Squiz-Box

A D library that handles compression / decompression and archiving.

TL;DR: See examples

In the Squiz-Box terminology, squiz refers to compression/decompression, while box refers to archiving. Squiz-Box API consists almost exclusively of range algorithms. squiz related algorithms map a range of bytes (range of byte chunks actually) to another range of bytes where one is the compressed stream of the other. box and unbox related algorithms map a range of file entries to/from a range of bytes.

Squiz-Box provides both compile time known structures and dynamic types over all algorithms.

Compression / decompression

Compression / decompression is provided by well-known libraries.

The range based design makes squizing suitable for streaming and for transforming data from any kind of source to any kind of destination.

Compression algorithms are represented by structs that share a common interface. Constructed objects from those structs carry parameters for compression / decompression and can instantiate a stream (class that derives SquizStream) that will carry the necessary state as well as input and output buffer.

To process these algorithms and streams as D ranges, you use the squiz function. The squiz function works for both compression and decompression and there are many helpers built upon it (deflate, inflate, compressXz, ...). See code examples below for usage.

Algorithms and formats

An algorithm refers to the compression algorithm properly, and format refers to a header and a trailer attached to the compressed data that gives information about decompression parameters and integrity checking. A raw format refers to data compressed without such header and trailer and can be used only in applications or archive that know how to decompress the stream by other means.

Algorithms	Library	Squiz structs	Compression helpers	Decompression Helpers	Available formats
Deflate	zlib	Deflate (compression), Inflate (decompression)	deflate, deflateGz, deflateRaw	inflate, inflateGz, inflateRaw	Zlib, Gzip, Raw
Bzip2	bzip2	CompressBzip2, DecompressBzip2	compressBzip2	decompressBzip2	Bzip2
LZMA (aka. LZMA2)	xz-utils	CompressLzma, DecompressLzma	compressXz, compressLzmaRaw	decompressXz, decompressLzmaRaw	Xz, Raw
LZMA1 (legacy compression)	xz-utils	CompressLzma, DecompressLzma			Xz, Lzma (legacy format), Raw
Zstandard	zstd	CompressZstd, DecompressZstd	compressZstd	decompressZstd	Zstandard

In addition, the LZMA1 and LZMA compression also support additional filters that transorm the data before the compression stage in order to increase the compression ratio:

• Delta
  • higher compression of repetitive binary data such as audio PCM, RGB...
• BCJ (Branch/Call/Jump)
  • higher compression of compiled executable
  • available for different architectures (X86, ARM, ...)

Archiving

Archiving is similar to the compression/decompression API: algorithms are described by structs that share a common interface and that you pass to the box and unbox function to process D ranges.

The box function generates a range of byte chunks from a range of entries, and the opposite is done by the unbox function. box consumes a range of BoxEntry, which is a class that can be derived (eg. FileBoxEntry, InfoBoxEntry). unbox returns a range of UnboxEntry, which has a byChunk method to get the data and an extractTo helper that extracts the data in the filesystem. See hereunder for code examples.

Archive Format	Struct	Box helper	Unbox helper
.tar	TarAlgo	boxTar	unboxTar
.tar.gz	TarGzAlgo	boxTarGz	unboxTarGz
.tar.bz2	TarBzip2Algo	boxTarBzip2	unboxTarBzip2
.tar.xz	TarXzAlgo	boxTarXz	unboxTarXz
.zip	ZipAlgo	boxZip	unboxZip

There is also WIP for 7z.

Squiz-Box will try to process the archive in a single pass whenever possible and will not allocate more memory than necessary. A consequence is that the ranges of UnboxEntry must be processed in the order they are in the archive. If one needs to store the entries in memory, some algorithms take a File or const(ubyte)[] parameter to describe the whole archive. Each UnboxEntry will reference the data source and seek as necessary. (e.g. unboxZip(File))

Archive update is not supported at this stage. It will be easily done in a single range expression once an implementation of BoxEntry that reads from UnboxEntry will be written (PR welcome).

Code examples

Compress data with zlib

import squiz_box;
import std.range;

const ubyte[] data = myDataToCompress();

auto algo = Deflate.init; // defaults to zlib format

only(data)              // InputRange of const(ubyte)[] (uncompressed)
  .squiz(algo)          // also InputRange of const(ubyte)[] (deflated)
  .sendOverNetwork();

// the following code is equivalent
only(data)
  .deflate()
  .sendOverNetwork();

Re-inflate data with zlib

import squiz_box;
import std.array;

const data = receiveFromNetwork() // InputRange of const(ubyte)[]
  .inflate()
  .join();                        // const(ubyte)[]

Create an archive from a directory

Zip a directory:

import squiz_box;

import std.algorithm;
import std.file;

const root = buildNormalizedPath(someDir);

dirEntries(root, SpanMode.breadth, false)
    .filter!(e => !e.isDir)
    .map!(e => fileEntry(e.name, root, null))    // range of FileBoxEntry
    .boxZip()                                    // range of bytes
    .writeBinaryFile("some-dir.zip");

// boxZip is identical to box(ZipAlgo.init)

Create a .tar.xz file from a directory (with little bit more control):

import squiz_box;

import std.algorithm;
import std.file;
import std.path;

const root = squizBoxDir;

// prefix all files path in the archive
// don't forget the trailing '/'!
const prefix = "squiz-box-12.5/";

const exclusion = [".git", ".dub", ".vscode", "libsquiz-box.a", "build"];

dirEntries(root, SpanMode.breadth, false)
    .filter!(e => !e.isDir)
    .filter!(e => !exclusion.any!(ex => e.name.canFind(ex)))
    .map!(e => fileEntry(e.name, root, prefix))
    .boxTarXz()
    .writeBinaryFile("squiz-box-12.5.tar.xz");

/// boxTarXz() is equivalent to boxTar().compressXz()

Extract an archive into a directory

import squiz_box;

const archive = "my-archive.tar.gz";
const extractionSite = "some-dir";

mkdir(extractionSite);

readBinaryFile(archive)
    .unboxTarGz()                         // range of UnboxEntry
    .each!(e => e.extractTo(extractionSite));

Download, list and extract archive

This examples uses requests to download an archive from the web, list the archive content and extract it with a single expression. (std.net.curl.byChunk would also work and woudn't require the const casting)

Thanks to D ranges laziness, the archive is extracted as the data download progresses. As such, it is possible to download and extract very large archives with minimal memory footprint (and without creating an intermediate file on disk).

import squiz_box;
import requests;

const url = "https://github.com/dlang/dmd/archive/master.tar.gz";
const dest = ".";

// Algorithm matched at runtime with url (using extension)
auto algo = boxAlgo(url);

size_t downloadSz;

auto rq = Request();
rq.useStreaming = true;
rq.get(url).receiveAsRange()
    .map!(c => cast(const(ubyte)[])c)             // type-casting to const is necessary
    .tee!(c => downloadSz += c.length)            // trace download size
    .unbox(algo)
    .tee!(e => writeln(buildPath(dest, e.path)))  // list archive content
    .each!(e => e.extractTo(dest));               // extract

Create archive, list and upload to web

This examples creates an archive and uses requests to upload it on the web. As in the previous example, the data is uploaded as the archive creation progresses.

import squiz_box;
import requests;

const postTo = "https://httpbin.org/post";
const fmt = ".tar.xz";
const src = "...";
const prefix;

size_t uploadSz;

// Algorithm matched at runtime (using extension)
auto algo = boxAlgo(fmt);

const exclusion = [".git", ".dub", ".vscode", "libsquiz-box.a", "build"];

auto archiveChunks = dirEntries(src, SpanMode.breadth, false)
    .filter!(e => !e.isDir)
    .filter!(e => !exclusion.any!(ex => e.name.canFind(ex)))
    .tee!(e => writeln(e.name))
    .map!(e => fileEntry(e.name, src, prefix))
    .box(algo)
    .tee!(c => uploadSz += c.length);

auto rq = Request();
auto resp = rq.post(postTo, archiveChunks, algo.mimetype);
enforce(resp.code < 300, format!"%s responded %s"(postTo, resp.code));

writefln!"POST %s - status %s (posted %s bytes)"(postTo, resp.code, uploadSz);

Full control over the streaming process

Sometimes, D ranges are not practical. Think of a receiver thread that receives data, compresses it and sends it over network with low latency. You will not wait to receive the full data before to start streaming, and a D range is probably not well suited to receive the data from a different thread. In that situation you can use the streaming API directly.

The following code gives the spirit of it.

import squiz_box;

// Zstandard is a good fit for low latency streaming
auto algo = CompressZstd.init;
auto stream = algo.initialize();

stream.input = nextDataChunk();
stream.output = buffer;

// something along this logic
while(true)
{
    const lastChunk = hasDataChunk() ? No.lastChunk : Yes.lastChunk;

    // algo.process will compress data from input to output.
    // along the way, stream.input and stream.output are reduced.
    // (e.g. stream.input = stream.input[processed .. $])
    // Depending on algorithm latency, it is possible to consume several mega-bytes of input
    // before starting to write any output.
    const streamEnded = algo.process(stream, lastChunk);

    if (streamEnded || stream.output.empty)
    {
        // send the filled buffer out and notify the stream that output space is available
        // (no need to zero-out the buffer)
        sendBufferOut(buffer[0 .. $ - stream.output.length]);
        stream.output = buffer;
    }

    if (stream.input.empty && hasDataChunk())
        // initialize or reset new input data
        stream.input = nextDataChunk();

    if (streamEnded)
        break;
}

// we can reset the stream and keep the allocated resources for a new round
algo.reset(stream);

// more streaming...

// finally we can release the resources
algo.end(stream);

Download, build, install

To use squiz-box, the easiest is to use the Dub package from the registry. On Linux, you will need liblzma, libbz2 and libzstd installed. On Windows, a compiled copy of these libraries is shipped with the package (only Windows 64 bit is supported).

Squiz-box is developped with Meson, which will build the C libraries if they are not found. If you want to use squiz-box in a Meson project, you should use it as a subproject.

To build, test on Linux:

meson builddir
cd builddir
ninja && ./squiz-test

To build test on Windows:

rem Visual Studio prompt is required (e.g. Windows Terminal)
meson builddir
cd builddir
ninja && squiz-test.exe