The Android Open Source Project | cfb3b27 | 2009-03-03 19:29:20 -0800 | [diff] [blame] | 1 | |
| 2 | NAME |
Elliott Hughes | 8645cf6 | 2021-12-08 15:07:46 -0800 | [diff] [blame] | 3 | bzip2, bunzip2 - a block-sorting file compressor, v1.0.8 |
The Android Open Source Project | cfb3b27 | 2009-03-03 19:29:20 -0800 | [diff] [blame] | 4 | bzcat - decompresses files to stdout |
| 5 | bzip2recover - recovers data from damaged bzip2 files |
| 6 | |
| 7 | |
| 8 | SYNOPSIS |
| 9 | bzip2 [ -cdfkqstvzVL123456789 ] [ filenames ... ] |
| 10 | bunzip2 [ -fkvsVL ] [ filenames ... ] |
| 11 | bzcat [ -s ] [ filenames ... ] |
| 12 | bzip2recover filename |
| 13 | |
| 14 | |
| 15 | DESCRIPTION |
| 16 | bzip2 compresses files using the Burrows-Wheeler block |
| 17 | sorting text compression algorithm, and Huffman coding. |
| 18 | Compression is generally considerably better than that |
| 19 | achieved by more conventional LZ77/LZ78-based compressors, |
| 20 | and approaches the performance of the PPM family of sta- |
| 21 | tistical compressors. |
| 22 | |
| 23 | The command-line options are deliberately very similar to |
| 24 | those of GNU gzip, but they are not identical. |
| 25 | |
| 26 | bzip2 expects a list of file names to accompany the com- |
| 27 | mand-line flags. Each file is replaced by a compressed |
| 28 | version of itself, with the name "original_name.bz2". |
| 29 | Each compressed file has the same modification date, per- |
| 30 | missions, and, when possible, ownership as the correspond- |
| 31 | ing original, so that these properties can be correctly |
| 32 | restored at decompression time. File name handling is |
| 33 | naive in the sense that there is no mechanism for preserv- |
| 34 | ing original file names, permissions, ownerships or dates |
| 35 | in filesystems which lack these concepts, or have serious |
| 36 | file name length restrictions, such as MS-DOS. |
| 37 | |
| 38 | bzip2 and bunzip2 will by default not overwrite existing |
| 39 | files. If you want this to happen, specify the -f flag. |
| 40 | |
| 41 | If no file names are specified, bzip2 compresses from |
| 42 | standard input to standard output. In this case, bzip2 |
| 43 | will decline to write compressed output to a terminal, as |
| 44 | this would be entirely incomprehensible and therefore |
| 45 | pointless. |
| 46 | |
| 47 | bunzip2 (or bzip2 -d) decompresses all specified files. |
| 48 | Files which were not created by bzip2 will be detected and |
| 49 | ignored, and a warning issued. bzip2 attempts to guess |
| 50 | the filename for the decompressed file from that of the |
| 51 | compressed file as follows: |
| 52 | |
| 53 | filename.bz2 becomes filename |
| 54 | filename.bz becomes filename |
| 55 | filename.tbz2 becomes filename.tar |
| 56 | filename.tbz becomes filename.tar |
| 57 | anyothername becomes anyothername.out |
| 58 | |
| 59 | If the file does not end in one of the recognised endings, |
| 60 | .bz2, .bz, .tbz2 or .tbz, bzip2 complains that it cannot |
| 61 | guess the name of the original file, and uses the original |
| 62 | name with .out appended. |
| 63 | |
| 64 | As with compression, supplying no filenames causes decom- |
| 65 | pression from standard input to standard output. |
| 66 | |
| 67 | bunzip2 will correctly decompress a file which is the con- |
| 68 | catenation of two or more compressed files. The result is |
| 69 | the concatenation of the corresponding uncompressed files. |
| 70 | Integrity testing (-t) of concatenated compressed files is |
| 71 | also supported. |
| 72 | |
| 73 | You can also compress or decompress files to the standard |
| 74 | output by giving the -c flag. Multiple files may be com- |
| 75 | pressed and decompressed like this. The resulting outputs |
| 76 | are fed sequentially to stdout. Compression of multiple |
| 77 | files in this manner generates a stream containing multi- |
| 78 | ple compressed file representations. Such a stream can be |
| 79 | decompressed correctly only by bzip2 version 0.9.0 or |
| 80 | later. Earlier versions of bzip2 will stop after decom- |
| 81 | pressing the first file in the stream. |
| 82 | |
| 83 | bzcat (or bzip2 -dc) decompresses all specified files to |
| 84 | the standard output. |
| 85 | |
| 86 | bzip2 will read arguments from the environment variables |
| 87 | BZIP2 and BZIP, in that order, and will process them |
| 88 | before any arguments read from the command line. This |
| 89 | gives a convenient way to supply default arguments. |
| 90 | |
| 91 | Compression is always performed, even if the compressed |
| 92 | file is slightly larger than the original. Files of less |
| 93 | than about one hundred bytes tend to get larger, since the |
| 94 | compression mechanism has a constant overhead in the |
| 95 | region of 50 bytes. Random data (including the output of |
| 96 | most file compressors) is coded at about 8.05 bits per |
| 97 | byte, giving an expansion of around 0.5%. |
| 98 | |
| 99 | As a self-check for your protection, bzip2 uses 32-bit |
| 100 | CRCs to make sure that the decompressed version of a file |
| 101 | is identical to the original. This guards against corrup- |
| 102 | tion of the compressed data, and against undetected bugs |
| 103 | in bzip2 (hopefully very unlikely). The chances of data |
| 104 | corruption going undetected is microscopic, about one |
| 105 | chance in four billion for each file processed. Be aware, |
| 106 | though, that the check occurs upon decompression, so it |
| 107 | can only tell you that something is wrong. It can't help |
| 108 | you recover the original uncompressed data. You can use |
| 109 | bzip2recover to try to recover data from damaged files. |
| 110 | |
| 111 | Return values: 0 for a normal exit, 1 for environmental |
| 112 | problems (file not found, invalid flags, I/O errors, &c), |
| 113 | 2 to indicate a corrupt compressed file, 3 for an internal |
| 114 | consistency error (eg, bug) which caused bzip2 to panic. |
| 115 | |
| 116 | |
| 117 | OPTIONS |
| 118 | -c --stdout |
| 119 | Compress or decompress to standard output. |
| 120 | |
| 121 | -d --decompress |
| 122 | Force decompression. bzip2, bunzip2 and bzcat are |
| 123 | really the same program, and the decision about |
| 124 | what actions to take is done on the basis of which |
| 125 | name is used. This flag overrides that mechanism, |
| 126 | and forces bzip2 to decompress. |
| 127 | |
| 128 | -z --compress |
| 129 | The complement to -d: forces compression, |
| 130 | regardless of the invocation name. |
| 131 | |
| 132 | -t --test |
| 133 | Check integrity of the specified file(s), but don't |
| 134 | decompress them. This really performs a trial |
| 135 | decompression and throws away the result. |
| 136 | |
| 137 | -f --force |
| 138 | Force overwrite of output files. Normally, bzip2 |
| 139 | will not overwrite existing output files. Also |
| 140 | forces bzip2 to break hard links to files, which it |
| 141 | otherwise wouldn't do. |
| 142 | |
| 143 | bzip2 normally declines to decompress files which |
| 144 | don't have the correct magic header bytes. If |
| 145 | forced (-f), however, it will pass such files |
| 146 | through unmodified. This is how GNU gzip behaves. |
| 147 | |
| 148 | -k --keep |
| 149 | Keep (don't delete) input files during compression |
| 150 | or decompression. |
| 151 | |
| 152 | -s --small |
| 153 | Reduce memory usage, for compression, decompression |
| 154 | and testing. Files are decompressed and tested |
| 155 | using a modified algorithm which only requires 2.5 |
| 156 | bytes per block byte. This means any file can be |
| 157 | decompressed in 2300k of memory, albeit at about |
| 158 | half the normal speed. |
| 159 | |
| 160 | During compression, -s selects a block size of |
| 161 | 200k, which limits memory use to around the same |
| 162 | figure, at the expense of your compression ratio. |
| 163 | In short, if your machine is low on memory (8 |
| 164 | megabytes or less), use -s for everything. See |
| 165 | MEMORY MANAGEMENT below. |
| 166 | |
| 167 | -q --quiet |
| 168 | Suppress non-essential warning messages. Messages |
| 169 | pertaining to I/O errors and other critical events |
| 170 | will not be suppressed. |
| 171 | |
| 172 | -v --verbose |
| 173 | Verbose mode -- show the compression ratio for each |
| 174 | file processed. Further -v's increase the ver- |
| 175 | bosity level, spewing out lots of information which |
| 176 | is primarily of interest for diagnostic purposes. |
| 177 | |
| 178 | -L --license -V --version |
| 179 | Display the software version, license terms and |
| 180 | conditions. |
| 181 | |
| 182 | -1 (or --fast) to -9 (or --best) |
| 183 | Set the block size to 100 k, 200 k .. 900 k when |
| 184 | compressing. Has no effect when decompressing. |
| 185 | See MEMORY MANAGEMENT below. The --fast and --best |
| 186 | aliases are primarily for GNU gzip compatibility. |
| 187 | In particular, --fast doesn't make things signifi- |
| 188 | cantly faster. And --best merely selects the |
| 189 | default behaviour. |
| 190 | |
| 191 | -- Treats all subsequent arguments as file names, even |
| 192 | if they start with a dash. This is so you can han- |
| 193 | dle files with names beginning with a dash, for |
| 194 | example: bzip2 -- -myfilename. |
| 195 | |
| 196 | --repetitive-fast --repetitive-best |
| 197 | These flags are redundant in versions 0.9.5 and |
| 198 | above. They provided some coarse control over the |
| 199 | behaviour of the sorting algorithm in earlier ver- |
| 200 | sions, which was sometimes useful. 0.9.5 and above |
| 201 | have an improved algorithm which renders these |
| 202 | flags irrelevant. |
| 203 | |
| 204 | |
| 205 | MEMORY MANAGEMENT |
| 206 | bzip2 compresses large files in blocks. The block size |
| 207 | affects both the compression ratio achieved, and the |
| 208 | amount of memory needed for compression and decompression. |
| 209 | The flags -1 through -9 specify the block size to be |
| 210 | 100,000 bytes through 900,000 bytes (the default) respec- |
| 211 | tively. At decompression time, the block size used for |
| 212 | compression is read from the header of the compressed |
| 213 | file, and bunzip2 then allocates itself just enough memory |
| 214 | to decompress the file. Since block sizes are stored in |
| 215 | compressed files, it follows that the flags -1 to -9 are |
| 216 | irrelevant to and so ignored during decompression. |
| 217 | |
| 218 | Compression and decompression requirements, in bytes, can |
| 219 | be estimated as: |
| 220 | |
| 221 | Compression: 400k + ( 8 x block size ) |
| 222 | |
| 223 | Decompression: 100k + ( 4 x block size ), or |
| 224 | 100k + ( 2.5 x block size ) |
| 225 | |
| 226 | Larger block sizes give rapidly diminishing marginal |
| 227 | returns. Most of the compression comes from the first two |
| 228 | or three hundred k of block size, a fact worth bearing in |
| 229 | mind when using bzip2 on small machines. It is also |
| 230 | important to appreciate that the decompression memory |
| 231 | requirement is set at compression time by the choice of |
| 232 | block size. |
| 233 | |
| 234 | For files compressed with the default 900k block size, |
| 235 | bunzip2 will require about 3700 kbytes to decompress. To |
| 236 | support decompression of any file on a 4 megabyte machine, |
| 237 | bunzip2 has an option to decompress using approximately |
| 238 | half this amount of memory, about 2300 kbytes. Decompres- |
| 239 | sion speed is also halved, so you should use this option |
| 240 | only where necessary. The relevant flag is -s. |
| 241 | |
| 242 | In general, try and use the largest block size memory con- |
| 243 | straints allow, since that maximises the compression |
| 244 | achieved. Compression and decompression speed are virtu- |
| 245 | ally unaffected by block size. |
| 246 | |
| 247 | Another significant point applies to files which fit in a |
| 248 | single block -- that means most files you'd encounter |
| 249 | using a large block size. The amount of real memory |
| 250 | touched is proportional to the size of the file, since the |
| 251 | file is smaller than a block. For example, compressing a |
| 252 | file 20,000 bytes long with the flag -9 will cause the |
| 253 | compressor to allocate around 7600k of memory, but only |
| 254 | touch 400k + 20000 * 8 = 560 kbytes of it. Similarly, the |
| 255 | decompressor will allocate 3700k but only touch 100k + |
| 256 | 20000 * 4 = 180 kbytes. |
| 257 | |
| 258 | Here is a table which summarises the maximum memory usage |
| 259 | for different block sizes. Also recorded is the total |
| 260 | compressed size for 14 files of the Calgary Text Compres- |
| 261 | sion Corpus totalling 3,141,622 bytes. This column gives |
| 262 | some feel for how compression varies with block size. |
| 263 | These figures tend to understate the advantage of larger |
| 264 | block sizes for larger files, since the Corpus is domi- |
| 265 | nated by smaller files. |
| 266 | |
| 267 | Compress Decompress Decompress Corpus |
| 268 | Flag usage usage -s usage Size |
| 269 | |
| 270 | -1 1200k 500k 350k 914704 |
| 271 | -2 2000k 900k 600k 877703 |
| 272 | -3 2800k 1300k 850k 860338 |
| 273 | -4 3600k 1700k 1100k 846899 |
| 274 | -5 4400k 2100k 1350k 845160 |
| 275 | -6 5200k 2500k 1600k 838626 |
| 276 | -7 6100k 2900k 1850k 834096 |
| 277 | -8 6800k 3300k 2100k 828642 |
| 278 | -9 7600k 3700k 2350k 828642 |
| 279 | |
| 280 | |
| 281 | RECOVERING DATA FROM DAMAGED FILES |
| 282 | bzip2 compresses files in blocks, usually 900kbytes long. |
| 283 | Each block is handled independently. If a media or trans- |
| 284 | mission error causes a multi-block .bz2 file to become |
| 285 | damaged, it may be possible to recover data from the |
| 286 | undamaged blocks in the file. |
| 287 | |
| 288 | The compressed representation of each block is delimited |
| 289 | by a 48-bit pattern, which makes it possible to find the |
| 290 | block boundaries with reasonable certainty. Each block |
| 291 | also carries its own 32-bit CRC, so damaged blocks can be |
| 292 | distinguished from undamaged ones. |
| 293 | |
| 294 | bzip2recover is a simple program whose purpose is to |
| 295 | search for blocks in .bz2 files, and write each block out |
| 296 | into its own .bz2 file. You can then use bzip2 -t to test |
| 297 | the integrity of the resulting files, and decompress those |
| 298 | which are undamaged. |
| 299 | |
| 300 | bzip2recover takes a single argument, the name of the dam- |
| 301 | aged file, and writes a number of files |
| 302 | "rec00001file.bz2", "rec00002file.bz2", etc, containing |
| 303 | the extracted blocks. The output filenames are |
| 304 | designed so that the use of wildcards in subsequent pro- |
| 305 | cessing -- for example, "bzip2 -dc rec*file.bz2 > recov- |
| 306 | ered_data" -- processes the files in the correct order. |
| 307 | |
| 308 | bzip2recover should be of most use dealing with large .bz2 |
| 309 | files, as these will contain many blocks. It is clearly |
| 310 | futile to use it on damaged single-block files, since a |
| 311 | damaged block cannot be recovered. If you wish to min- |
| 312 | imise any potential data loss through media or transmis- |
| 313 | sion errors, you might consider compressing with a smaller |
| 314 | block size. |
| 315 | |
| 316 | |
| 317 | PERFORMANCE NOTES |
| 318 | The sorting phase of compression gathers together similar |
| 319 | strings in the file. Because of this, files containing |
| 320 | very long runs of repeated symbols, like "aabaabaabaab |
| 321 | ..." (repeated several hundred times) may compress more |
| 322 | slowly than normal. Versions 0.9.5 and above fare much |
| 323 | better than previous versions in this respect. The ratio |
| 324 | between worst-case and average-case compression time is in |
| 325 | the region of 10:1. For previous versions, this figure |
| 326 | was more like 100:1. You can use the -vvvv option to mon- |
| 327 | itor progress in great detail, if you want. |
| 328 | |
| 329 | Decompression speed is unaffected by these phenomena. |
| 330 | |
| 331 | bzip2 usually allocates several megabytes of memory to |
| 332 | operate in, and then charges all over it in a fairly ran- |
| 333 | dom fashion. This means that performance, both for com- |
| 334 | pressing and decompressing, is largely determined by the |
| 335 | speed at which your machine can service cache misses. |
| 336 | Because of this, small changes to the code to reduce the |
| 337 | miss rate have been observed to give disproportionately |
| 338 | large performance improvements. I imagine bzip2 will per- |
| 339 | form best on machines with very large caches. |
| 340 | |
| 341 | |
| 342 | CAVEATS |
| 343 | I/O error messages are not as helpful as they could be. |
| 344 | bzip2 tries hard to detect I/O errors and exit cleanly, |
| 345 | but the details of what the problem is sometimes seem |
| 346 | rather misleading. |
| 347 | |
Elliott Hughes | 8645cf6 | 2021-12-08 15:07:46 -0800 | [diff] [blame] | 348 | This manual page pertains to version 1.0.8 of bzip2. Com- |
The Android Open Source Project | cfb3b27 | 2009-03-03 19:29:20 -0800 | [diff] [blame] | 349 | pressed data created by this version is entirely forwards |
| 350 | and backwards compatible with the previous public |
| 351 | releases, versions 0.1pl2, 0.9.0, 0.9.5, 1.0.0, 1.0.1, |
Nick Kralevich | 172b266 | 2010-09-20 17:21:30 -0700 | [diff] [blame] | 352 | 1.0.2 and above, but with the following exception: 0.9.0 |
The Android Open Source Project | cfb3b27 | 2009-03-03 19:29:20 -0800 | [diff] [blame] | 353 | and above can correctly decompress multiple concatenated |
| 354 | compressed files. 0.1pl2 cannot do this; it will stop |
| 355 | after decompressing just the first file in the stream. |
| 356 | |
| 357 | bzip2recover versions prior to 1.0.2 used 32-bit integers |
| 358 | to represent bit positions in compressed files, so they |
| 359 | could not handle compressed files more than 512 megabytes |
| 360 | long. Versions 1.0.2 and above use 64-bit ints on some |
| 361 | platforms which support them (GNU supported targets, and |
| 362 | Windows). To establish whether or not bzip2recover was |
| 363 | built with such a limitation, run it without arguments. |
| 364 | In any event you can build yourself an unlimited version |
| 365 | if you can recompile it with MaybeUInt64 set to be an |
| 366 | unsigned 64-bit integer. |
| 367 | |
| 368 | |
| 369 | AUTHOR |
Elliott Hughes | 8645cf6 | 2021-12-08 15:07:46 -0800 | [diff] [blame] | 370 | Julian Seward, jseward@acm.org |
The Android Open Source Project | cfb3b27 | 2009-03-03 19:29:20 -0800 | [diff] [blame] | 371 | |
Elliott Hughes | 8645cf6 | 2021-12-08 15:07:46 -0800 | [diff] [blame] | 372 | https://sourceware.org/bzip2/ |
The Android Open Source Project | cfb3b27 | 2009-03-03 19:29:20 -0800 | [diff] [blame] | 373 | |
| 374 | The ideas embodied in bzip2 are due to (at least) the fol- |
| 375 | lowing people: Michael Burrows and David Wheeler (for the |
| 376 | block sorting transformation), David Wheeler (again, for |
| 377 | the Huffman coder), Peter Fenwick (for the structured cod- |
| 378 | ing model in the original bzip, and many refinements), and |
| 379 | Alistair Moffat, Radford Neal and Ian Witten (for the |
| 380 | arithmetic coder in the original bzip). I am much |
| 381 | indebted for their help, support and advice. See the man- |
| 382 | ual in the source distribution for pointers to sources of |
| 383 | documentation. Christian von Roques encouraged me to look |
| 384 | for faster sorting algorithms, so as to speed up compres- |
| 385 | sion. Bela Lubkin encouraged me to improve the worst-case |
| 386 | compression performance. Donna Robinson XMLised the docu- |
| 387 | mentation. The bz* scripts are derived from those of GNU |
| 388 | gzip. Many people sent patches, helped with portability |
| 389 | problems, lent machines, gave advice and were generally |
| 390 | helpful. |
| 391 | |