Next: , Up: (dir)

Lzip Manual

This manual is for Lzip (version 1.19, 13 April 2017).

Copyright © 2008-2017 Antonio Diaz Diaz.

This manual is free documentation: you have unlimited permission to copy, distribute and modify it.

Next: , Previous: Top, Up: Top

1 Introduction

Lzip is a lossless data compressor with a user interface similar to the one of gzip or bzip2. Lzip can compress about as fast as gzip (lzip -0), or compress most files more than bzip2 (lzip -9). Decompression speed is intermediate between gzip and bzip2. Lzip is better than gzip and bzip2 from a data recovery perspective.

The lzip file format is designed for data sharing and long-term archiving, taking into account both data integrity and decoder availability:

A nice feature of the lzip format is that a corrupt byte is easier to repair the nearer it is from the beginning of the file. Therefore, with the help of lziprecover, losing an entire archive just because of a corrupt byte near the beginning is a thing of the past.

The member trailer stores the 32-bit CRC of the original data, the size of the original data and the size of the member. These values, together with the end-of-stream marker, provide a 3 factor integrity checking which guarantees that the decompressed version of the data is identical to the original. This guards against corruption of the compressed data, and against undetected bugs in lzip (hopefully very unlikely). The chances of data corruption going undetected are microscopic. Be aware, though, that the check occurs upon decompression, so it can only tell you that something is wrong. It can't help you recover the original uncompressed data.

Lzip uses the same well-defined exit status values used by bzip2, which makes it safer than compressors returning ambiguous warning values (like gzip) when it is used as a back end for other programs like tar or zutils.

Lzip will automatically use the smallest possible dictionary size for each file without exceeding the given limit. Keep in mind that the decompression memory requirement is affected at compression time by the choice of dictionary size limit.

The amount of memory required for compression is about 1 or 2 times the dictionary size limit (1 if input file size is less than dictionary size limit, else 2) plus 9 times the dictionary size really used. The option '-0' is special and only requires about 1.5 MiB at most. The amount of memory required for decompression is about 46 kB larger than the dictionary size really used.

When compressing, lzip replaces every file given in the command line with a compressed version of itself, with the name "original_name.lz". When decompressing, lzip attempts to guess the name for the decompressed file from that of the compressed file as follows:

filename.lz becomes filename
filename.tlz becomes filename.tar
anyothername becomes anyothername.out

(De)compressing a file is much like copying or moving it; therefore lzip preserves the access and modification dates, permissions, and, when possible, ownership of the file just as "cp -p" does. (If the user ID or the group ID can't be duplicated, the file permission bits S_ISUID and S_ISGID are cleared).

Lzip is able to read from some types of non regular files if the '--stdout' option is specified.

If no file names are specified, lzip compresses (or decompresses) from standard input to standard output. In this case, lzip will decline to write compressed output to a terminal, as this would be entirely incomprehensible and therefore pointless.

Lzip will correctly decompress a file which is the concatenation of two or more compressed files. The result is the concatenation of the corresponding uncompressed files. Integrity testing of concatenated compressed files is also supported.

Lzip can produce multimember files, and lziprecover can safely recover the undamaged members in case of file damage. Lzip can also split the compressed output in volumes of a given size, even when reading from standard input. This allows the direct creation of multivolume compressed tar archives.

Lzip is able to compress and decompress streams of unlimited size by automatically creating multimember output. The members so created are large, about 2 PiB each.

LANGUAGE NOTE: Uncompressed = not compressed = plain data; it may never have been compressed. Decompressed is used to refer to data which have undergone the process of decompression.

Next: , Previous: Introduction, Up: Top

2 Invoking lzip

The format for running lzip is:

     lzip [options] [files]

'-' used as a file argument means standard input. It can be mixed with other files and is read just once, the first time it appears in the command line.

Lzip supports the following options:

Print an informative help message describing the options and exit.
Print the version number of lzip on the standard output and exit.

Exit with error status 2 if any remaining input is detected after decompressing the last member. Such remaining input is usually trailing garbage that can be safely ignored. See concat-example.
-b bytes
Set the member size limit to bytes. A small member size may degrade compression ratio, so use it only when needed. Valid values range from 100 kB to 2 PiB. Defaults to 2 PiB.
Compress or decompress to standard output; keep input files unchanged. If compressing several files, each file is compressed independently. This option is needed when reading from a named pipe (fifo) or from a device. Use it also to recover as much of the uncompressed data as possible when decompressing a corrupt file.
Decompress the specified file(s). If a file does not exist or can't be opened, lzip continues decompressing the rest of the files. If a file fails to decompress, lzip exits immediately without decompressing the rest of the files.
Force overwrite of output files.
Force re-compression of files whose name already has the '.lz' or '.tlz' suffix.
Keep (don't delete) input files during compression or decompression.
Print the uncompressed size, compressed size and percentage saved of the specified file(s). Trailing data are ignored. The values produced are correct even for multimember files. If more than one file is given, a final line containing the cumulative sizes is printed. With '-v', the dictionary size, the number of members in the file, and the amount of trailing data (if any) are also printed. With '-vv', the positions and sizes of each member in multimember files are also printed. '-lq' can be used to verify quickly (without decompressing) the structural integrity of the specified files. (Use '--test' to verify the data integrity). '-alq' additionally verifies that none of the specified files contain trailing data.
-m bytes
Set the match length limit in bytes. After a match this long is found, the search is finished. Valid values range from 5 to 273. Larger values usually give better compression ratios but longer compression times.
-o file
When reading from standard input and '--stdout' has not been specified, use 'file' as the virtual name of the uncompressed file. This produces a file named 'file' when decompressing, a file named 'file.lz' when compressing, and several files named 'file00001.lz', 'file00002.lz', etc, when compressing and splitting the output in volumes.
Quiet operation. Suppress all messages.
-s bytes
Set the dictionary size limit in bytes. Lzip will use the smallest possible dictionary size for each file without exceeding this limit. Valid values range from 4 KiB to 512 MiB. Values 12 to 29 are interpreted as powers of two, meaning 2^12 to 2^29 bytes. Note that dictionary sizes are quantized. If the specified size does not match one of the valid sizes, it will be rounded upwards by adding up to (bytes / 8) to it.

For maximum compression you should use a dictionary size limit as large as possible, but keep in mind that the decompression memory requirement is affected at compression time by the choice of dictionary size limit.

-S bytes
Split the compressed output into several volume files with names 'original_name00001.lz', 'original_name00002.lz', etc, and set the volume size limit to bytes. Each volume is a complete, maybe multimember, lzip file. A small volume size may degrade compression ratio, so use it only when needed. Valid values range from 100 kB to 4 EiB.
Check integrity of the specified file(s), but don't decompress them. This really performs a trial decompression and throws away the result. Use it together with '-v' to see information about the file(s). If a file fails the test, does not exist, can't be opened, or is a terminal, lzip continues checking the rest of the files.
Verbose mode.
When compressing, show the compression ratio for each file processed. A second '-v' shows the progress of compression.
When decompressing or testing, further -v's (up to 4) increase the verbosity level, showing status, compression ratio, dictionary size, trailer contents (CRC, data size, member size), and up to 6 bytes of trailing data (if any) both in hexadecimal and as a string of printable ASCII characters.
-0 .. -9
Set the compression parameters (dictionary size and match length limit) as shown in the table below. The default compression level is '-6'. Note that '-9' can be much slower than '-0'. These options have no effect when decompressing.

The bidimensional parameter space of LZMA can't be mapped to a linear scale optimal for all files. If your files are large, very repetitive, etc, you may need to use the '--dictionary-size' and '--match-length' options directly to achieve optimal performance.

Level Dictionary size Match length limit
-0 64 KiB 16 bytes
-1 1 MiB 5 bytes
-2 1.5 MiB 6 bytes
-3 2 MiB 8 bytes
-4 3 MiB 12 bytes
-5 4 MiB 20 bytes
-6 8 MiB 36 bytes
-7 16 MiB 68 bytes
-8 24 MiB 132 bytes
-9 32 MiB 273 bytes

Aliases for GNU gzip compatibility.

Numbers given as arguments to options may be followed by a multiplier and an optional 'B' for "byte".

Table of SI and binary prefixes (unit multipliers):

Prefix Value | Prefix Value
k kilobyte (10^3 = 1000) | Ki kibibyte (2^10 = 1024)
M megabyte (10^6) | Mi mebibyte (2^20)
G gigabyte (10^9) | Gi gibibyte (2^30)
T terabyte (10^12) | Ti tebibyte (2^40)
P petabyte (10^15) | Pi pebibyte (2^50)
E exabyte (10^18) | Ei exbibyte (2^60)
Z zettabyte (10^21) | Zi zebibyte (2^70)
Y yottabyte (10^24) | Yi yobibyte (2^80)

Exit status: 0 for a normal exit, 1 for environmental problems (file not found, invalid flags, I/O errors, etc), 2 to indicate a corrupt or invalid input file, 3 for an internal consistency error (eg, bug) which caused lzip to panic.

Next: , Previous: Invoking lzip, Up: Top

3 Design, development and testing of lzip

There are two ways of constructing a software design: One way is to make it so simple that there are obviously no deficiencies and the other way is to make it so complicated that there are no obvious deficiencies. The first method is far more difficult.
-- C.A.R. Hoare

Lzip has been designed, written and tested with great care to be the standard general-purpose compressor for unix-like systems. This chapter describes the lessons learned from previous compressors (gzip and bzip2), and their application to the design of lzip.

3.1 Format design

When gzip was designed in 1992, computers and operating systems were much less capable than they are today. Gzip tried to work around some of those limitations, like 8.3 file names, with additional fields in its file format.

Today those limitations have mostly disappeared, and the format of gzip has proved to be unnecessarily complicated. It includes fields that were never used, others that have lost their usefulness, and finally others that have become too limited.

Bzip2 was designed 5 years later, and its format is simpler than the one of gzip.

Probably the worst defect of the gzip format from the point of view of data safety is the variable size of its header. If the byte at offset 3 (flags) of a gzip member gets corrupted, it may become very difficult to recover the data, even if the compressed blocks are intact, because it can't be known with certainty where the compressed blocks begin.

By contrast, the header of a lzip member has a fixed length of 6. The LZMA stream in a lzip member always starts at offset 6, making it trivial to recover the data even if the whole header becomes corrupt.

Bzip2 also provides a header of fixed length and marks the begin and end of each compressed block with six magic bytes, making it possible to find the compressed blocks even in case of file damage. But bzip2 does not store the size of each compressed block, as lzip does.

Lzip provides better data recovery capabilities than any other gzip-like compressor because its format has been designed from the beginning to be simple and safe. It also helps that the LZMA data stream as used by lzip is extraordinarily safe. It provides embedded error detection. Any distance larger than the dictionary size acts as a forbidden symbol, allowing the decompressor to detect the approximate position of errors, and leaving very little work for the check sequence (CRC and data sizes) in the detection of errors. Lzip is usually able to detect all posible bit-flips in the compressed data without resorting to the check sequence. It would be very difficult to write an automatic recovery tool like lziprecover for the gzip format. And, as far as I know, it has never been written.

Lzip, like gzip and bzip2, uses a CRC32 to check the integrity of the decompressed data because it provides more accurate error detection than CRC64 up to a compressed size of about 16 GiB, a size larger than that of most files. In the case of lzip, the additional detection capability of the decompressor reduces the probability of undetected errors more than a million times, making CRC32 more accurate than CRC64 up to about 20 PiB of compressed size.

The lzip format is designed for long-term archiving. Therefore it excludes any unneeded features that may interfere with the future extraction of the uncompressed data.

3.1.1 Gzip format (mis)features not present in lzip

'Multiple algorithms'
Gzip provides a CM (Compression Method) field that has never been used because it is a bad idea to begin with. New compression methods may require additional fields, making it impossible to implement new methods and, at the same time, keep the same format. This field does not solve the problem of format proliferation; it just makes the problem less obvious.
'Optional fields in header'
Unless special precautions are taken, optional fields are generally a bad idea because they produce a header of variable size. The gzip header has 2 fields that, in addition to being optional, are zero-terminated. This means that if any byte inside the field gets zeroed, or if the terminating zero gets altered, gzip won't be able to find neither the header CRC nor the compressed blocks.
'Optional CRC for the header'
Using an optional checksum for the header is not only a bad idea, it is an error; it may prevent the extraction of perfectly good data. For example, if the checksum is used and the bit enabling it is reset by a bit-flip, the header will appear to be intact (in spite of being corrupt) while the compressed blocks will appear to be totally unrecoverable (in spite of being intact). Very misleading indeed.

3.1.2 Lzip format improvements over gzip and bzip2

'64-bit size field'
Probably the most frequently reported shortcoming of the gzip format is that it only stores the least significant 32 bits of the uncompressed size. The size of any file larger than 4 GiB gets truncated.

Bzip2 does not store the uncompressed size of the file.

The lzip format provides a 64-bit field for the uncompressed size. Additionaly, lzip produces multimember output automatically when the size is too large for a single member, allowing for an unlimited uncompressed size.

'Distributed index'
The lzip format provides a distributed index that, among other things, helps plzip to decompress several times faster than pigz and helps lziprecover do its job. Neither the gzip format nor the bzip2 format do provide an index.

A distributed index is safer and more scalable than a monolithic index. The monolithic index introduces a single point of failure in the compressed file and may limit the number of members or the total uncompressed size.

3.2 Quality of implementation

'Accurate and robust error detection'
The lzip format provides 3 factor integrity checking and the decompressors report mismatches in each factor separately. This way if just one byte in one factor fails but the other two factors match the data, it probably means that the data are intact and the corruption just affects the mismatching factor (CRC or data size) in the check sequence.
'Multiple implementations'
Just like the lzip format provides 3 factor protection against undetected data corruption, the development methodology of the lzip family of compressors provides 3 factor protection against undetected programming errors.

Three related but independent compressor implementations, lzip, clzip and minilzip/lzlib, are developed concurrently. Every stable release of any of them is subjected to a hundred hours of intensive testing to verify that it produces identical output to the other two. This guarantees that all three implement the same algorithm, and makes it unlikely that any of them may contain serious undiscovered errors. In fact, no errors have been discovered in lzip since 2009.

Additionally, the three implementations have been extensively tested with unzcrash, valgrind and 'american fuzzy lop' without finding a single vulnerability or false negative.

'Dictionary size'
Lzip automatically uses the smallest possible dictionary size for each file. In addition to reducing the amount of memory required for decompression, this feature also minimizes the probability of being affected by RAM errors during compression.
'Exit status'
Returning a warning status of 2 is a design flaw of compress that leaked into the design of gzip. Both bzip2 and lzip are free from this flaw.

Next: , Previous: Quality assurance, Up: Top

4 File format

Perfection is reached, not when there is no longer anything to add, but when there is no longer anything to take away.
-- Antoine de Saint-Exupery

In the diagram below, a box like this:
|   | <-- the vertical bars might be missing

represents one byte; a box like this:

|              |

represents a variable number of bytes.

A lzip file consists of a series of "members" (compressed data sets). The members simply appear one after another in the file, with no additional information before, between, or after them.

Each member has the following structure:

| ID string | VN | DS | LZMA stream | CRC32 |   Data size   |  Member size  |

All multibyte values are stored in little endian order.

'ID string (the "magic" bytes)'
A four byte string, identifying the lzip format, with the value "LZIP" (0x4C, 0x5A, 0x49, 0x50).
'VN (version number, 1 byte)'
Just in case something needs to be modified in the future. 1 for now.
'DS (coded dictionary size, 1 byte)'
The dictionary size is calculated by taking a power of 2 (the base size) and substracting from it a fraction between 0/16 and 7/16 of the base size.
Bits 4-0 contain the base 2 logarithm of the base size (12 to 29).
Bits 7-5 contain the numerator of the fraction (0 to 7) to substract from the base size to obtain the dictionary size.
Example: 0xD3 = 2^19 - 6 * 2^15 = 512 KiB - 6 * 32 KiB = 320 KiB
Valid values for dictionary size range from 4 KiB to 512 MiB.
'LZMA stream'
The LZMA stream, finished by an end of stream marker. Uses default values for encoder properties. See Stream format, for a complete description.
'CRC32 (4 bytes)'
CRC of the uncompressed original data.
'Data size (8 bytes)'
Size of the uncompressed original data.
'Member size (8 bytes)'
Total size of the member, including header and trailer. This field acts as a distributed index, allows the verification of stream integrity, and facilitates safe recovery of undamaged members from multimember files.

Next: , Previous: File format, Up: Top

5 Algorithm

In spite of its name (Lempel-Ziv-Markov chain-Algorithm), LZMA is not a concrete algorithm; it is more like "any algorithm using the LZMA coding scheme". For example, the option '-0' of lzip uses the scheme in almost the simplest way possible; issuing the longest match it can find, or a literal byte if it can't find a match. Inversely, a much more elaborated way of finding coding sequences of minimum size than the one currently used by lzip could be developed, and the resulting sequence could also be coded using the LZMA coding scheme.

Lzip currently implements two variants of the LZMA algorithm; fast (used by option '-0') and normal (used by all other compression levels).

The high compression of LZMA comes from combining two basic, well-proven compression ideas: sliding dictionaries (LZ77/78) and markov models (the thing used by every compression algorithm that uses a range encoder or similar order-0 entropy coder as its last stage) with segregation of contexts according to what the bits are used for.

Lzip is a two stage compressor. The first stage is a Lempel-Ziv coder, which reduces redundancy by translating chunks of data to their corresponding distance-length pairs. The second stage is a range encoder that uses a different probability model for each type of data; distances, lengths, literal bytes, etc.

Here is how it works, step by step:

1) The member header is written to the output stream.

2) The first byte is coded literally, because there are no previous bytes to which the match finder can refer to.

3) The main encoder advances to the next byte in the input data and calls the match finder.

4) The match finder fills an array with the minimum distances before the current byte where a match of a given length can be found.

5) Go back to step 3 until a sequence (formed of pairs, repeated distances and literal bytes) of minimum price has been formed. Where the price represents the number of output bits produced.

6) The range encoder encodes the sequence produced by the main encoder and sends the produced bytes to the output stream.

7) Go back to step 3 until the input data are finished or until the member or volume size limits are reached.

8) The range encoder is flushed.

9) The member trailer is written to the output stream.

10) If there are more data to compress, go back to step 1.

The ideas embodied in lzip are due to (at least) the following people: Abraham Lempel and Jacob Ziv (for the LZ algorithm), Andrey Markov (for the definition of Markov chains), G.N.N. Martin (for the definition of range encoding), Igor Pavlov (for putting all the above together in LZMA), and Julian Seward (for bzip2's CLI).

Next: , Previous: Algorithm, Up: Top

6 Format of the LZMA stream in lzip files

The LZMA algorithm has three parameters, called "special LZMA properties", to adjust it for some kinds of binary data. These parameters are; 'literal_context_bits' (with a default value of 3), 'literal_pos_state_bits' (with a default value of 0), and 'pos_state_bits' (with a default value of 2). As a general purpose compressor, lzip only uses the default values for these parameters. In particular 'literal_pos_state_bits' has been optimized away and does not even appear in the code.

Lzip also finishes the LZMA stream with an "End Of Stream" marker (the distance-length pair 0xFFFFFFFFU, 2), which in conjunction with the "member size" field in the member trailer allows the verification of stream integrity. The LZMA stream in lzip files always has these two features (default properties and EOS marker) and is referred to in this document as LZMA-302eos or LZMA-lzip.

The second stage of LZMA is a range encoder that uses a different probability model for each type of symbol; distances, lengths, literal bytes, etc. Range encoding conceptually encodes all the symbols of the message into one number. Unlike Huffman coding, which assigns to each symbol a bit-pattern and concatenates all the bit-patterns together, range encoding can compress one symbol to less than one bit. Therefore the compressed data produced by a range encoder can't be split in pieces that could be individually described.

It seems that the only way of describing the LZMA-302eos stream is describing the algorithm that decodes it. And given the many details about the range decoder that need to be described accurately, the source code of a real decoder seems the only appropriate reference to use.

What follows is a description of the decoding algorithm for LZMA-302eos streams using as reference the source code of "lzd", an educational decompressor for lzip files which can be downloaded from the lzip download directory. The source code of lzd is included in appendix A. See Reference source code.

6.1 What is coded

The LZMA stream includes literals, matches and repeated matches (matches reusing a recently used distance). There are 7 different coding sequences:

Bit sequence Name Description
0 + byte literal literal byte
1 + 0 + len + dis match distance-length pair
1 + 1 + 0 + 0 shortrep 1 byte match at latest used distance
1 + 1 + 0 + 1 + len rep0 len bytes match at latest used distance
1 + 1 + 1 + 0 + len rep1 len bytes match at second latest used distance
1 + 1 + 1 + 1 + 0 + len rep2 len bytes match at third latest used distance
1 + 1 + 1 + 1 + 1 + len rep3 len bytes match at fourth latest used distance

In the following tables, multibit sequences are coded in normal order, from MSB to LSB, except where noted otherwise.

Lengths (the 'len' in the table above) are coded as follows:

Bit sequence Description
0 + 3 bits lengths from 2 to 9
1 + 0 + 3 bits lengths from 10 to 17
1 + 1 + 8 bits lengths from 18 to 273

The coding of distances is a little more complicated, so I'll begin explaining a simpler version of the encoding.

Imagine you need to code a number from 0 to 2^32 - 1, and you want to do it in a way that produces shorter codes for the smaller numbers. You may first send the position of the most significant bit that is set to 1, which you may find by making a bit scan from the left (from the MSB). A position of 0 means that the number is 0 (no bit is set), 1 means the LSB is the first bit set (the number is 1), and 32 means the MSB is set (i.e., the number is >= 0x80000000). Let's call this bit position a "slot". Then, if slot is > 1, you send the remaining slot - 1 bits. Let's call these bits "direct_bits" because they are coded directly by value instead of indirectly by position.

The inconvenient of this simple method is that it needs 6 bits to code the slot, but it just uses 33 of the 64 possible values, wasting almost half of the codes.

The intelligent trick of LZMA is that it encodes the position of the most significant bit set, along with the value of the next bit, in the same 6 bits that would take to encode the position alone. This seems to need 66 slots (2 * position + next_bit), but for slots 0 and 1 there is no next bit, so the number of needed slots is 64 (0 to 63).

The 6 bits representing this "slot number" are then context-coded. If the distance is >= 4, the remaining bits are coded as follows. 'direct_bits' is the amount of remaining bits (from 0 to 30) needed to form a complete distance, and is calculated as (slot >> 1) - 1. If a distance needs 6 or more direct_bits, the last 4 bits are coded separately. The last piece (all the direct_bits for distances 4 to 127 or the last 4 bits for distances >= 128) is context-coded in reverse order (from LSB to MSB). For distances >= 128, the 'direct_bits - 4' part is coded with fixed 0.5 probability.

Bit sequence Description
slot distances from 0 to 3
slot + direct_bits distances from 4 to 127
slot + (direct_bits - 4) + 4 bits distances from 128 to 2^32 - 1

6.2 The coding contexts

These contexts ('Bit_model' in the source), are integers or arrays of integers representing the probability of the corresponding bit being 0.

The indices used in these arrays are:

A state machine ('State' in the source) with 12 states (0 to 11), coding the latest 2 to 4 types of sequences processed. The initial state is 0.
Value of the 2 least significant bits of the current position in the decoded data.
Value of the 3 most significant bits of the latest byte decoded.
Coded value of length (length - 2), with a maximum of 3. The resulting value is in the range 0 to 3.

In the following table, '!literal' is any sequence except a literal byte. 'rep' is any one of 'rep0', 'rep1', 'rep2' or 'rep3'. The types of previous sequences corresponding to each state are:

State Types of previous sequences
0 literal, literal, literal
1 match, literal, literal
2 rep or (!literal, shortrep), literal, literal
3 literal, shortrep, literal, literal
4 match, literal
5 rep or (!literal, shortrep), literal
6 literal, shortrep, literal
7 literal, match
8 literal, rep
9 literal, shortrep
10 !literal, match
11 !literal, (rep or shortrep)

The contexts for decoding the type of coding sequence are:

Name Indices Used when
bm_match state, pos_state sequence start
bm_rep state after sequence 1
bm_rep0 state after sequence 11
bm_rep1 state after sequence 111
bm_rep2 state after sequence 1111
bm_len state, pos_state after sequence 110

The contexts for decoding distances are:

Name Indices Used when
bm_dis_slot len_state, bit tree distance start
bm_dis reverse bit tree after slots 4 to 13
bm_align reverse bit tree for distances >= 128, after fixed probability bits

There are two separate sets of contexts for lengths ('Len_model' in the source). One for normal matches, the other for repeated matches. The contexts in each Len_model are (see 'decode_len' in the source):

Name Indices Used when
choice1 none length start
choice2 none after sequence 1
bm_low pos_state, bit tree after sequence 0
bm_mid pos_state, bit tree after sequence 10
bm_high bit tree after sequence 11

The context array 'bm_literal' is special. In principle it acts as a normal bit tree context, the one selected by 'literal_state'. But if the previous decoded byte was not a literal, two other bit tree contexts are used depending on the value of each bit in 'match_byte' (the byte at the latest used distance), until a bit is decoded that is different from its corresponding bit in 'match_byte'. After the first difference is found, the rest of the byte is decoded using the normal bit tree context. (See 'decode_matched' in the source).

6.3 The range decoder

The LZMA stream is consumed one byte at a time by the range decoder. (See 'normalize' in the source). Every byte consumed produces a variable number of decoded bits, depending on how well these bits agree with their context. (See 'decode_bit' in the source).

The range decoder state consists of two unsigned 32-bit variables; range (representing the most significant part of the range size not yet decoded), and code (representing the current point within range). range is initialized to (2^32 - 1), and code is initialized to 0.

The range encoder produces a first 0 byte that must be ignored by the range decoder. This is done by shifting 5 bytes in the initialization of code instead of 4. (See the 'Range_decoder' constructor in the source).

6.4 Decoding the LZMA stream

After decoding the member header and obtaining the dictionary size, the range decoder is initialized and then the LZMA decoder enters a loop (See 'decode_member' in the source) where it invokes the range decoder with the appropriate contexts to decode the different coding sequences (matches, repeated matches, and literal bytes), until the "End Of Stream" marker is decoded.

Next: , Previous: Stream format, Up: Top

7 Extra data appended to the file

Sometimes extra data are found appended to a lzip file after the last member. Such trailing data may be:

Trailing data are in no way part of the lzip file format, but tools reading lzip files are expected to behave as correctly and usefully as possible in the presence of trailing data.

Trailing data can be safely ignored in most cases. In some cases, like that of user-added data, they are expected to be ignored. In those cases where a file containing trailing data must be rejected, the option '--trailing-error' can be used. See --trailing-error.

Next: , Previous: Trailing data, Up: Top

8 A small tutorial with examples

WARNING! Even if lzip is bug-free, other causes may result in a corrupt compressed file (bugs in the system libraries, memory errors, etc). Therefore, if the data you are going to compress are important, give the '--keep' option to lzip and don't remove the original file until you verify the compressed file with a command like 'lzip -cd file.lz | cmp file -'.

Example 1: Replace a regular file with its compressed version 'file.lz' and show the compression ratio.
     lzip -v file

Example 2: Like example 1 but the created 'file.lz' is multimember with a member size of 1 MiB. The compression ratio is not shown.
     lzip -b 1MiB file

Example 3: Restore a regular file from its compressed version 'file.lz'. If the operation is successful, 'file.lz' is removed.
     lzip -d file.lz

Example 4: Verify the integrity of the compressed file 'file.lz' and show status.
     lzip -tv file.lz

Example 5: Compress a whole device in /dev/sdc and send the output to 'file.lz'.
     lzip -c /dev/sdc > file.lz

Example 6: The right way of concatenating the decompressed output of two or more compressed files. See Trailing data.
     Don't do this
       cat file1.lz file2.lz file3.lz | lzip -d
     Do this instead
       lzip -cd file1.lz file2.lz file3.lz

Example 7: Decompress 'file.lz' partially until 10 KiB of decompressed data are produced.
     lzip -cd file.lz | dd bs=1024 count=10

Example 8: Decompress 'file.lz' partially from decompressed byte 10000 to decompressed byte 15000 (5000 bytes are produced).
     lzip -cd file.lz | dd bs=1000 skip=10 count=5

Example 9: Create a multivolume compressed tar archive with a volume size of 1440 KiB.
     tar -c some_directory | lzip -S 1440KiB -o volume_name

Example 10: Extract a multivolume compressed tar archive.
     lzip -cd volume_name*.lz | tar -xf -

Example 11: Create a multivolume compressed backup of a large database file with a volume size of 650 MB, where each volume is a multimember file with a member size of 32 MiB.
     lzip -b 32MiB -S 650MB big_db

Next: , Previous: Examples, Up: Top

9 Reporting bugs

There are probably bugs in lzip. There are certainly errors and omissions in this manual. If you report them, they will get fixed. If you don't, no one will ever know about them and they will remain unfixed for all eternity, if not longer.

If you find a bug in lzip, please send electronic mail to Include the version number, which you can find by running lzip --version.

Next: , Previous: Problems, Up: Top

Appendix A Reference source code

/*  Lzd - Educational decompressor for the lzip format
    Copyright (C) 2013-2017 Antonio Diaz Diaz.

    This program is free software. Redistribution and use in source and
    binary forms, with or without modification, are permitted provided
    that the following conditions are met:

    1. Redistributions of source code must retain the above copyright
    notice, this list of conditions and the following disclaimer.

    2. Redistributions in binary form must reproduce the above copyright
    notice, this list of conditions and the following disclaimer in the
    documentation and/or other materials provided with the distribution.

    This program is distributed in the hope that it will be useful,
    but WITHOUT ANY WARRANTY; without even the implied warranty of
    Exit status: 0 for a normal exit, 1 for environmental problems
    (file not found, invalid flags, I/O errors, etc), 2 to indicate a
    corrupt or invalid input file.

#include <algorithm>
#include <cerrno>
#include <cstdio>
#include <cstdlib>
#include <cstring>
#include <stdint.h>
#include <unistd.h>
#if defined(__MSVCRT__) || defined(__OS2__) || defined(_MSC_VER)
#include <fcntl.h>
#include <io.h>

class State
  int st;

  enum { states = 12 };
  State() : st( 0 ) {}
  int operator()() const { return st; }
  bool is_char() const { return st < 7; }

  void set_char()
    static const int next[states] = { 0, 0, 0, 0, 1, 2, 3, 4, 5, 6, 4, 5 };
    st = next[st];
  void set_match()     { st = ( st < 7 ) ? 7 : 10; }
  void set_rep()       { st = ( st < 7 ) ? 8 : 11; }
  void set_short_rep() { st = ( st < 7 ) ? 9 : 11; }

enum {
  min_dictionary_size = 1 << 12,
  max_dictionary_size = 1 << 29,
  literal_context_bits = 3,
  literal_pos_state_bits = 0,				// not used
  pos_state_bits = 2,
  pos_states = 1 << pos_state_bits,
  pos_state_mask = pos_states - 1,

  len_states = 4,
  dis_slot_bits = 6,
  start_dis_model = 4,
  end_dis_model = 14,
  modeled_distances = 1 << (end_dis_model / 2),		// 128
  dis_align_bits = 4,
  dis_align_size = 1 << dis_align_bits,

  len_low_bits = 3,
  len_mid_bits = 3,
  len_high_bits = 8,
  len_low_symbols = 1 << len_low_bits,
  len_mid_symbols = 1 << len_mid_bits,
  len_high_symbols = 1 << len_high_bits,
  max_len_symbols = len_low_symbols + len_mid_symbols + len_high_symbols,

  min_match_len = 2,					// must be 2

  bit_model_move_bits = 5,
  bit_model_total_bits = 11,
  bit_model_total = 1 << bit_model_total_bits };

struct Bit_model
  int probability;
  Bit_model() : probability( bit_model_total / 2 ) {}

struct Len_model
  Bit_model choice1;
  Bit_model choice2;
  Bit_model bm_low[pos_states][len_low_symbols];
  Bit_model bm_mid[pos_states][len_mid_symbols];
  Bit_model bm_high[len_high_symbols];

class CRC32
  uint32_t data[256];		// Table of CRCs of all 8-bit messages.

    for( unsigned n = 0; n < 256; ++n )
      unsigned c = n;
      for( int k = 0; k < 8; ++k )
        { if( c & 1 ) c = 0xEDB88320U ^ ( c >> 1 ); else c >>= 1; }
      data[n] = c;

  void update_buf( uint32_t & crc, const uint8_t * const buffer,
                   const int size ) const
    for( int i = 0; i < size; ++i )
      crc = data[(crc^buffer[i])&0xFF] ^ ( crc >> 8 );

const CRC32 crc32;

typedef uint8_t File_header[6];	// 0-3 magic, 4 version, 5 coded_dict_size

typedef uint8_t File_trailer[20];
			//  0-3  CRC32 of the uncompressed data
			//  4-11 size of the uncompressed data
			// 12-19 member size including header and trailer

class Range_decoder
  uint32_t code;
  uint32_t range;

  Range_decoder() : code( 0 ), range( 0xFFFFFFFFU )
    for( int i = 0; i < 5; ++i ) code = (code << 8) | get_byte();

  uint8_t get_byte() { return std::getc( stdin ); }

  unsigned decode( const int num_bits )
    unsigned symbol = 0;
    for( int i = num_bits; i > 0; --i )
      range >>= 1;
      symbol <<= 1;
      if( code >= range ) { code -= range; symbol |= 1; }
      if( range <= 0x00FFFFFFU )			// normalize
        { range <<= 8; code = (code << 8) | get_byte(); }
    return symbol;

  unsigned decode_bit( Bit_model & bm )
    unsigned symbol;
    const uint32_t bound = ( range >> bit_model_total_bits ) * bm.probability;
    if( code < bound )
      range = bound;
      bm.probability += (bit_model_total - bm.probability) >> bit_model_move_bits;
      symbol = 0;
      range -= bound;
      code -= bound;
      bm.probability -= bm.probability >> bit_model_move_bits;
      symbol = 1;
    if( range <= 0x00FFFFFFU )				// normalize
      { range <<= 8; code = (code << 8) | get_byte(); }
    return symbol;

  unsigned decode_tree( Bit_model bm[], const int num_bits )
    unsigned symbol = 1;
    for( int i = 0; i < num_bits; ++i )
      symbol = ( symbol << 1 ) | decode_bit( bm[symbol] );
    return symbol - (1 << num_bits);

  unsigned decode_tree_reversed( Bit_model bm[], const int num_bits )
    unsigned symbol = decode_tree( bm, num_bits );
    unsigned reversed_symbol = 0;
    for( int i = 0; i < num_bits; ++i )
      reversed_symbol = ( reversed_symbol << 1 ) | ( symbol & 1 );
      symbol >>= 1;
    return reversed_symbol;

  unsigned decode_matched( Bit_model bm[], const unsigned match_byte )
    unsigned symbol = 1;
    for( int i = 7; i >= 0; --i )
      const unsigned match_bit = ( match_byte >> i ) & 1;
      const unsigned bit = decode_bit( bm[symbol+(match_bit<<8)+0x100] );
      symbol = ( symbol << 1 ) | bit;
      if( match_bit != bit )
        while( symbol < 0x100 )
          symbol = ( symbol << 1 ) | decode_bit( bm[symbol] );
    return symbol & 0xFF;

  unsigned decode_len( Len_model & lm, const int pos_state )
    if( decode_bit( lm.choice1 ) == 0 )
      return decode_tree( lm.bm_low[pos_state], len_low_bits );
    if( decode_bit( lm.choice2 ) == 0 )
      return len_low_symbols +
             decode_tree( lm.bm_mid[pos_state], len_mid_bits );
    return len_low_symbols + len_mid_symbols +
           decode_tree( lm.bm_high, len_high_bits );

class LZ_decoder
  unsigned long long partial_data_pos;
  Range_decoder rdec;
  const unsigned dictionary_size;
  uint8_t * const buffer;	// output buffer
  unsigned pos;			// current pos in buffer
  unsigned stream_pos;		// first byte not yet written to stdout
  uint32_t crc_;
  bool pos_wrapped;

  void flush_data();

  uint8_t peek( const unsigned distance ) const
    if( pos > distance ) return buffer[pos - distance - 1];
    if( pos_wrapped ) return buffer[dictionary_size + pos - distance - 1];
    return 0;			// prev_byte of first byte

  void put_byte( const uint8_t b )
    buffer[pos] = b;
    if( ++pos >= dictionary_size ) flush_data();

  explicit LZ_decoder( const unsigned dict_size )
    partial_data_pos( 0 ),
    dictionary_size( dict_size ),
    buffer( new uint8_t[dictionary_size] ),
    pos( 0 ),
    stream_pos( 0 ),
    crc_( 0xFFFFFFFFU ),
    pos_wrapped( false )

  ~LZ_decoder() { delete[] buffer; }

  unsigned crc() const { return crc_ ^ 0xFFFFFFFFU; }
  unsigned long long data_position() const { return partial_data_pos + pos; }

  bool decode_member();

void LZ_decoder::flush_data()
  if( pos > stream_pos )
    const unsigned size = pos - stream_pos;
    crc32.update_buf( crc_, buffer + stream_pos, size );
    errno = 0;
    if( std::fwrite( buffer + stream_pos, 1, size, stdout ) != size )
      { std::fprintf( stderr, "Write error: %s\n", std::strerror( errno ) );
        std::exit( 1 ); }
    if( pos >= dictionary_size )
      { partial_data_pos += pos; pos = 0; pos_wrapped = true; }
    stream_pos = pos;

bool LZ_decoder::decode_member()		// Returns false if error
  Bit_model bm_literal[1<<literal_context_bits][0x300];
  Bit_model bm_match[State::states][pos_states];
  Bit_model bm_rep[State::states];
  Bit_model bm_rep0[State::states];
  Bit_model bm_rep1[State::states];
  Bit_model bm_rep2[State::states];
  Bit_model bm_len[State::states][pos_states];
  Bit_model bm_dis_slot[len_states][1<<dis_slot_bits];
  Bit_model bm_dis[modeled_distances-end_dis_model+1];
  Bit_model bm_align[dis_align_size];
  Len_model match_len_model;
  Len_model rep_len_model;
  unsigned rep0 = 0;			// rep[0-3] latest four distances
  unsigned rep1 = 0;			// used for efficient coding of
  unsigned rep2 = 0;			// repeated distances
  unsigned rep3 = 0;
  State state;

  while( !std::feof( stdin ) && !std::ferror( stdin ) )
    const int pos_state = data_position() & pos_state_mask;
    if( rdec.decode_bit( bm_match[state()][pos_state] ) == 0 )	// 1st bit
      const uint8_t prev_byte = peek( 0 );
      const int literal_state = prev_byte >> ( 8 - literal_context_bits );
      Bit_model * const bm = bm_literal[literal_state];
      if( state.is_char() )
        put_byte( rdec.decode_tree( bm, 8 ) );
        put_byte( rdec.decode_matched( bm, peek( rep0 ) ) );
    else					// match or repeated match
      int len;
      if( rdec.decode_bit( bm_rep[state()] ) != 0 )		// 2nd bit
        if( rdec.decode_bit( bm_rep0[state()] ) == 0 )		// 3rd bit
          if( rdec.decode_bit( bm_len[state()][pos_state] ) == 0 ) // 4th bit
            { state.set_short_rep(); put_byte( peek( rep0 ) ); continue; }
          unsigned distance;
          if( rdec.decode_bit( bm_rep1[state()] ) == 0 )	// 4th bit
            distance = rep1;
            if( rdec.decode_bit( bm_rep2[state()] ) == 0 )	// 5th bit
              distance = rep2;
              { distance = rep3; rep3 = rep2; }
            rep2 = rep1;
          rep1 = rep0;
          rep0 = distance;
        len = min_match_len + rdec.decode_len( rep_len_model, pos_state );
      else					// match
        rep3 = rep2; rep2 = rep1; rep1 = rep0;
        len = min_match_len + rdec.decode_len( match_len_model, pos_state );
        const int len_state = std::min( len - min_match_len, len_states - 1 );
        rep0 = rdec.decode_tree( bm_dis_slot[len_state], dis_slot_bits );
        if( rep0 >= start_dis_model )
          const unsigned dis_slot = rep0;
          const int direct_bits = ( dis_slot >> 1 ) - 1;
          rep0 = ( 2 | ( dis_slot & 1 ) ) << direct_bits;
          if( dis_slot < end_dis_model )
            rep0 += rdec.decode_tree_reversed( bm_dis + ( rep0 - dis_slot ),
                                               direct_bits );
            rep0 += rdec.decode( direct_bits - dis_align_bits ) << dis_align_bits;
            rep0 += rdec.decode_tree_reversed( bm_align, dis_align_bits );
            if( rep0 == 0xFFFFFFFFU )		// marker found
              return ( len == min_match_len );	// End Of Stream marker
        if( rep0 >= dictionary_size || ( rep0 >= pos && !pos_wrapped ) )
          { flush_data(); return false; }
      for( int i = 0; i < len; ++i ) put_byte( peek( rep0 ) );
  return false;

int main( const int argc, const char * const argv[] )
  if( argc > 1 )
    std::printf( "Lzd %s - Educational decompressor for the lzip format.\n",
                 PROGVERSION );
    std::printf( "Study the source to learn how a lzip decompressor works.\n"
                 "See the lzip manual for an explanation of the code.\n"
                 "It is not safe to use lzd for any real work.\n"
                 "\nUsage: %s < file.lz > file\n", argv[0] );
    std::printf( "Lzd decompresses from standard input to standard output.\n"
                 "\nCopyright (C) 2017 Antonio Diaz Diaz.\n"
                 "This is free software: you are free to change and redistribute it.\n"
                 "There is NO WARRANTY, to the extent permitted by law.\n"
                 "Report bugs to\n"
                 "Lzd home page:\n" );
    return 0;

#if defined(__MSVCRT__) || defined(__OS2__) || defined(_MSC_VER)
  setmode( fileno( stdin ), O_BINARY );
  setmode( fileno( stdout ), O_BINARY );

  for( bool first_member = true; ; first_member = false )
    File_header header;				// verify header
    for( int i = 0; i < 6; ++i ) header[i] = std::getc( stdin );
    if( std::feof( stdin ) || std::memcmp( header, "LZIP\x01", 5 ) != 0 )
      if( first_member )
        { std::fputs( "Bad magic number (file not in lzip format).\n", stderr );
          return 2; }
    unsigned dict_size = 1 << ( header[5] & 0x1F );
    dict_size -= ( dict_size / 16 ) * ( ( header[5] >> 5 ) & 7 );
    if( dict_size < min_dictionary_size || dict_size > max_dictionary_size )
      { std::fputs( "Invalid dictionary size in member header.\n", stderr );
        return 2; }

    LZ_decoder decoder( dict_size );		// decode LZMA stream
    if( !decoder.decode_member() )
      { std::fputs( "Data error\n", stderr ); return 2; }

    File_trailer trailer;			// verify trailer
    for( int i = 0; i < 20; ++i ) trailer[i] = std::getc( stdin );
    unsigned crc = 0;
    for( int i = 3; i >= 0; --i ) { crc <<= 8; crc += trailer[i]; }
    unsigned long long data_size = 0;
    for( int i = 11; i >= 4; --i ) { data_size <<= 8; data_size += trailer[i]; }
    if( crc != decoder.crc() || data_size != decoder.data_position() )
      { std::fputs( "CRC error\n", stderr ); return 2; }

  if( std::fclose( stdout ) != 0 )
    { std::fprintf( stderr, "Can't close stdout: %s\n", std::strerror( errno ) );
      return 1; }
  return 0;

Previous: Reference source code, Up: Top

Concept index