabet -- because the model adapts as it goes.
my $symbols = adaptive_ac_decode($bits_fh, \@alphabet);
my $symbols = adaptive_ac_decode($bitstring, \@alphabet);
Performs Adaptive Arithmetic Coding decoding using the provided alphabet and a string of 1s and 0s.
It takes two parameters: $bitstring, representing a string of 1s and 0s containing the adaptive arithmetic coded data, and \@alphabet, representing the array of distinct sorted symbols that appear in the encoded data.
The function returns the decoded sequence of symbols.
my $symbols = lzw_encode($string);
Performs Lempel-Ziv-Welch (LZW) encoding on the provided string.
It takes a single parameter, $string, representing the data to be encoded.
The function returns an array-ref of symbols (dictionary indices).
my $string = lzw_decode(\@symbols);
Performs Lempel-Ziv-Welch (LZW) decoding on the provided symbols. Inverse of lzw_encode().
The function returns the decoded string.
my $int32 = crc32($data);
my $int32 = crc32($data, $prev_crc32);
Compute the CRC32 checksum of a given string, as used by GZIP and ZLIB. The optional second argument allows a checksum to be computed incrementally, over successive chunks:
my $crc = 0;
$crc = crc32($chunk1, $crc);
$crc = crc32($chunk2, $crc);
my $int32 = adler32($data);
my $int32 = adler32($data, $prev_adler32);
Compute the Adler-32 checksum of a given string, as used by ZLIB. Like crc32(), it can be computed incrementally by passing the running checksum as the second argument; note that its identity value is 1, not 0.
my $bit = read_bit($fh, \$buffer);
Reads a single bit from a file-handle $fh (MSB order).
The function stores the extra bits inside the $buffer, reading one character at a time from the file-handle. Pass the same \$buffer scalar-ref on every call so left-over bits from a partially consumed byte carry over correctly to the next read.
my $bit = read_bit_lsb($fh, \$buffer);
Reads a single bit from a file-handle $fh (LSB order).
The function stores the extra bits inside the $buffer, reading one character at a time from the file-handle.
my $bitstring = read_bits($fh, $bits_len);
Reads a specified number of bits ($bits_len) from a file-handle ($fh) and returns them as a string, in MSB order.
my $bitstring = read_bits_lsb($fh, $bits_len);
Reads a specified number of bits ($bits_len) from a file-handle ($fh) and returns them as a string, in LSB order.
my $bitstring = int2bits($symbol, $size)
Convert a non-negative integer to a bitstring of width $size, in MSB order.
Example:
int2bits(5, 8); # "00000101"
bits2int_from_string(int2bits(5, 8)); # see note below
Note: to turn a bitstring back into a packed binary string (rather than reading it from a file-handle with bits2int()), use pack('B*', $bitstring).
my $bitstring = int2bits_lsb($symbol, $size)
Convert a non-negative integer to a bitstring of width $size, in LSB order.
my $string = int2bytes($symbol, $size);
Convert a non-negative integer to a byte-string of width $size, in MSB order. For example, int2bytes(1, 4) returns the 4-byte big-endian encoding of 1.
my $string = int2bytes_lsb($symbol, $size);
Convert a non-negative integer to a byte-string of width $size, in LSB order (little-endian).
my $integer = bits2int($fh, $size, \$buffer);
Read $size bits from a file-handle $fh and convert them to an integer, in MSB order. Inverse of int2bits().
The function stores the extra bits inside the $buffer, reading one character at a time from the file-handle.
my $integer = bits2int_lsb($fh, $size, \$buffer);
Read $size bits from a file-handle $fh and convert them to an integer, in LSB order. Inverse of int2bits_lsb().
The function stores the extra bits inside the $buffer, reading one character at a time from the file-handle.
my $integer = bytes2int($fh, $n);
my $integer = bytes2int($str, $n);
Read $n bytes from a file-handle $fh or from a string $str and convert them to an integer, in MSB order.
my $integer = bytes2int_lsb($fh, $n);
my $integer = bytes2int_lsb($str, $n);
Read $n bytes from a file-handle $fh or from a string $str and convert them to an integer, in LSB order.
my $symbols = string2symbols($string)
Returns an array-ref of code points, given a string. Equivalent to [unpack('C*', $string)].
my $string = symbols2string(\@symbols)
Returns a string, given an array-ref of code points. Equivalent to pack('C*', @$symbols). Inverse of string2symbols().
my $string = read_null_terminated($fh)
Read a string from file-handle $fh that ends with a NULL character ("\0"). The terminator itself is consumed but not included in the returned string.
my $bitstring_enc = binary_vrl_encode($bitstring);
Given a string of 1s and 0s, returns back a bitstring of 1s and 0s encoded using variable run-length encoding.
my $bitstring = binary_vrl_decode($bitstring_enc);
Given an encoded bitstring, returned by binary_vrl_encode(), gives back the decoded string of 1s and 0s.
my $indices = bwt_sort($string);
my $indices = bwt_sort($string, $lookahead_len);
Low-level function that sorts the rotations of a given string using the Burrows-Wheeler Transform (BWT) algorithm.
It takes two parameters: $string, which is the input string to be transformed, and $lookahead_len (optional), representing the length of look-ahead during sorting, which trades memory for speed (O(n * $lookahead_len) space). By default, $lookahead_len is 128.
The function returns an array-ref of indices.
There is probably no need to call this function explicitly. Use bwt_encode() instead!
my $indices = bwt_sort_symbolic(\@symbols);
Low-level function that sorts the rotations of a sequence of symbolic elements using the Burrows-Wheeler Transform (BWT) algorithm. Unlike bwt_sort(), this uses O(n) space with no look-ahead parameter, and is somewhat slower for large inputs.
It takes a single parameter \@symbols, which represents the input sequence of symbolic elements. The function returns an array-ref of indices.
There is probably no need to call this function explicitly. Use bwt_encode_symbolic() instead!
my $dict = huffman_from_freq(\%freq);
my ($dict, $rev_dict) = huffman_from_freq(\%freq);
Low-level function that constructs Huffman prefix codes, based on the frequency of symbols provided in a hash table.
It takes a single parameter, \%freq, representing the hash table where keys are symbols, and values are their corresponding frequencies (such as the hash-ref returned by frequencies()).
The function returns two values: $dict, which is the mapping of symbols to Huffman codes, and $rev_dict, which holds the reverse mapping of Huffman codes to symbols.
The prefix codes are in canonical form, as defined in RFC 1951 (Section 3.2.2).
my $dict = huffman_from_symbols(\@symbols);
my ($dict, $rev_dict) = huffman_from_symbols(\@symbols);
Low-level function that constructs Huffman prefix codes, given an array-ref of symbols. Equivalent to huffman_from_freq(frequencies(\@symbols)).
It takes a single parameter, \@symbols, from which it computes the frequency of each symbol and generates the corresponding Huffman prefix codes.
The function returns two values: $dict, which is the mapping of symbols to Huffman codes, and $rev_dict, which holds the reverse mapping of Huffman codes to symbols.
The prefix codes are in canonical form, as defined in RFC 1951 (Section 3.2.2).
my $dict = huffman_from_code_lengths(\@code_lengths);
my ($dict, $rev_dict) = huffman_from_code_lengths(\@code_lengths);
my $dict = huffman_from_code_lengths(\%code_lengths);
my ($dict, $rev_dict) = huffman_from_code_lengths(\%code_lengths);
Low-level function that constructs a dictionary of canonical prefix codes as defined in RFC 1951 (Section 3.2.2), given an array-ref of code lengths or a hash-ref of symbol => length values.
It takes a single parameter, \@code_lengths, where entry $i in the array corresponds to the code length for symbol $i (a length of 0 means the symbol is unused).
Similarly, when a hash-ref table is given, \%code_lengths, keys are the symbols and values are the code lengths. This variant is useful for large symbols, where an array indexed by symbol value would be wastefully sparse.
In list context, the function returns two values: $dict, which is the mapping of symbols to Huffman codes, and $rev_dict, which holds the reverse mapping of Huffman codes to symbols.
In scalar context, it returns only the $dict table.
This is the function used to reconstruct Huffman codes purely from the list of code lengths transmitted in a DEFLATE header, without needing to re-run the frequency-based algorithm.
my $bitstring = huffman_encode(\@symbols, $dict);
Low-level function that performs Huffman encoding on a sequence of symbols using a provided dictionary, returned by huffman_from_freq(), huffman_from_symbols(), or huffman_from_code_lengths().
It takes two parameters: \@symbols, representing the sequence of symbols to be encoded, and $dict, representing the Huffman dictionary mapping symbols to their corresponding Huffman codes.
The function returns a concatenated string of 1s and 0s, representing the Huffman-encoded sequence of symbols.
my $symbols = huffman_decode($bitstring, $rev_dict);
Low-level function that decodes a Huffman-encoded binary string into a sequence of symbols using a provided reverse dictionary.
It takes two parameters: $bitstring, representing the Huffman-encoded string of 1s and 0s, as returned by huffman_encode(), and $rev_dict, representing the reverse dictionary mapping Huffman codes to their corresponding symbols.
The function returns the decoded sequence of symbols as an array-ref.
Putting the last few functions together, without using create_huffman_entry()'s bookkeeping:
my @symbols = (65, 65, 66, 65, 67, 65, 66);
my ($dict, $rev_dict) = huffman_from_symbols(\@symbols);
my $bitstring = huffman_encode(\@symbols, $dict);
my $decoded = huffman_decode($bitstring, $rev_dict);
"@$decoded" eq "@symbols" or die "mismatch";
my ($literals, $distances, $lengths, $matches) = lz77_encode($string);
my ($literals, $distances, $lengths, $matches) = lz77_encode(\@symbols);
Low-level function that combines LZSS with ideas from the LZ4 method (interleaving literal runs and matches into parallel arrays, rather than a single flat token stream).
The function returns four values:
$literals # array-ref of uncompressed symbols
$distances # array-ref of back-reference distances
$lengths # array-ref of literal-run lengths
$matches # array-ref of match lengths
The output can be decoded with lz77_decode() and lz77_decode_symbolic(), respectively. lz77_encode_symbolic() is currently the same underlying function as lz77_encode(), since lz77_encode() already dispatches on whether it was given a string or an array-ref.
Ignores $LZ_MAX_LEN, always using unlimited match lengths (see "$LZ_MAX_LEN").
my $string = lz77_decode(\@literals, \@distances, \@lengths, \@matches);
my $symbols = lz77_decode_symbolic(\@literals, \@distances, \@lengths, \@matches);
Low-level function that performs decoding using the provided literals, distances, lengths and matches, returned by LZ77 encoding.
Inverse of lz77_encode() and lz77_encode_symbolic(), respectively.
# Standard version -- best compression
my ($literals, $distances, $lengths) = lzss_encode($data, %params);
my ($literals, $distances, $lengths) = lzss_encode(\@symbols, %params);
# Faster version -- lower chain length, still string-hash based
my ($literals, $distances, $lengths) = lzss_encode_fast($data, %params);
my ($literals, $distances, $lengths) = lzss_encode_fast(\@symbols, %params);
Low-level function that applies the LZSS (Lempel-Ziv-Storer-Szymanski) algorithm on the provided data. See "CHOOSING AN LZSS ENCODER" for how this compares to lzss_encode_fast() and lzss_encode_hash4().
The accepted %params are:
min_len => $LZ_MIN_LEN,
max_len => $LZ_MAX_LEN,
max_dist => $LZ_MAX_DIST,
max_chain_len => $LZ_MAX_CHAIN_LEN,
Any key omitted from %params falls back to the corresponding package variable (see "PACKAGE VARIABLES").
The function returns three values:
$literals # array-ref of uncompressed symbols
$distances # array-ref of back-reference distances
$lengths # array-ref of match lengths
The output can be decoded with lzss_decode() and lzss_decode_symbolic(), respectively.
my ($literals, $distances, $lengths) = lzss_encode_hash4($data, %params);
Low-level function that applies a fast LZSS compression algorithm using a fixed 4-byte hash window and an O(1) flat-array hash table.
Unlike lzss_encode_fast(), it uses integer hashing instead of string keys and skips indexing the bytes covered by a match, providing an LZ4-style performance trade-off (fastest of the three encoders, at some cost to compression ratio).
The accepted %params are:
max_len => $LZ_MAX_LEN,
max_dist => $LZ_MAX_DIST,
The minimum match length is fixed at 4 bytes and is not configurable. Symbolic-array input is not supported -- only plain byte strings.
The function returns three values:
$literals # array-ref of uncompressed bytes
$distances # array-ref of back-reference distances
$lengths # array-ref of match lengths
The output can be decoded with lzss_decode().
my $string = lzss_decode(\@literals, \@distances, \@lengths);
my $symbols = lzss_decode_symbolic(\@literals, \@distances, \@lengths);
Low-level function that decodes the LZSS encoding, using the provided literals, distances, and lengths of matched sub-strings.
Inverse of lzss_encode(), lzss_encode_fast(), and lzss_encode_hash4() (all three produce output in the same ($literals, $distances, $lengths) shape).
# Returns a binary string
my $string = deflate_encode(\@literals, \@distances, \@lengths);
my $string = deflate_encode(\@literals, \@distances, \@lengths, \&create_ac_entry);
Low-level function that encodes the results returned by lzss_encode(), lzss_encode_fast(), or lzss_encode_hash4(), using a DEFLATE-like approach, combined with Huffman coding by default (or the given $entropy_sub).
# Huffman decoding
my ($literals, $distances, $lengths) = deflate_decode($fh);
my ($literals, $distances, $lengths) = deflate_decode($string);
# Arithmetic decoding
my ($literals, $distances, $lengths) = deflate_decode($fh, \&decode_ac_entry);
my ($literals, $distances, $lengths) = deflate_decode($string, \&decode_ac_entry);
Inverse of deflate_encode().
my ($DISTANCE_SYMBOLS, $LENGTH_SYMBOLS, $LENGTH_INDICES) = make_deflate_tables($max_dist, $max_len);
Low-level function that returns a list of tables used in encoding the relative back-reference distances and lengths returned by lzss_encode() and lzss_encode_fast().
When no arguments are provided:
$max_dist = $Compression::Util::LZ_MAX_DIST
$max_len = $Compression::Util::LZ_MAX_LEN
There is no need to call this function explicitly. Use deflate_encode() instead!
my $index = find_deflate_index($value, $DISTANCE_SYMBOLS);
Low-level function that returns the index inside the DEFLATE tables (as returned by make_deflate_tables()) for a given value.
my $bt0_header = deflate_create_block_type_0_header($chunk);
Creates the header for a DEFLATE block of type 0 (uncompressed), as a bitstring, without including the block code number 00.
The length of the $chunk must not exceed 2^16 - 1.
To create a DEFLATE block of type 0, including the content, use:
my $block_type_0 = pack('b*', '00') . pack('b*', $bt0_header) . $chunk;
which can be recovered as:
open my $fh, '<:raw', \$block_type_0;
my ($buffer, $search_window) = ('', '');
my $chunk = deflate_extract_next_block($fh, \$buffer, \$search_window);
my $bitstring = deflate_create_block_type_1($literals, $distances, $lengths);
Creates a DEFLATE block of type 1 (fixed prefix-codes), as a bitstring, given the array-refs of literals, distances and lengths, returned by lzss_encode().
This type of block uses fixed prefix-codes and is pretty fast, at the cost of a somewhat lower compression ratio than type 2.
my $bitstring = deflate_create_block_type_2($literals, $distances, $lengths);
Creates a DEFLATE block of type 2 (dynamic prefix-codes), as a bitstring, given the array-refs of literals, distances and lengths, returned by lzss_encode().
This type of block uses dynamic prefix-codes (Huffman codes) and produces good compression ratio on most inputs.
my $data = deflate_extract_block_type_0($fh, \$buffer, \$search_window);
Given an input filehandle, it extracts a DEFLATE block of type 0 (uncompressed).
my ($buffer, $search_window) = ('', '');
my $block_type = bits2int_lsb($fh, 2, \$buffer);
$block_type == 0 or die "Not a block of type 0";
my $decoded_chunk = deflate_extract_block_type_0($fh, \$buffer, \$search_window);
my $data = deflate_extract_block_type_1($fh, \$buffer, \$search_window);
Given an input filehandle, a bitstring buffer and a search window, it extracts a DEFLATE block of type 1 (fixed prefix-codes).
my ($buffer, $search_window) = ('', '');
my $block_type = bits2int_lsb($fh, 2, \$buffer);
$block_type == 1 or die "Not a block of type 1";
my $decoded_chunk = deflate_extract_block_type_1($fh, \$buffer, \$search_window);
my $data = deflate_extract_block_type_2($fh, \$buffer, \$search_window);
Given an input filehandle, a bitstring buffer and a search window, it extracts a DEFLATE block of type 2 (dynamic prefix-codes).
my ($buffer, $search_window) = ('', '');
my $block_type = bits2int_lsb($fh, 2, \$buffer);
$block_type == 2 or die "Not a block of type 2";
my $decoded_chunk = deflate_extract_block_type_2($fh, \$buffer, \$search_window);
my $data = deflate_extract_next_block($fh, \$buffer, \$search_window);
Given an input filehandle, a bitstring buffer and a search window, it extracts the next DEFLATE block, dispatching automatically to deflate_extract_block_type_0(), deflate_extract_block_type_1(), or deflate_extract_block_type_2() based on the block-type bits. The next two bits in the input file-handle (or in the bitstring buffer) must contain the block-type number.
This is the function gzip_decompress() and zlib_decompress() use internally to walk a stream one block at a time; $search_window accumulates the last 32 KiB of decoded output across calls, as required for resolving back-references near a block boundary.
Each function can be exported individually, as:
use Compression::Util qw(bwt_compress);
By specifying the :all keyword, will export all the exportable functions:
use Compression::Util qw(:all);
The package variables ($VERBOSE, $LZ_MIN_LEN, $LZ_MAX_LEN, $LZ_MAX_DIST, $LZ_MAX_CHAIN_LEN, $LZ_MAX_CHAIN_WIDTH) are not included in :all; import them explicitly by name if needed (see "PACKAGE VARIABLES").
Nothing is exported by default.
The functions can be combined in various ways, easily creating novel compression methods, as illustrated in the following examples.
my $enc = lzss_compress($str, \&mrl_compress_symbolic);
my $dec = lzss_decompress($enc, \&mrl_decompress_symbolic);
my $enc = lz77_compress($str, \&obh_encode);
my $dec = lz77_decompress($enc, \&obh_decode);
my $enc = lzss_compress($str, \&bwt_compress_symbolic);
my $dec = lzss_decompress($enc, \&bwt_decompress_symbolic);
my $enc = bwt_compress($str, \&lzss_compress_symbolic);
my $dec = bwt_decompress($enc, \&lzss_decompress_symbolic);
my $enc = lzw_compress($str, \&fibonacci_encode);
my $dec = lzw_decompress($enc, \&fibonacci_decode);
my $enc = bwt_compress($str, sub ($s) { lz77_compress_symbolic($s, \&mrl_compress_symbolic) });
my $dec = bwt_decompress($enc, sub ($s) { lz77_decompress_symbolic($s, \&mrl_decompress_symbolic) });
Combining LZ77 + BWT compression + Fibonacci encoding + Huffman coding + OBH encoding + MRL compression:
# Compression
my $enc = do {
my ($literals, $distances, $lengths, $matches) = lz77_encode($str);
bwt_compress(symbols2string($literals))
. fibonacci_encode($lengths)
. create_huffman_entry($matches)
. obh_encode($distances, \&mrl_compress_symbolic);
};
# Decompression
my $dec = do {
open my $fh, '<:raw', \$enc;
my $literals = string2symbols(bwt_decompress($fh));
my $lengths = fibonacci_decode($fh);
my $matches = decode_huffman_entry($fh);
my $distances = obh_decode($fh, \&mrl_decompress_symbolic);
lz77_decode($literals, $distances, $lengths, $matches);
};
Delta coding is a good fit whenever the data is naturally a sequence of integers rather than text -- for example, a column of sorted database IDs, or timestamps:
my @ids = (1000, 1001, 1002, 1050, 1051, 1052, 2000);
my $enc = delta_encode(\@ids);
my $dec = delta_decode($enc);
"@$dec" eq "@ids" or die "mismatch";
Adaptive Arithmetic Coding avoids storing an explicit frequency table, which tends to pay off on short inputs, while plain Huffman coding (via create_huffman_entry()) is typically faster and does just as well once the input is large enough to amortize the table's size:
sub smart_compress ($data) {
length($data) < 1024
? ("A", create_adaptive_ac_entry(string2symbols($data)))
: ("H", create_huffman_entry(string2symbols($data)));
}
sub smart_decompress ($tag, $fh) {
my $symbols =
$tag eq 'A' ? decode_adaptive_ac_entry($fh)
: $tag eq 'H' ? decode_huffman_entry($fh)
: die "unknown tag: $tag";
symbols2string($symbols);
}
use Compression::Util qw(:all);
# Favor speed: shallower search, fewer chain positions remembered
{
local $Compression::Util::LZ_MAX_CHAIN_LEN = 4;
my $fast_enc = lzss_compress($data);
}
# Favor ratio: deeper search (default is 32)
{
local $Compression::Util::LZ_MAX_CHAIN_LEN = 128;
my $tight_enc = lzss_compress($data);
}
# Fastest possible: LZ4-style O(1) hashing
my $quick_enc = lzss_compress($data, \&create_huffman_entry, \&lzss_encode_hash4);
-
DEFLATE Compressed Data Format Specification https://datatracker.ietf.org/doc/html/rfc1951
-
GZIP file format specification https://datatracker.ietf.org/doc/html/rfc1952
-
ZLIB Compressed Data Format Specification https://datatracker.ietf.org/doc/html/rfc1950
-
BZIP2 Format Specification, by Joe Tsai: https://github.com/dsnet/compress/blob/master/doc/bzip2-format.pdf
-
LZ4 Frame format https://github.com/lz4/lz4/blob/dev/doc/lz4_Frame_format.md
-
LZ4 Block format https://github.com/lz4/lz4/blob/dev/doc/lz4_Block_format.md
-
Data Compression (Summer 2023) - Lecture 4 - The Unix 'compress' Program: https://youtube.com/watch?v=1cJL9Va80Pk
-
Data Compression (Summer 2023) - Lecture 5 - Basic Techniques: https://youtube.com/watch?v=TdFWb8mL5Gk
-
Data Compression (Summer 2023) - Lecture 11 - DEFLATE (gzip): https://youtube.com/watch?v=SJPvNi4HrWQ
-
Data Compression (Summer 2023) - Lecture 12 - The Burrows-Wheeler Transform (BWT): https://youtube.com/watch?v=rQ7wwh4HRZM
-
Data Compression (Summer 2023) - Lecture 13 - BZip2: https://youtube.com/watch?v=cvoZbBZ3M2A
-
Data Compression (Summer 2023) - Lecture 15 - Infinite Precision in Finite Bits: https://youtube.com/watch?v=EqKbT3QdtOI
-
Information Retrieval WS 17/18, Lecture 4: Compression, Codes, Entropy: https://youtube.com/watch?v=A_F94FV21Ek
-
COMP526 7-5 SS7.4 Run length encoding: https://youtube.com/watch?v=3jKLjmV1bL8
-
COMP526 Unit 7-6 2020-03-24 Compression - Move-to-front transform: https://youtube.com/watch?v=Q2pinaj3i9Y
-
Basic arithmetic coder in C++: https://github.com/billbird/arith32
Please report any bugs or feature requests to: https://github.com/trizen/Compression-Util.
-
Compress::Zlib, IO::Compress::Gzip, IO::Compress::Bzip2 -- bindings to the corresponding C libraries, generally much faster than this pure-Perl implementation for production use.
-
Compress::Raw::Zlib -- low-level bindings, if only speed matters and not portability.
Use Compression::Util when you want a pure-Perl implementation with no XS/C dependencies, when you want to inspect or hand-tune every stage of the pipeline, or for learning how these algorithms work.
Daniel "Trizen" Șuteu
Special thanks to professor Bill Bird for the awesome YouTube lectures on data compression.
This library is free software; you can redistribute it and/or modify it under the same terms as Perl itself, either Perl version 5.38.2 or, at your option, any later version of Perl 5 you may have available.