summaryrefslogtreecommitdiffstats
path: root/doc/lzip.texi
diff options
context:
space:
mode:
Diffstat (limited to '')
-rw-r--r--doc/lzip.texi1783
1 files changed, 1783 insertions, 0 deletions
diff --git a/doc/lzip.texi b/doc/lzip.texi
new file mode 100644
index 0000000..144b525
--- /dev/null
+++ b/doc/lzip.texi
@@ -0,0 +1,1783 @@
+\input texinfo @c -*-texinfo-*-
+@c %**start of header
+@setfilename lzip.info
+@documentencoding ISO-8859-15
+@settitle Lzip Manual
+@finalout
+@c %**end of header
+
+@set UPDATED 24 January 2022
+@set VERSION 1.23
+
+@dircategory Compression
+@direntry
+* Lzip: (lzip). LZMA lossless data compressor
+@end direntry
+
+
+@ifnothtml
+@titlepage
+@title Lzip
+@subtitle LZMA lossless data compressor
+@subtitle for Lzip version @value{VERSION}, @value{UPDATED}
+@author by Antonio Diaz Diaz
+
+@page
+@vskip 0pt plus 1filll
+@end titlepage
+
+@contents
+@end ifnothtml
+
+@ifnottex
+@node Top
+@top
+
+This manual is for Lzip (version @value{VERSION}, @value{UPDATED}).
+
+@menu
+* Introduction:: Purpose and features of lzip
+* Output:: Meaning of lzip's output
+* Invoking lzip:: Command line interface
+* Quality assurance:: Design, development, and testing of lzip
+* Algorithm:: How lzip compresses the data
+* File format:: Detailed format of the compressed file
+* Stream format:: Format of the LZMA stream in lzip files
+* Trailing data:: Extra data appended to the file
+* Examples:: A small tutorial with examples
+* Problems:: Reporting bugs
+* Reference source code:: Source code illustrating stream format
+* Concept index:: Index of concepts
+@end menu
+
+@sp 1
+Copyright @copyright{} 2008-2022 Antonio Diaz Diaz.
+
+This manual is free documentation: you have unlimited permission to copy,
+distribute, and modify it.
+@end ifnottex
+
+
+@node Introduction
+@chapter Introduction
+@cindex introduction
+
+@uref{http://www.nongnu.org/lzip/lzip.html,,Lzip}
+is a lossless data compressor with a user interface similar to the one
+of gzip or bzip2. Lzip uses a simplified form of the 'Lempel-Ziv-Markov
+chain-Algorithm' (LZMA) stream format and provides a 3 factor integrity
+checking to maximize interoperability and optimize safety. Lzip can compress
+about as fast as gzip @w{(lzip -0)} or compress most files more than bzip2
+@w{(lzip -9)}. Decompression speed is intermediate between gzip and bzip2.
+Lzip is better than gzip and bzip2 from a data recovery perspective. Lzip
+has been designed, written, and tested with great care to replace gzip and
+bzip2 as the standard general-purpose compressed format for unix-like
+systems.
+
+For compressing/decompressing large files on multiprocessor machines
+@uref{http://www.nongnu.org/lzip/manual/plzip_manual.html,,plzip} can be
+much faster than lzip at the cost of a slightly reduced compression ratio.
+@ifnothtml
+@xref{Top,plzip manual,,plzip}.
+@end ifnothtml
+
+For creation and manipulation of compressed tar archives
+@uref{http://www.nongnu.org/lzip/manual/tarlz_manual.html,,tarlz} can be more
+efficient than using tar and plzip because tarlz is able to keep the
+alignment between tar members and lzip members.
+@ifnothtml
+@xref{Top,tarlz manual,,tarlz}.
+@end ifnothtml
+
+The lzip file format is designed for data sharing and long-term archiving,
+taking into account both data integrity and decoder availability:
+
+@itemize @bullet
+@item
+The lzip format provides very safe integrity checking and some data
+recovery means. The program
+@uref{http://www.nongnu.org/lzip/manual/lziprecover_manual.html#Data-safety,,lziprecover}
+can repair bit flip errors (one of the most common forms of data corruption)
+in lzip files, and provides data recovery capabilities, including
+error-checked merging of damaged copies of a file.
+@ifnothtml
+@xref{Data safety,,,lziprecover}.
+@end ifnothtml
+
+@item
+The lzip format is as simple as possible (but not simpler). The lzip
+manual provides the source code of a simple decompressor along with a
+detailed explanation of how it works, so that with the only help of the
+lzip manual it would be possible for a digital archaeologist to extract
+the data from a lzip file long after quantum computers eventually
+render LZMA obsolete.
+
+@item
+Additionally the lzip reference implementation is copylefted, which
+guarantees that it will remain free forever.
+@end itemize
+
+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 for each file the largest dictionary size that
+does not exceed neither the file size nor the limit given. 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
+@samp{-0} is special and only requires about @w{1.5 MiB} at most. The
+amount of memory required for decompression is about @w{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:
+
+@multitable {anyothername} {becomes} {anyothername.out}
+@item filename.lz @tab becomes @tab filename
+@item filename.tlz @tab becomes @tab filename.tar
+@item anyothername @tab becomes @tab anyothername.out
+@end multitable
+
+(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 @w{@samp{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 either the
+option @samp{-c} or the option @samp{-o} is specified.
+
+Lzip will refuse to read compressed data from a terminal or write compressed
+data to a terminal, as this would be entirely incomprehensible and might
+leave the terminal in an abnormal state.
+
+Lzip will correctly decompress a file which is the concatenation of two or
+more compressed files. The result is the concatenation of the corresponding
+decompressed 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 @w{2 PiB} each.
+
+
+@node Output
+@chapter Meaning of lzip's output
+@cindex output
+
+The output of lzip looks like this:
+
+@example
+lzip -v foo
+ foo: 6.676:1, 14.98% ratio, 85.02% saved, 450560 in, 67493 out.
+
+lzip -tvvv foo.lz
+ foo.lz: 6.676:1, 14.98% ratio, 85.02% saved. 450560 out, 67493 in. ok
+@end example
+
+The meaning of each field is as follows:
+
+@table @code
+@item N:1
+The compression ratio @w{(uncompressed_size / compressed_size)}, shown as
+@w{N to 1}.
+
+@item ratio
+The inverse compression ratio @w{(compressed_size / uncompressed_size)},
+shown as a percentage. A decimal ratio is easily obtained by moving the
+decimal point two places to the left; @w{14.98% = 0.1498}.
+
+@item saved
+The space saved by compression @w{(1 - ratio)}, shown as a percentage.
+
+@item in
+Size of the input data. This is the uncompressed size when compressing, or
+the compressed size when decompressing or testing. Note that lzip always
+prints the uncompressed size before the compressed size when compressing,
+decompressing, testing, or listing.
+
+@item out
+Size of the output data. This is the compressed size when compressing, or
+the decompressed size when decompressing or testing.
+
+@end table
+
+When decompressing or testing at verbosity level 4 (-vvvv), the dictionary
+size used to compress the file and the CRC32 of the uncompressed data are
+also shown.
+
+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.
+
+
+@node Invoking lzip
+@chapter Invoking lzip
+@cindex invoking
+@cindex options
+@cindex usage
+@cindex version
+
+The format for running lzip is:
+
+@example
+lzip [@var{options}] [@var{files}]
+@end example
+
+@noindent
+If no file names are specified, lzip compresses (or decompresses) from
+standard input to standard output. A hyphen @samp{-} used as a @var{file}
+argument means standard input. It can be mixed with other @var{files} and is
+read just once, the first time it appears in the command line.
+
+lzip supports the following
+@uref{http://www.nongnu.org/arg-parser/manual/arg_parser_manual.html#Argument-syntax,,options}:
+@ifnothtml
+@xref{Argument syntax,,,arg_parser}.
+@end ifnothtml
+
+@table @code
+@item -h
+@itemx --help
+Print an informative help message describing the options and exit.
+
+@item -V
+@itemx --version
+Print the version number of lzip on the standard output and exit.
+This version number should be included in all bug reports.
+
+@anchor{--trailing-error}
+@item -a
+@itemx --trailing-error
+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. @xref{concat-example}.
+
+@item -b @var{bytes}
+@itemx --member-size=@var{bytes}
+When compressing, set the member size limit to @var{bytes}. It is advisable
+to keep members smaller than RAM size so that they can be repaired with
+lziprecover in case of corruption. A small member size may degrade
+compression ratio, so use it only when needed. Valid values range from
+@w{100 kB} to @w{2 PiB}. Defaults to @w{2 PiB}.
+
+@item -c
+@itemx --stdout
+Compress or decompress to standard output; keep input files unchanged. If
+compressing several files, each file is compressed independently. (The
+output consists of a sequence of independently compressed members). This
+option (or @samp{-o}) is needed when reading from a named pipe (fifo) or
+from a device. Use it also to recover as much of the decompressed data as
+possible when decompressing a corrupt file. @samp{-c} overrides @samp{-o}
+and @samp{-S}. @samp{-c} has no effect when testing or listing.
+
+@item -d
+@itemx --decompress
+Decompress the files specified. If a file does not exist, can't be opened,
+or the destination file already exists and @samp{--force} has not been
+specified, lzip continues decompressing the rest of the files and exits with
+error status 1. If a file fails to decompress, or is a terminal, lzip exits
+immediately with error status 2 without decompressing the rest of the files.
+A terminal is considered an uncompressed file, and therefore invalid.
+
+@item -f
+@itemx --force
+Force overwrite of output files.
+
+@item -F
+@itemx --recompress
+When compressing, force re-compression of files whose name already has
+the @samp{.lz} or @samp{.tlz} suffix.
+
+@item -k
+@itemx --keep
+Keep (don't delete) input files during compression or decompression.
+
+@item -l
+@itemx --list
+Print the uncompressed size, compressed size, and percentage saved of the
+files specified. 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 @samp{-v}, the dictionary
+size, the number of members in the file, and the amount of trailing data (if
+any) are also printed. With @samp{-vv}, the positions and sizes of each
+member in multimember files are also printed.
+
+If any file is damaged, does not exist, can't be opened, or is not regular,
+the final exit status will be @w{> 0}. @samp{-lq} can be used to verify
+quickly (without decompressing) the structural integrity of the files
+specified. (Use @samp{--test} to verify the data integrity). @samp{-alq}
+additionally verifies that none of the files specified contain trailing data.
+
+@item -m @var{bytes}
+@itemx --match-length=@var{bytes}
+When compressing, 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.
+
+@item -o @var{file}
+@itemx --output=@var{file}
+If @samp{-c} has not been also specified, write the (de)compressed output to
+@var{file}; keep input files unchanged. If compressing several files, each
+file is compressed independently. (The output consists of a sequence of
+independently compressed members). This option (or @samp{-c}) is needed when
+reading from a named pipe (fifo) or from a device. @w{@samp{-o -}} is
+equivalent to @samp{-c}. @samp{-o} has no effect when testing or listing.
+
+In order to keep backward compatibility with lzip versions prior to 1.22,
+when compressing from standard input and no other file names are given, the
+extension @samp{.lz} is appended to @var{file} unless it already ends in
+@samp{.lz} or @samp{.tlz}. This feature will be removed in a future version
+of lzip. Meanwhile, redirection may be used instead of @samp{-o} to write
+the compressed output to a file without the extension @samp{.lz} in its
+name: @w{@samp{lzip < file > foo}}.
+
+When compressing and splitting the output in volumes, @var{file} is used as
+a prefix, and several files named @samp{@var{file}00001.lz},
+@samp{@var{file}00002.lz}, etc, are created. In this case, only one input
+file is allowed.
+
+@item -q
+@itemx --quiet
+Quiet operation. Suppress all messages.
+
+@item -s @var{bytes}
+@itemx --dictionary-size=@var{bytes}
+When compressing, set the dictionary size limit in bytes. Lzip will use
+for each file the largest dictionary size that does not exceed neither
+the file size nor this limit. Valid values range from @w{4 KiB} to
+@w{512 MiB}. Values 12 to 29 are interpreted as powers of two, meaning
+2^12 to 2^29 bytes. Dictionary sizes are quantized so that they can be
+coded in just one byte (@pxref{coded-dict-size}). If the size specified
+does not match one of the valid sizes, it will be rounded upwards by
+adding up to @w{(@var{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.
+
+@item -S @var{bytes}
+@itemx --volume-size=@var{bytes}
+When compressing, and @samp{-c} has not been also specified, split the
+compressed output into several volume files with names
+@samp{original_name00001.lz}, @samp{original_name00002.lz}, etc, and set the
+volume size limit to @var{bytes}. Input files are kept unchanged. 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 @w{100 kB} to @w{4 EiB}.
+
+@item -t
+@itemx --test
+Check integrity of the files specified, but don't decompress them. This
+really performs a trial decompression and throws away the result. Use it
+together with @samp{-v} to see information about the files. 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. A final diagnostic is shown at
+verbosity level 1 or higher if any file fails the test when testing
+multiple files.
+
+@item -v
+@itemx --verbose
+Verbose mode.@*
+When compressing, show the compression ratio and size for each file
+processed.@*
+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.@*
+Two or more @samp{-v} options show the progress of (de)compression.
+
+@item -0 .. -9
+Compression level. Set the compression parameters (dictionary size and
+match length limit) as shown in the table below. The default compression
+level is @samp{-6}, equivalent to @w{@samp{-s8MiB -m36}}. Note that
+@samp{-9} can be much slower than @samp{-0}. These options have no
+effect when decompressing, testing, or listing.
+
+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 options @samp{--dictionary-size} and
+@samp{--match-length} directly to achieve optimal performance.
+
+If several compression levels or @samp{-s} or @samp{-m} options are
+given, the last setting is used. For example @w{@samp{-9 -s64MiB}} is
+equivalent to @w{@samp{-s64MiB -m273}}
+
+@multitable {Level} {Dictionary size (-s)} {Match length limit (-m)}
+@item Level @tab Dictionary size (-s) @tab Match length limit (-m)
+@item -0 @tab 64 KiB @tab 16 bytes
+@item -1 @tab 1 MiB @tab 5 bytes
+@item -2 @tab 1.5 MiB @tab 6 bytes
+@item -3 @tab 2 MiB @tab 8 bytes
+@item -4 @tab 3 MiB @tab 12 bytes
+@item -5 @tab 4 MiB @tab 20 bytes
+@item -6 @tab 8 MiB @tab 36 bytes
+@item -7 @tab 16 MiB @tab 68 bytes
+@item -8 @tab 24 MiB @tab 132 bytes
+@item -9 @tab 32 MiB @tab 273 bytes
+@end multitable
+
+@item --fast
+@itemx --best
+Aliases for GNU gzip compatibility.
+
+@item --loose-trailing
+When decompressing, testing, or listing, allow trailing data whose first
+bytes are so similar to the magic bytes of a lzip header that they can
+be confused with a corrupt header. Use this option if a file triggers a
+"corrupt header" error and the cause is not indeed a corrupt header.
+
+@end table
+
+Numbers given as arguments to options may be followed by a multiplier
+and an optional @samp{B} for "byte".
+
+Table of SI and binary prefixes (unit multipliers):
+
+@multitable {Prefix} {kilobyte (10^3 = 1000)} {|} {Prefix} {kibibyte (2^10 = 1024)}
+@item Prefix @tab Value @tab | @tab Prefix @tab Value
+@item k @tab kilobyte (10^3 = 1000) @tab | @tab Ki @tab kibibyte (2^10 = 1024)
+@item M @tab megabyte (10^6) @tab | @tab Mi @tab mebibyte (2^20)
+@item G @tab gigabyte (10^9) @tab | @tab Gi @tab gibibyte (2^30)
+@item T @tab terabyte (10^12) @tab | @tab Ti @tab tebibyte (2^40)
+@item P @tab petabyte (10^15) @tab | @tab Pi @tab pebibyte (2^50)
+@item E @tab exabyte (10^18) @tab | @tab Ei @tab exbibyte (2^60)
+@item Z @tab zettabyte (10^21) @tab | @tab Zi @tab zebibyte (2^70)
+@item Y @tab yottabyte (10^24) @tab | @tab Yi @tab yobibyte (2^80)
+@end multitable
+
+@sp 1
+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 (e.g., bug) which caused
+lzip to panic.
+
+
+@node Quality assurance
+@chapter Design, development, and testing of lzip
+@cindex quality assurance
+
+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 is developed by volunteers who lack the resources required for
+extensive testing in all circumstances. It is up to you to test lzip before
+using it in mission-critical applications. However, a compressor like lzip
+is not a toy, and maintaining it is not a hobby. Many people's data depend
+on it. Therefore the lzip file format has been reviewed carefully and is
+believed to be free from negligent design errors.
+
+Lzip has been designed, written, and tested with great care to replace gzip
+and bzip2 as the standard general-purpose compressed format for unix-like
+systems. This chapter describes the lessons learned from these previous
+formats, and their application to the design of lzip.
+
+@sp 1
+@section Format design
+
+When gzip was designed in 1992, computers and operating systems were much
+less capable than they are today. The designers of gzip tried to work around
+some of those limitations, like 8.3 file names, with additional fields in
+the 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 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.
+
+Lziprecover is able to provide unique data recovery capabilities because the
+lzip format is extraordinarily safe. The simple and safe design of the file
+format complements the embedded error detection provided by the LZMA data
+stream. 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
+possible bit flips in the compressed data without resorting to the check
+sequence. It would be 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 optimal accuracy in the detection of
+errors up to a compressed size of about @w{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 several
+million times more, resulting in a combined integrity checking optimally
+accurate for any member size produced by lzip. Preliminary results suggest
+that the lzip format is safe enough to be used in critical safety avionics
+systems.
+
+The lzip format is designed for long-term archiving. Therefore it excludes
+any unneeded features that may interfere with the future extraction of the
+decompressed data.
+
+@sp 1
+@subsection Gzip format (mis)features not present in lzip
+
+@table @samp
+@item 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.
+
+@item 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.
+
+@item Optional CRC for the header
+
+Using an optional CRC for the header is not only a bad idea, it is an error;
+it circumvents the Hamming distance (HD) of the CRC and may prevent the
+extraction of perfectly good data. For example, if the CRC 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.
+
+@item Metadata
+
+The gzip format stores some metadata, like the modification time of the
+original file or the operating system on which compression took place. This
+complicates reproducible compression (obtaining identical compressed output
+from identical input).
+
+@end table
+
+@subsection Lzip format improvements over gzip and bzip2
+
+@table @samp
+@item 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 @w{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.
+Additionally, lzip produces multimember output automatically when the size
+is too large for a single member, allowing for an unlimited uncompressed
+size.
+
+@item 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.
+
+@end table
+
+@section Quality of implementation
+
+@table @samp
+@item Accurate and robust error detection
+
+The lzip format provides 3 factor integrity checking, and the decompressors
+report mismatches in each factor separately. This method detects most false
+positives for corruption. 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, data size, or
+member size) in the member trailer.
+
+@item 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 tested 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
+@uref{http://www.nongnu.org/lzip/manual/lziprecover_manual.html#Unzcrash,,unzcrash},
+valgrind, and @samp{american fuzzy lop} without finding a single
+vulnerability or false negative.
+@ifnothtml
+@xref{Unzcrash,,,lziprecover}.
+@end ifnothtml
+
+@item Dictionary size
+
+Lzip automatically adapts the dictionary size to the size of 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. @c key4_mask
+
+@item 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.
+
+@end table
+
+
+@node Algorithm
+@chapter Algorithm
+@cindex 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". LZMA compression consists in describing the uncompressed data as a
+succession of coding sequences from the set shown in Section @samp{What is
+coded} (@pxref{what-is-coded}), and then encoding them using a range
+encoder. For example, the option @samp{-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 @samp{-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 bytes produced 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.
+
+@sp 1
+During compression, lzip reads data in large blocks (one dictionary size at
+a time). Therefore it may block for up to tens of seconds any process
+feeding data to it through a pipe. This is normal. The blocking intervals
+get longer with higher compression levels because dictionary size increases
+(and compression speed decreases) with compression level.
+
+@noindent
+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).
+
+
+@node File format
+@chapter File format
+@cindex 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
+
+@sp 1
+In the diagram below, a box like this:
+
+@verbatim
++---+
+| | <-- the vertical bars might be missing
++---+
+@end verbatim
+
+represents one byte; a box like this:
+
+@verbatim
++==============+
+| |
++==============+
+@end verbatim
+
+represents a variable number of bytes.
+
+@sp 1
+A lzip file consists of a series of independent "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 can
+encode in compressed form up to @w{16 EiB - 1 byte} of uncompressed data.
+The size of a multimember file is unlimited.
+
+Each member has the following structure:
+
+@verbatim
++--+--+--+--+----+----+=============+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+| ID string | VN | DS | LZMA stream | CRC32 | Data size | Member size |
++--+--+--+--+----+----+=============+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
+@end verbatim
+
+All multibyte values are stored in little endian order.
+
+@table @samp
+@item ID string (the "magic" bytes)
+A four byte string, identifying the lzip format, with the value "LZIP"
+(0x4C, 0x5A, 0x49, 0x50).
+
+@item VN (version number, 1 byte)
+Just in case something needs to be modified in the future. 1 for now.
+
+@anchor{coded-dict-size}
+@item DS (coded dictionary size, 1 byte)
+The dictionary size is calculated by taking a power of 2 (the base size)
+and subtracting 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 subtract
+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.
+
+@item LZMA stream
+The LZMA stream, finished by an "End Of Stream" marker. Uses default values
+for encoder properties. @xref{Stream format}, for a complete description.
+
+@item CRC32 (4 bytes)
+Cyclic Redundancy Check (CRC) of the original uncompressed data.
+
+@item Data size (8 bytes)
+Size of the original uncompressed data.
+
+@item 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 the safe recovery of undamaged members from multimember files.
+Member size should be limited to @w{2 PiB} to prevent the data size field
+from overflowing.
+
+@end table
+
+
+@node Stream format
+@chapter Format of the LZMA stream in lzip files
+@cindex format of the LZMA stream
+
+The LZMA algorithm has three parameters, called "special LZMA
+properties", to adjust it for some kinds of binary data. These
+parameters are: @samp{literal_context_bits} (with a default value of 3),
+@samp{literal_pos_state_bits} (with a default value of 0), and
+@samp{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 @samp{literal_pos_state_bits} has been optimized away and
+does not even appear in the code.
+
+Lzip finishes the LZMA stream with an "End Of Stream" (EOS) marker (the
+distance-length pair @w{0xFFFFFFFFU, 2}), which in conjunction with the
+@samp{member size} field in the member trailer allows the verification of
+stream integrity. The EOS marker is the only marker allowed in lzip files.
+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. This simplified form of the LZMA stream format has been chosen
+to maximize interoperability and safety.
+
+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 described individually.
+
+It seems that the only way of describing the LZMA-302eos stream is to
+describe 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 decompressor 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. Lzd is written in C++11 and its source code is included in
+appendix A. @xref{Reference source code}.
+
+@sp 1
+@section What is coded
+
+@anchor{what-is-coded}
+The LZMA stream includes literals, matches, and repeated matches (matches
+reusing a recently used distance). There are 7 different coding sequences:
+
+@multitable @columnfractions .35 .14 .51
+@headitem Bit sequence @tab Name @tab Description
+@item 0 + byte @tab literal @tab literal byte
+@item 1 + 0 + len + dis @tab match @tab distance-length pair
+@item 1 + 1 + 0 + 0 @tab shortrep @tab 1 byte match at latest used distance
+@item 1 + 1 + 0 + 1 + len @tab rep0 @tab len bytes match at latest used distance
+@item 1 + 1 + 1 + 0 + len @tab rep1 @tab len bytes match at second
+latest used distance
+@item 1 + 1 + 1 + 1 + 0 + len @tab rep2 @tab len bytes match at third
+latest used distance
+@item 1 + 1 + 1 + 1 + 1 + len @tab rep3 @tab len bytes match at fourth
+latest used distance
+@end multitable
+
+@sp 1
+In the following tables, multibit sequences are coded in normal order,
+from most significant bit (MSB) to least significant bit (LSB), except
+where noted otherwise.
+
+Lengths (the @samp{len} in the table above) are coded as follows:
+
+@multitable @columnfractions .5 .5
+@headitem Bit sequence @tab Description
+@item 0 + 3 bits @tab lengths from 2 to 9
+@item 1 + 0 + 3 bits @tab lengths from 10 to 17
+@item 1 + 1 + 8 bits @tab lengths from 18 to 273
+@end multitable
+
+@sp 1
+The coding of distances is a little more complicated, so I'll begin by
+explaining a simpler version of the encoding.
+
+Imagine you need to encode a number from 0 to @w{2^32 - 1}, and you want to
+do it in a way that produces shorter codes for the smaller numbers. You may
+first encode 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 @w{>= 0x80000000}). Then, if the position is @w{>= 2}, you encode
+the remaining @w{position - 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 encode the
+position, 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 in what it calls a "slot"
+the position of the most significant bit set, along with the value of the
+next bit, using the same 6 bits that would take to encode the position
+alone. This seems to need 66 slots (twice the number of positions), but for
+positions 0 and 1 there is no next bit, so the number of slots needed is 64
+(0 to 63).
+
+The 6 bits representing this "slot number" are then context-coded. If
+the distance is @w{>= 4}, the remaining bits are encoded as follows.
+@samp{direct_bits} is the amount of remaining bits (from 1 to 30) needed
+to form a complete distance, and is calculated as @w{(slot >> 1) - 1}.
+If a distance needs 6 or more direct_bits, the last 4 bits are encoded
+separately. The last piece (all the direct_bits for distances 4 to 127,
+or the last 4 bits for distances @w{>= 128}) is context-coded in reverse
+order (from LSB to MSB). For distances @w{>= 128}, the
+@w{@samp{direct_bits - 4}} part is encoded with fixed 0.5 probability.
+
+@multitable @columnfractions .5 .5
+@headitem Bit sequence @tab Description
+@item slot @tab distances from 0 to 3
+@item slot + direct_bits @tab distances from 4 to 127
+@item slot + (direct_bits - 4) + 4 bits @tab distances from 128 to
+2^32 - 1
+@end multitable
+
+@sp 1
+@section The coding contexts
+
+These contexts (@samp{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:
+
+@table @samp
+@item state
+A state machine (@samp{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.
+
+@item pos_state
+Value of the 2 least significant bits of the current position in the
+decoded data.
+
+@item literal_state
+Value of the 3 most significant bits of the latest byte decoded.
+
+@item len_state
+Coded value of the current match length @w{(length - 2)}, with a maximum
+of 3. The resulting value is in the range 0 to 3.
+
+@end table
+
+
+The types of previous sequences corresponding to each state are shown in the
+following table. @samp{!literal} is any sequence except a literal byte.
+@samp{rep} is any one of @samp{rep0}, @samp{rep1}, @samp{rep2}, or
+@samp{rep3}. The last type in each line is the most recent.
+
+@multitable {State} {rep or (!literal, shortrep), literal, literal}
+@headitem State @tab Types of previous sequences
+@item 0 @tab literal, literal, literal
+@item 1 @tab match, literal, literal
+@item 2 @tab rep or (!literal, shortrep), literal, literal
+@item 3 @tab literal, shortrep, literal, literal
+@item 4 @tab match, literal
+@item 5 @tab rep or (!literal, shortrep), literal
+@item 6 @tab literal, shortrep, literal
+@item 7 @tab literal, match
+@item 8 @tab literal, rep
+@item 9 @tab literal, shortrep
+@item 10 @tab !literal, match
+@item 11 @tab !literal, (rep or shortrep)
+@end multitable
+
+@sp 1
+The contexts for decoding the type of coding sequence are:
+
+@multitable @columnfractions .2 .35 .45
+@headitem Name @tab Indices @tab Used when
+@item bm_match @tab state, pos_state @tab sequence start
+@item bm_rep @tab state @tab after sequence 1
+@item bm_rep0 @tab state @tab after sequence 11
+@item bm_rep1 @tab state @tab after sequence 111
+@item bm_rep2 @tab state @tab after sequence 1111
+@item bm_len @tab state, pos_state @tab after sequence 110
+@end multitable
+
+@sp 1
+The contexts for decoding distances are:
+
+@multitable @columnfractions .2 .3 .5
+@headitem Name @tab Indices @tab Used when
+@item bm_dis_slot @tab len_state, bit tree @tab distance start
+@item bm_dis @tab reverse bit tree @tab after slots 4 to 13
+@item bm_align @tab reverse bit tree @tab for distances >= 128, after
+fixed probability bits
+@end multitable
+
+@sp 1
+There are two separate sets of contexts for lengths (@samp{Len_model} in
+the source). One for normal matches, the other for repeated matches. The
+contexts in each Len_model are (see @samp{decode_len} in the source):
+
+@multitable @columnfractions .2 .4 .4
+@headitem Name @tab Indices @tab Used when
+@item choice1 @tab none @tab length start
+@item choice2 @tab none @tab after sequence 1
+@item bm_low @tab pos_state, bit tree @tab after sequence 0
+@item bm_mid @tab pos_state, bit tree @tab after sequence 10
+@item bm_high @tab bit tree @tab after sequence 11
+@end multitable
+
+@sp 1
+The context array @samp{bm_literal} is special. In principle it acts as
+a normal bit tree context, the one selected by @samp{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
+@samp{match_byte} (the byte at the latest used distance), until a bit is
+decoded that is different from its corresponding bit in
+@samp{match_byte}. After the first difference is found, the rest of the
+byte is decoded using the normal bit tree context. (See
+@samp{decode_matched} in the source).
+
+@sp 1
+@section The range decoder
+
+The LZMA stream is consumed one byte at a time by the range decoder.
+(See @samp{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 @samp{decode_bit} in the source).
+
+The range decoder state consists of two unsigned 32-bit variables:
+@samp{range} (representing the most significant part of the range size
+not yet decoded) and @samp{code} (representing the current point within
+@samp{range}). @samp{range} is initialized to @w{2^32 - 1}, and
+@samp{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
+@samp{code} instead of 4. (See the @samp{Range_decoder} constructor in
+the source).
+
+@sp 1
+@section Decoding and verifying 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 @samp{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.
+
+Once the "End Of Stream" marker has been decoded, the decompressor reads and
+decodes the member trailer, and verifies that the three integrity factors
+stored there (CRC, data size, and member size) match those computed from the
+data.
+
+
+@node Trailing data
+@chapter Extra data appended to the file
+@cindex trailing data
+
+Sometimes extra data are found appended to a lzip file after the last
+member. Such trailing data may be:
+
+@itemize @bullet
+@item
+Padding added to make the file size a multiple of some block size, for
+example when writing to a tape. It is safe to append any amount of
+padding zero bytes to a lzip file.
+
+@item
+Useful data added by the user; a cryptographically secure hash, a
+description of file contents, etc. It is safe to append any amount of
+text to a lzip file as long as none of the first four bytes of the text
+match the corresponding byte in the string "LZIP", and the text does not
+contain any zero bytes (null characters). Nonzero bytes and zero bytes
+can't be safely mixed in trailing data.
+
+@item
+Garbage added by some not totally successful copy operation.
+
+@item
+Malicious data added to the file in order to make its total size and
+hash value (for a chosen hash) coincide with those of another file.
+
+@item
+In rare cases, trailing data could be the corrupt header of another
+member. In multimember or concatenated files the probability of
+corruption happening in the magic bytes is 5 times smaller than the
+probability of getting a false positive caused by the corruption of the
+integrity information itself. Therefore it can be considered to be below
+the noise level. Additionally, the test used by lzip to discriminate
+trailing data from a corrupt header has a Hamming distance (HD) of 3,
+and the 3 bit flips must happen in different magic bytes for the test to
+fail. In any case, the option @samp{--trailing-error} guarantees that
+any corrupt header will be detected.
+@end itemize
+
+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
+@samp{--trailing-error} can be used. @xref{--trailing-error}.
+
+
+@node Examples
+@chapter A small tutorial with examples
+@cindex 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
+option @samp{--keep} to lzip and don't remove the original file until you
+verify the compressed file with a command like
+@w{@samp{lzip -cd file.lz | cmp file -}}. Most RAM errors happening during
+compression can only be detected by comparing the compressed file with the
+original because the corruption happens before lzip compresses the RAM
+contents, resulting in a valid compressed file containing wrong data.
+
+@sp 1
+@noindent
+Example 1: Extract all the files from archive @samp{foo.tar.lz}.
+
+@example
+ tar -xf foo.tar.lz
+or
+ lzip -cd foo.tar.lz | tar -xf -
+@end example
+
+@sp 1
+@noindent
+Example 2: Replace a regular file with its compressed version @samp{file.lz}
+and show the compression ratio.
+
+@example
+lzip -v file
+@end example
+
+@sp 1
+@noindent
+Example 3: Like example 2 but the created @samp{file.lz} is multimember with
+a member size of @w{1 MiB}. The compression ratio is not shown.
+
+@example
+lzip -b 1MiB file
+@end example
+
+@sp 1
+@noindent
+Example 4: Restore a regular file from its compressed version
+@samp{file.lz}. If the operation is successful, @samp{file.lz} is removed.
+
+@example
+lzip -d file.lz
+@end example
+
+@sp 1
+@noindent
+Example 5: Verify the integrity of the compressed file @samp{file.lz} and
+show status.
+
+@example
+lzip -tv file.lz
+@end example
+
+@sp 1
+@anchor{concat-example}
+@noindent
+Example 6: The right way of concatenating the decompressed output of two or
+more compressed files. @xref{Trailing data}.
+
+@example
+Don't do this
+ cat file1.lz file2.lz file3.lz | lzip -d -
+Do this instead
+ lzip -cd file1.lz file2.lz file3.lz
+@end example
+
+@sp 1
+@noindent
+Example 7: Decompress @samp{file.lz} partially until @w{10 KiB} of
+decompressed data are produced.
+
+@example
+lzip -cd file.lz | dd bs=1024 count=10
+@end example
+
+@sp 1
+@noindent
+Example 8: Decompress @samp{file.lz} partially from decompressed byte at
+offset 10000 to decompressed byte at offset 14999 (5000 bytes are produced).
+
+@example
+lzip -cd file.lz | dd bs=1000 skip=10 count=5
+@end example
+
+@sp 1
+@noindent
+Example 9: Compress a whole device in /dev/sdc and send the output to
+@samp{file.lz}.
+
+@example
+ lzip -c /dev/sdc > file.lz
+or
+ lzip /dev/sdc -o file.lz
+@end example
+
+@sp 1
+@noindent
+Example 10: Create a multivolume compressed tar archive with a volume size
+of @w{1440 KiB}.
+
+@example
+tar -c some_directory | lzip -S 1440KiB -o volume_name -
+@end example
+
+@sp 1
+@noindent
+Example 11: Extract a multivolume compressed tar archive.
+
+@example
+lzip -cd volume_name*.lz | tar -xf -
+@end example
+
+@sp 1
+@noindent
+Example 12: Create a multivolume compressed backup of a large database file
+with a volume size of @w{650 MB}, where each volume is a multimember file
+with a member size of @w{32 MiB}.
+
+@example
+lzip -b 32MiB -S 650MB big_db
+@end example
+
+
+@node Problems
+@chapter Reporting bugs
+@cindex bugs
+@cindex getting help
+
+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
+@email{lzip-bug@@nongnu.org}. Include the version number, which you can
+find by running @w{@samp{lzip --version}}.
+
+
+@node Reference source code
+@appendix Reference source code
+@cindex reference source code
+
+@verbatim
+/* Lzd - Educational decompressor for the lzip format
+ Copyright (C) 2013-2022 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
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
+*/
+/*
+ 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 __DJGPP__
+#include <fcntl.h>
+#include <io.h>
+#endif
+
+
+class State
+ {
+ int st;
+
+public:
+ enum { states = 12 };
+ State() : st( 0 ) {}
+ int operator()() const { return st; }
+ bool is_char() const { return st < 7; }
+
+ void set_char()
+ {
+ 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.
+
+public:
+ CRC32()
+ {
+ 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 Lzip_header[6]; // 0-3 magic bytes
+ // 4 version
+ // 5 coded dictionary size
+typedef uint8_t Lzip_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
+ {
+ unsigned long long member_pos;
+ uint32_t code;
+ uint32_t range;
+
+public:
+ Range_decoder() : member_pos( 6 ), code( 0 ), range( 0xFFFFFFFFU )
+ {
+ for( int i = 0; i < 5; ++i ) code = ( code << 8 ) | get_byte();
+ }
+
+ uint8_t get_byte() { ++member_pos; return std::getc( stdin ); }
+ unsigned long long member_position() const { return member_pos; }
+
+ 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;
+ }
+ else
+ {
+ 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] );
+ break;
+ }
+ }
+ 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();
+ }
+
+public:
+ 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; }
+ uint8_t get_byte() { return rdec.get_byte(); }
+ unsigned long long member_position() const
+ { return rdec.member_position(); }
+
+ 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 );
+ 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
+ {
+ // literal byte
+ 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 ) );
+ else
+ put_byte( rdec.decode_matched( bm, peek( rep0 ) ) );
+ state.set_char();
+ continue;
+ }
+ // 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; }
+ }
+ else
+ {
+ unsigned distance;
+ if( rdec.decode_bit( bm_rep1[state()] ) == 0 ) // 4th bit
+ distance = rep1;
+ else
+ {
+ if( rdec.decode_bit( bm_rep2[state()] ) == 0 ) // 5th bit
+ distance = rep2;
+ else
+ { distance = rep3; rep3 = rep2; }
+ rep2 = rep1;
+ }
+ rep1 = rep0;
+ rep0 = distance;
+ }
+ state.set_rep();
+ 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 );
+ else
+ {
+ 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
+ {
+ flush_data();
+ return ( len == min_match_len ); // End Of Stream marker
+ }
+ }
+ }
+ state.set_match();
+ if( rep0 >= dictionary_size || ( rep0 >= pos && !pos_wrapped ) )
+ { flush_data(); return false; }
+ }
+ for( int i = 0; i < len; ++i ) put_byte( peek( rep0 ) );
+ }
+ flush_data();
+ return false;
+ }
+
+
+int main( const int argc, const char * const argv[] )
+ {
+ if( argc > 2 || ( argc == 2 && std::strcmp( argv[1], "-d" ) != 0 ) )
+ {
+ std::printf(
+ "Lzd %s - Educational decompressor for the lzip format.\n"
+ "Study the source to learn how a lzip decompressor works.\n"
+ "See the lzip manual for an explanation of the code.\n"
+ "\nUsage: %s [-d] < file.lz > file\n"
+ "Lzd decompresses from standard input to standard output.\n"
+ "\nCopyright (C) 2022 Antonio Diaz Diaz.\n"
+ "License 2-clause BSD.\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 lzip-bug@nongnu.org\n"
+ "Lzd home page: http://www.nongnu.org/lzip/lzd.html\n",
+ PROGVERSION, argv[0] );
+ return 0;
+ }
+
+#if defined __MSVCRT__ || defined __OS2__ || defined __DJGPP__
+ setmode( STDIN_FILENO, O_BINARY );
+ setmode( STDOUT_FILENO, O_BINARY );
+#endif
+
+ for( bool first_member = true; ; first_member = false )
+ {
+ Lzip_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; }
+ break; // ignore trailing data
+ }
+ 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; }
+
+ Lzip_trailer trailer; // verify trailer
+ for( int i = 0; i < 20; ++i ) trailer[i] = decoder.get_byte();
+ int retval = 0;
+ unsigned crc = 0;
+ for( int i = 3; i >= 0; --i ) crc = ( crc << 8 ) + trailer[i];
+ if( crc != decoder.crc() )
+ { std::fputs( "CRC mismatch\n", stderr ); retval = 2; }
+
+ unsigned long long data_size = 0;
+ for( int i = 11; i >= 4; --i )
+ data_size = ( data_size << 8 ) + trailer[i];
+ if( data_size != decoder.data_position() )
+ { std::fputs( "Data size mismatch\n", stderr ); retval = 2; }
+
+ unsigned long long member_size = 0;
+ for( int i = 19; i >= 12; --i )
+ member_size = ( member_size << 8 ) + trailer[i];
+ if( member_size != decoder.member_position() )
+ { std::fputs( "Member size mismatch\n", stderr ); retval = 2; }
+ if( retval ) return retval;
+ }
+
+ if( std::fclose( stdout ) != 0 )
+ { std::fprintf( stderr, "Error closing stdout: %s\n",
+ std::strerror( errno ) ); return 1; }
+ return 0;
+ }
+@end verbatim
+
+
+@node Concept index
+@unnumbered Concept index
+
+@printindex cp
+
+@bye