• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1<?xml version="1.0"?> <!-- -*- sgml -*- -->
2<!DOCTYPE book PUBLIC "-//OASIS//DTD DocBook XML V4.2//EN"
3  "http://www.oasis-open.org/docbook/xml/4.2/docbookx.dtd"[
4
5<!-- various strings, dates etc. common to all docs -->
6<!ENTITY % common-ents SYSTEM "entities.xml"> %common-ents;
7]>
8
9<book lang="en" id="userman" xreflabel="bzip2 Manual">
10
11 <bookinfo>
12  <title>bzip2 and libbzip2, version 1.0.6</title>
13  <subtitle>A program and library for data compression</subtitle>
14  <copyright>
15   <year>&bz-lifespan;</year>
16   <holder>Julian Seward</holder>
17  </copyright>
18  <releaseinfo>Version &bz-version; of &bz-date;</releaseinfo>
19
20  <authorgroup>
21   <author>
22    <firstname>Julian</firstname>
23    <surname>Seward</surname>
24    <affiliation>
25     <orgname>&bz-url;</orgname>
26    </affiliation>
27   </author>
28  </authorgroup>
29
30  <legalnotice>
31
32  <para>This program, <computeroutput>bzip2</computeroutput>, the
33  associated library <computeroutput>libbzip2</computeroutput>, and
34  all documentation, are copyright &copy; &bz-lifespan; Julian Seward.
35  All rights reserved.</para>
36
37  <para>Redistribution and use in source and binary forms, with
38  or without modification, are permitted provided that the
39  following conditions are met:</para>
40
41  <itemizedlist mark='bullet'>
42
43   <listitem><para>Redistributions of source code must retain the
44   above copyright notice, this list of conditions and the
45   following disclaimer.</para></listitem>
46
47   <listitem><para>The origin of this software must not be
48   misrepresented; you must not claim that you wrote the original
49   software.  If you use this software in a product, an
50   acknowledgment in the product documentation would be
51   appreciated but is not required.</para></listitem>
52
53   <listitem><para>Altered source versions must be plainly marked
54   as such, and must not be misrepresented as being the original
55   software.</para></listitem>
56
57   <listitem><para>The name of the author may not be used to
58   endorse or promote products derived from this software without
59   specific prior written permission.</para></listitem>
60
61  </itemizedlist>
62
63  <para>THIS SOFTWARE IS PROVIDED BY THE AUTHOR "AS IS" AND ANY
64  EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO,
65  THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A
66  PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
67  AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
68  EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
69  TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
70  DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
71  ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
72  LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
73  IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
74  THE POSSIBILITY OF SUCH DAMAGE.</para>
75
76 <para>PATENTS: To the best of my knowledge,
77 <computeroutput>bzip2</computeroutput> and
78 <computeroutput>libbzip2</computeroutput> do not use any patented
79 algorithms.  However, I do not have the resources to carry
80 out a patent search.  Therefore I cannot give any guarantee of
81 the above statement.
82 </para>
83
84</legalnotice>
85
86</bookinfo>
87
88
89
90<chapter id="intro" xreflabel="Introduction">
91<title>Introduction</title>
92
93<para><computeroutput>bzip2</computeroutput> compresses files
94using the Burrows-Wheeler block-sorting text compression
95algorithm, and Huffman coding.  Compression is generally
96considerably better than that achieved by more conventional
97LZ77/LZ78-based compressors, and approaches the performance of
98the PPM family of statistical compressors.</para>
99
100<para><computeroutput>bzip2</computeroutput> is built on top of
101<computeroutput>libbzip2</computeroutput>, a flexible library for
102handling compressed data in the
103<computeroutput>bzip2</computeroutput> format.  This manual
104describes both how to use the program and how to work with the
105library interface.  Most of the manual is devoted to this
106library, not the program, which is good news if your interest is
107only in the program.</para>
108
109<itemizedlist mark='bullet'>
110
111 <listitem><para><xref linkend="using"/> describes how to use
112 <computeroutput>bzip2</computeroutput>; this is the only part
113 you need to read if you just want to know how to operate the
114 program.</para></listitem>
115
116 <listitem><para><xref linkend="libprog"/> describes the
117 programming interfaces in detail, and</para></listitem>
118
119 <listitem><para><xref linkend="misc"/> records some
120 miscellaneous notes which I thought ought to be recorded
121 somewhere.</para></listitem>
122
123</itemizedlist>
124
125</chapter>
126
127
128<chapter id="using" xreflabel="How to use bzip2">
129<title>How to use bzip2</title>
130
131<para>This chapter contains a copy of the
132<computeroutput>bzip2</computeroutput> man page, and nothing
133else.</para>
134
135<sect1 id="name" xreflabel="NAME">
136<title>NAME</title>
137
138<itemizedlist mark='bullet'>
139
140 <listitem><para><computeroutput>bzip2</computeroutput>,
141  <computeroutput>bunzip2</computeroutput> - a block-sorting file
142  compressor, v1.0.6</para></listitem>
143
144 <listitem><para><computeroutput>bzcat</computeroutput> -
145   decompresses files to stdout</para></listitem>
146
147 <listitem><para><computeroutput>bzip2recover</computeroutput> -
148   recovers data from damaged bzip2 files</para></listitem>
149
150</itemizedlist>
151
152</sect1>
153
154
155<sect1 id="synopsis" xreflabel="SYNOPSIS">
156<title>SYNOPSIS</title>
157
158<itemizedlist mark='bullet'>
159
160 <listitem><para><computeroutput>bzip2</computeroutput> [
161  -cdfkqstvzVL123456789 ] [ filenames ...  ]</para></listitem>
162
163 <listitem><para><computeroutput>bunzip2</computeroutput> [
164  -fkvsVL ] [ filenames ...  ]</para></listitem>
165
166 <listitem><para><computeroutput>bzcat</computeroutput> [ -s ] [
167  filenames ...  ]</para></listitem>
168
169 <listitem><para><computeroutput>bzip2recover</computeroutput>
170  filename</para></listitem>
171
172</itemizedlist>
173
174</sect1>
175
176
177<sect1 id="description" xreflabel="DESCRIPTION">
178<title>DESCRIPTION</title>
179
180<para><computeroutput>bzip2</computeroutput> compresses files
181using the Burrows-Wheeler block sorting text compression
182algorithm, and Huffman coding.  Compression is generally
183considerably better than that achieved by more conventional
184LZ77/LZ78-based compressors, and approaches the performance of
185the PPM family of statistical compressors.</para>
186
187<para>The command-line options are deliberately very similar to
188those of GNU <computeroutput>gzip</computeroutput>, but they are
189not identical.</para>
190
191<para><computeroutput>bzip2</computeroutput> expects a list of
192file names to accompany the command-line flags.  Each file is
193replaced by a compressed version of itself, with the name
194<computeroutput>original_name.bz2</computeroutput>.  Each
195compressed file has the same modification date, permissions, and,
196when possible, ownership as the corresponding original, so that
197these properties can be correctly restored at decompression time.
198File name handling is naive in the sense that there is no
199mechanism for preserving original file names, permissions,
200ownerships or dates in filesystems which lack these concepts, or
201have serious file name length restrictions, such as
202MS-DOS.</para>
203
204<para><computeroutput>bzip2</computeroutput> and
205<computeroutput>bunzip2</computeroutput> will by default not
206overwrite existing files.  If you want this to happen, specify
207the <computeroutput>-f</computeroutput> flag.</para>
208
209<para>If no file names are specified,
210<computeroutput>bzip2</computeroutput> compresses from standard
211input to standard output.  In this case,
212<computeroutput>bzip2</computeroutput> will decline to write
213compressed output to a terminal, as this would be entirely
214incomprehensible and therefore pointless.</para>
215
216<para><computeroutput>bunzip2</computeroutput> (or
217<computeroutput>bzip2 -d</computeroutput>) decompresses all
218specified files.  Files which were not created by
219<computeroutput>bzip2</computeroutput> will be detected and
220ignored, and a warning issued.
221<computeroutput>bzip2</computeroutput> attempts to guess the
222filename for the decompressed file from that of the compressed
223file as follows:</para>
224
225<itemizedlist mark='bullet'>
226
227 <listitem><para><computeroutput>filename.bz2 </computeroutput>
228  becomes
229  <computeroutput>filename</computeroutput></para></listitem>
230
231 <listitem><para><computeroutput>filename.bz </computeroutput>
232  becomes
233  <computeroutput>filename</computeroutput></para></listitem>
234
235 <listitem><para><computeroutput>filename.tbz2</computeroutput>
236  becomes
237  <computeroutput>filename.tar</computeroutput></para></listitem>
238
239 <listitem><para><computeroutput>filename.tbz </computeroutput>
240  becomes
241  <computeroutput>filename.tar</computeroutput></para></listitem>
242
243 <listitem><para><computeroutput>anyothername </computeroutput>
244  becomes
245  <computeroutput>anyothername.out</computeroutput></para></listitem>
246
247</itemizedlist>
248
249<para>If the file does not end in one of the recognised endings,
250<computeroutput>.bz2</computeroutput>,
251<computeroutput>.bz</computeroutput>,
252<computeroutput>.tbz2</computeroutput> or
253<computeroutput>.tbz</computeroutput>,
254<computeroutput>bzip2</computeroutput> complains that it cannot
255guess the name of the original file, and uses the original name
256with <computeroutput>.out</computeroutput> appended.</para>
257
258<para>As with compression, supplying no filenames causes
259decompression from standard input to standard output.</para>
260
261<para><computeroutput>bunzip2</computeroutput> will correctly
262decompress a file which is the concatenation of two or more
263compressed files.  The result is the concatenation of the
264corresponding uncompressed files.  Integrity testing
265(<computeroutput>-t</computeroutput>) of concatenated compressed
266files is also supported.</para>
267
268<para>You can also compress or decompress files to the standard
269output by giving the <computeroutput>-c</computeroutput> flag.
270Multiple files may be compressed and decompressed like this.  The
271resulting outputs are fed sequentially to stdout.  Compression of
272multiple files in this manner generates a stream containing
273multiple compressed file representations.  Such a stream can be
274decompressed correctly only by
275<computeroutput>bzip2</computeroutput> version 0.9.0 or later.
276Earlier versions of <computeroutput>bzip2</computeroutput> will
277stop after decompressing the first file in the stream.</para>
278
279<para><computeroutput>bzcat</computeroutput> (or
280<computeroutput>bzip2 -dc</computeroutput>) decompresses all
281specified files to the standard output.</para>
282
283<para><computeroutput>bzip2</computeroutput> will read arguments
284from the environment variables
285<computeroutput>BZIP2</computeroutput> and
286<computeroutput>BZIP</computeroutput>, in that order, and will
287process them before any arguments read from the command line.
288This gives a convenient way to supply default arguments.</para>
289
290<para>Compression is always performed, even if the compressed
291file is slightly larger than the original.  Files of less than
292about one hundred bytes tend to get larger, since the compression
293mechanism has a constant overhead in the region of 50 bytes.
294Random data (including the output of most file compressors) is
295coded at about 8.05 bits per byte, giving an expansion of around
2960.5%.</para>
297
298<para>As a self-check for your protection,
299<computeroutput>bzip2</computeroutput> uses 32-bit CRCs to make
300sure that the decompressed version of a file is identical to the
301original.  This guards against corruption of the compressed data,
302and against undetected bugs in
303<computeroutput>bzip2</computeroutput> (hopefully very unlikely).
304The chances of data corruption going undetected is microscopic,
305about one chance in four billion for each file processed.  Be
306aware, though, that the check occurs upon decompression, so it
307can only tell you that something is wrong.  It can't help you
308recover the original uncompressed data.  You can use
309<computeroutput>bzip2recover</computeroutput> to try to recover
310data from damaged files.</para>
311
312<para>Return values: 0 for a normal exit, 1 for environmental
313problems (file not found, invalid flags, I/O errors, etc.), 2
314to indicate a corrupt compressed file, 3 for an internal
315consistency error (eg, bug) which caused
316<computeroutput>bzip2</computeroutput> to panic.</para>
317
318</sect1>
319
320
321<sect1 id="options" xreflabel="OPTIONS">
322<title>OPTIONS</title>
323
324<variablelist>
325
326 <varlistentry>
327 <term><computeroutput>-c --stdout</computeroutput></term>
328 <listitem><para>Compress or decompress to standard
329  output.</para></listitem>
330 </varlistentry>
331
332 <varlistentry>
333 <term><computeroutput>-d --decompress</computeroutput></term>
334 <listitem><para>Force decompression.
335  <computeroutput>bzip2</computeroutput>,
336  <computeroutput>bunzip2</computeroutput> and
337  <computeroutput>bzcat</computeroutput> are really the same
338  program, and the decision about what actions to take is done on
339  the basis of which name is used.  This flag overrides that
340  mechanism, and forces bzip2 to decompress.</para></listitem>
341 </varlistentry>
342
343 <varlistentry>
344 <term><computeroutput>-z --compress</computeroutput></term>
345 <listitem><para>The complement to
346  <computeroutput>-d</computeroutput>: forces compression,
347  regardless of the invokation name.</para></listitem>
348 </varlistentry>
349
350 <varlistentry>
351 <term><computeroutput>-t --test</computeroutput></term>
352 <listitem><para>Check integrity of the specified file(s), but
353  don't decompress them.  This really performs a trial
354  decompression and throws away the result.</para></listitem>
355 </varlistentry>
356
357 <varlistentry>
358 <term><computeroutput>-f --force</computeroutput></term>
359 <listitem><para>Force overwrite of output files.  Normally,
360  <computeroutput>bzip2</computeroutput> will not overwrite
361  existing output files.  Also forces
362  <computeroutput>bzip2</computeroutput> to break hard links to
363  files, which it otherwise wouldn't do.</para>
364  <para><computeroutput>bzip2</computeroutput> normally declines
365  to decompress files which don't have the correct magic header
366  bytes. If forced (<computeroutput>-f</computeroutput>),
367  however, it will pass such files through unmodified. This is
368  how GNU <computeroutput>gzip</computeroutput> behaves.</para>
369 </listitem>
370 </varlistentry>
371
372 <varlistentry>
373 <term><computeroutput>-k --keep</computeroutput></term>
374 <listitem><para>Keep (don't delete) input files during
375  compression or decompression.</para></listitem>
376 </varlistentry>
377
378 <varlistentry>
379 <term><computeroutput>-s --small</computeroutput></term>
380 <listitem><para>Reduce memory usage, for compression,
381  decompression and testing.  Files are decompressed and tested
382  using a modified algorithm which only requires 2.5 bytes per
383  block byte.  This means any file can be decompressed in 2300k
384  of memory, albeit at about half the normal speed.</para>
385  <para>During compression, <computeroutput>-s</computeroutput>
386  selects a block size of 200k, which limits memory use to around
387  the same figure, at the expense of your compression ratio.  In
388  short, if your machine is low on memory (8 megabytes or less),
389  use <computeroutput>-s</computeroutput> for everything.  See
390  <xref linkend="memory-management"/> below.</para></listitem>
391 </varlistentry>
392
393 <varlistentry>
394 <term><computeroutput>-q --quiet</computeroutput></term>
395 <listitem><para>Suppress non-essential warning messages.
396  Messages pertaining to I/O errors and other critical events
397  will not be suppressed.</para></listitem>
398 </varlistentry>
399
400 <varlistentry>
401 <term><computeroutput>-v --verbose</computeroutput></term>
402 <listitem><para>Verbose mode -- show the compression ratio for
403  each file processed.  Further
404  <computeroutput>-v</computeroutput>'s increase the verbosity
405  level, spewing out lots of information which is primarily of
406  interest for diagnostic purposes.</para></listitem>
407 </varlistentry>
408
409 <varlistentry>
410 <term><computeroutput>-L --license -V --version</computeroutput></term>
411 <listitem><para>Display the software version, license terms and
412  conditions.</para></listitem>
413 </varlistentry>
414
415 <varlistentry>
416 <term><computeroutput>-1</computeroutput> (or
417 <computeroutput>--fast</computeroutput>) to
418 <computeroutput>-9</computeroutput> (or
419 <computeroutput>-best</computeroutput>)</term>
420 <listitem><para>Set the block size to 100 k, 200 k ...  900 k
421  when compressing.  Has no effect when decompressing.  See <xref
422  linkend="memory-management" /> below.  The
423  <computeroutput>--fast</computeroutput> and
424  <computeroutput>--best</computeroutput> aliases are primarily
425  for GNU <computeroutput>gzip</computeroutput> compatibility.
426  In particular, <computeroutput>--fast</computeroutput> doesn't
427  make things significantly faster.  And
428  <computeroutput>--best</computeroutput> merely selects the
429  default behaviour.</para></listitem>
430 </varlistentry>
431
432 <varlistentry>
433 <term><computeroutput>--</computeroutput></term>
434 <listitem><para>Treats all subsequent arguments as file names,
435  even if they start with a dash.  This is so you can handle
436  files with names beginning with a dash, for example:
437  <computeroutput>bzip2 --
438  -myfilename</computeroutput>.</para></listitem>
439 </varlistentry>
440
441 <varlistentry>
442 <term><computeroutput>--repetitive-fast</computeroutput></term>
443 <term><computeroutput>--repetitive-best</computeroutput></term>
444 <listitem><para>These flags are redundant in versions 0.9.5 and
445  above.  They provided some coarse control over the behaviour of
446  the sorting algorithm in earlier versions, which was sometimes
447  useful.  0.9.5 and above have an improved algorithm which
448  renders these flags irrelevant.</para></listitem>
449 </varlistentry>
450
451</variablelist>
452
453</sect1>
454
455
456<sect1 id="memory-management" xreflabel="MEMORY MANAGEMENT">
457<title>MEMORY MANAGEMENT</title>
458
459<para><computeroutput>bzip2</computeroutput> compresses large
460files in blocks.  The block size affects both the compression
461ratio achieved, and the amount of memory needed for compression
462and decompression.  The flags <computeroutput>-1</computeroutput>
463through <computeroutput>-9</computeroutput> specify the block
464size to be 100,000 bytes through 900,000 bytes (the default)
465respectively.  At decompression time, the block size used for
466compression is read from the header of the compressed file, and
467<computeroutput>bunzip2</computeroutput> then allocates itself
468just enough memory to decompress the file.  Since block sizes are
469stored in compressed files, it follows that the flags
470<computeroutput>-1</computeroutput> to
471<computeroutput>-9</computeroutput> are irrelevant to and so
472ignored during decompression.</para>
473
474<para>Compression and decompression requirements, in bytes, can be
475estimated as:</para>
476<programlisting>
477Compression:   400k + ( 8 x block size )
478
479Decompression: 100k + ( 4 x block size ), or
480               100k + ( 2.5 x block size )
481</programlisting>
482
483<para>Larger block sizes give rapidly diminishing marginal
484returns.  Most of the compression comes from the first two or
485three hundred k of block size, a fact worth bearing in mind when
486using <computeroutput>bzip2</computeroutput> on small machines.
487It is also important to appreciate that the decompression memory
488requirement is set at compression time by the choice of block
489size.</para>
490
491<para>For files compressed with the default 900k block size,
492<computeroutput>bunzip2</computeroutput> will require about 3700
493kbytes to decompress.  To support decompression of any file on a
4944 megabyte machine, <computeroutput>bunzip2</computeroutput> has
495an option to decompress using approximately half this amount of
496memory, about 2300 kbytes.  Decompression speed is also halved,
497so you should use this option only where necessary.  The relevant
498flag is <computeroutput>-s</computeroutput>.</para>
499
500<para>In general, try and use the largest block size memory
501constraints allow, since that maximises the compression achieved.
502Compression and decompression speed are virtually unaffected by
503block size.</para>
504
505<para>Another significant point applies to files which fit in a
506single block -- that means most files you'd encounter using a
507large block size.  The amount of real memory touched is
508proportional to the size of the file, since the file is smaller
509than a block.  For example, compressing a file 20,000 bytes long
510with the flag <computeroutput>-9</computeroutput> will cause the
511compressor to allocate around 7600k of memory, but only touch
512400k + 20000 * 8 = 560 kbytes of it.  Similarly, the decompressor
513will allocate 3700k but only touch 100k + 20000 * 4 = 180
514kbytes.</para>
515
516<para>Here is a table which summarises the maximum memory usage
517for different block sizes.  Also recorded is the total compressed
518size for 14 files of the Calgary Text Compression Corpus
519totalling 3,141,622 bytes.  This column gives some feel for how
520compression varies with block size.  These figures tend to
521understate the advantage of larger block sizes for larger files,
522since the Corpus is dominated by smaller files.</para>
523
524<programlisting>
525        Compress   Decompress   Decompress   Corpus
526Flag     usage      usage       -s usage     Size
527
528 -1      1200k       500k         350k      914704
529 -2      2000k       900k         600k      877703
530 -3      2800k      1300k         850k      860338
531 -4      3600k      1700k        1100k      846899
532 -5      4400k      2100k        1350k      845160
533 -6      5200k      2500k        1600k      838626
534 -7      6100k      2900k        1850k      834096
535 -8      6800k      3300k        2100k      828642
536 -9      7600k      3700k        2350k      828642
537</programlisting>
538
539</sect1>
540
541
542<sect1 id="recovering" xreflabel="RECOVERING DATA FROM DAMAGED FILES">
543<title>RECOVERING DATA FROM DAMAGED FILES</title>
544
545<para><computeroutput>bzip2</computeroutput> compresses files in
546blocks, usually 900kbytes long.  Each block is handled
547independently.  If a media or transmission error causes a
548multi-block <computeroutput>.bz2</computeroutput> file to become
549damaged, it may be possible to recover data from the undamaged
550blocks in the file.</para>
551
552<para>The compressed representation of each block is delimited by
553a 48-bit pattern, which makes it possible to find the block
554boundaries with reasonable certainty.  Each block also carries
555its own 32-bit CRC, so damaged blocks can be distinguished from
556undamaged ones.</para>
557
558<para><computeroutput>bzip2recover</computeroutput> is a simple
559program whose purpose is to search for blocks in
560<computeroutput>.bz2</computeroutput> files, and write each block
561out into its own <computeroutput>.bz2</computeroutput> file.  You
562can then use <computeroutput>bzip2 -t</computeroutput> to test
563the integrity of the resulting files, and decompress those which
564are undamaged.</para>
565
566<para><computeroutput>bzip2recover</computeroutput> takes a
567single argument, the name of the damaged file, and writes a
568number of files <computeroutput>rec0001file.bz2</computeroutput>,
569<computeroutput>rec0002file.bz2</computeroutput>, etc, containing
570the extracted blocks.  The output filenames are designed so that
571the use of wildcards in subsequent processing -- for example,
572<computeroutput>bzip2 -dc rec*file.bz2 &#62;
573recovered_data</computeroutput> -- lists the files in the correct
574order.</para>
575
576<para><computeroutput>bzip2recover</computeroutput> should be of
577most use dealing with large <computeroutput>.bz2</computeroutput>
578files, as these will contain many blocks.  It is clearly futile
579to use it on damaged single-block files, since a damaged block
580cannot be recovered.  If you wish to minimise any potential data
581loss through media or transmission errors, you might consider
582compressing with a smaller block size.</para>
583
584</sect1>
585
586
587<sect1 id="performance" xreflabel="PERFORMANCE NOTES">
588<title>PERFORMANCE NOTES</title>
589
590<para>The sorting phase of compression gathers together similar
591strings in the file.  Because of this, files containing very long
592runs of repeated symbols, like "aabaabaabaab ..."  (repeated
593several hundred times) may compress more slowly than normal.
594Versions 0.9.5 and above fare much better than previous versions
595in this respect.  The ratio between worst-case and average-case
596compression time is in the region of 10:1.  For previous
597versions, this figure was more like 100:1.  You can use the
598<computeroutput>-vvvv</computeroutput> option to monitor progress
599in great detail, if you want.</para>
600
601<para>Decompression speed is unaffected by these
602phenomena.</para>
603
604<para><computeroutput>bzip2</computeroutput> usually allocates
605several megabytes of memory to operate in, and then charges all
606over it in a fairly random fashion.  This means that performance,
607both for compressing and decompressing, is largely determined by
608the speed at which your machine can service cache misses.
609Because of this, small changes to the code to reduce the miss
610rate have been observed to give disproportionately large
611performance improvements.  I imagine
612<computeroutput>bzip2</computeroutput> will perform best on
613machines with very large caches.</para>
614
615</sect1>
616
617
618
619<sect1 id="caveats" xreflabel="CAVEATS">
620<title>CAVEATS</title>
621
622<para>I/O error messages are not as helpful as they could be.
623<computeroutput>bzip2</computeroutput> tries hard to detect I/O
624errors and exit cleanly, but the details of what the problem is
625sometimes seem rather misleading.</para>
626
627<para>This manual page pertains to version &bz-version; of
628<computeroutput>bzip2</computeroutput>.  Compressed data created by
629this version is entirely forwards and backwards compatible with the
630previous public releases, versions 0.1pl2, 0.9.0 and 0.9.5, 1.0.0,
6311.0.1, 1.0.2 and 1.0.3, but with the following exception: 0.9.0 and
632above can correctly decompress multiple concatenated compressed files.
6330.1pl2 cannot do this; it will stop after decompressing just the first
634file in the stream.</para>
635
636<para><computeroutput>bzip2recover</computeroutput> versions
637prior to 1.0.2 used 32-bit integers to represent bit positions in
638compressed files, so it could not handle compressed files more
639than 512 megabytes long.  Versions 1.0.2 and above use 64-bit ints
640on some platforms which support them (GNU supported targets, and
641Windows). To establish whether or not
642<computeroutput>bzip2recover</computeroutput> was built with such
643a limitation, run it without arguments. In any event you can
644build yourself an unlimited version if you can recompile it with
645<computeroutput>MaybeUInt64</computeroutput> set to be an
646unsigned 64-bit integer.</para>
647
648</sect1>
649
650
651
652<sect1 id="author" xreflabel="AUTHOR">
653<title>AUTHOR</title>
654
655<para>Julian Seward,
656<computeroutput>&bz-email;</computeroutput></para>
657
658<para>The ideas embodied in
659<computeroutput>bzip2</computeroutput> are due to (at least) the
660following people: Michael Burrows and David Wheeler (for the
661block sorting transformation), David Wheeler (again, for the
662Huffman coder), Peter Fenwick (for the structured coding model in
663the original <computeroutput>bzip</computeroutput>, and many
664refinements), and Alistair Moffat, Radford Neal and Ian Witten
665(for the arithmetic coder in the original
666<computeroutput>bzip</computeroutput>).  I am much indebted for
667their help, support and advice.  See the manual in the source
668distribution for pointers to sources of documentation.  Christian
669von Roques encouraged me to look for faster sorting algorithms,
670so as to speed up compression.  Bela Lubkin encouraged me to
671improve the worst-case compression performance.
672Donna Robinson XMLised the documentation.
673Many people sent
674patches, helped with portability problems, lent machines, gave
675advice and were generally helpful.</para>
676
677</sect1>
678
679</chapter>
680
681
682
683<chapter id="libprog" xreflabel="Programming with libbzip2">
684<title>
685Programming with <computeroutput>libbzip2</computeroutput>
686</title>
687
688<para>This chapter describes the programming interface to
689<computeroutput>libbzip2</computeroutput>.</para>
690
691<para>For general background information, particularly about
692memory use and performance aspects, you'd be well advised to read
693<xref linkend="using"/> as well.</para>
694
695
696<sect1 id="top-level" xreflabel="Top-level structure">
697<title>Top-level structure</title>
698
699<para><computeroutput>libbzip2</computeroutput> is a flexible
700library for compressing and decompressing data in the
701<computeroutput>bzip2</computeroutput> data format.  Although
702packaged as a single entity, it helps to regard the library as
703three separate parts: the low level interface, and the high level
704interface, and some utility functions.</para>
705
706<para>The structure of
707<computeroutput>libbzip2</computeroutput>'s interfaces is similar
708to that of Jean-loup Gailly's and Mark Adler's excellent
709<computeroutput>zlib</computeroutput> library.</para>
710
711<para>All externally visible symbols have names beginning
712<computeroutput>BZ2_</computeroutput>.  This is new in version
7131.0.  The intention is to minimise pollution of the namespaces of
714library clients.</para>
715
716<para>To use any part of the library, you need to
717<computeroutput>#include &lt;bzlib.h&gt;</computeroutput>
718into your sources.</para>
719
720
721
722<sect2 id="ll-summary" xreflabel="Low-level summary">
723<title>Low-level summary</title>
724
725<para>This interface provides services for compressing and
726decompressing data in memory.  There's no provision for dealing
727with files, streams or any other I/O mechanisms, just straight
728memory-to-memory work.  In fact, this part of the library can be
729compiled without inclusion of
730<computeroutput>stdio.h</computeroutput>, which may be helpful
731for embedded applications.</para>
732
733<para>The low-level part of the library has no global variables
734and is therefore thread-safe.</para>
735
736<para>Six routines make up the low level interface:
737<computeroutput>BZ2_bzCompressInit</computeroutput>,
738<computeroutput>BZ2_bzCompress</computeroutput>, and
739<computeroutput>BZ2_bzCompressEnd</computeroutput> for
740compression, and a corresponding trio
741<computeroutput>BZ2_bzDecompressInit</computeroutput>,
742<computeroutput>BZ2_bzDecompress</computeroutput> and
743<computeroutput>BZ2_bzDecompressEnd</computeroutput> for
744decompression.  The <computeroutput>*Init</computeroutput>
745functions allocate memory for compression/decompression and do
746other initialisations, whilst the
747<computeroutput>*End</computeroutput> functions close down
748operations and release memory.</para>
749
750<para>The real work is done by
751<computeroutput>BZ2_bzCompress</computeroutput> and
752<computeroutput>BZ2_bzDecompress</computeroutput>.  These
753compress and decompress data from a user-supplied input buffer to
754a user-supplied output buffer.  These buffers can be any size;
755arbitrary quantities of data are handled by making repeated calls
756to these functions.  This is a flexible mechanism allowing a
757consumer-pull style of activity, or producer-push, or a mixture
758of both.</para>
759
760</sect2>
761
762
763<sect2 id="hl-summary" xreflabel="High-level summary">
764<title>High-level summary</title>
765
766<para>This interface provides some handy wrappers around the
767low-level interface to facilitate reading and writing
768<computeroutput>bzip2</computeroutput> format files
769(<computeroutput>.bz2</computeroutput> files).  The routines
770provide hooks to facilitate reading files in which the
771<computeroutput>bzip2</computeroutput> data stream is embedded
772within some larger-scale file structure, or where there are
773multiple <computeroutput>bzip2</computeroutput> data streams
774concatenated end-to-end.</para>
775
776<para>For reading files,
777<computeroutput>BZ2_bzReadOpen</computeroutput>,
778<computeroutput>BZ2_bzRead</computeroutput>,
779<computeroutput>BZ2_bzReadClose</computeroutput> and
780<computeroutput>BZ2_bzReadGetUnused</computeroutput> are
781supplied.  For writing files,
782<computeroutput>BZ2_bzWriteOpen</computeroutput>,
783<computeroutput>BZ2_bzWrite</computeroutput> and
784<computeroutput>BZ2_bzWriteFinish</computeroutput> are
785available.</para>
786
787<para>As with the low-level library, no global variables are used
788so the library is per se thread-safe.  However, if I/O errors
789occur whilst reading or writing the underlying compressed files,
790you may have to consult <computeroutput>errno</computeroutput> to
791determine the cause of the error.  In that case, you'd need a C
792library which correctly supports
793<computeroutput>errno</computeroutput> in a multithreaded
794environment.</para>
795
796<para>To make the library a little simpler and more portable,
797<computeroutput>BZ2_bzReadOpen</computeroutput> and
798<computeroutput>BZ2_bzWriteOpen</computeroutput> require you to
799pass them file handles (<computeroutput>FILE*</computeroutput>s)
800which have previously been opened for reading or writing
801respectively.  That avoids portability problems associated with
802file operations and file attributes, whilst not being much of an
803imposition on the programmer.</para>
804
805</sect2>
806
807
808<sect2 id="util-fns-summary" xreflabel="Utility functions summary">
809<title>Utility functions summary</title>
810
811<para>For very simple needs,
812<computeroutput>BZ2_bzBuffToBuffCompress</computeroutput> and
813<computeroutput>BZ2_bzBuffToBuffDecompress</computeroutput> are
814provided.  These compress data in memory from one buffer to
815another buffer in a single function call.  You should assess
816whether these functions fulfill your memory-to-memory
817compression/decompression requirements before investing effort in
818understanding the more general but more complex low-level
819interface.</para>
820
821<para>Yoshioka Tsuneo
822(<computeroutput>tsuneo@rr.iij4u.or.jp</computeroutput>) has
823contributed some functions to give better
824<computeroutput>zlib</computeroutput> compatibility.  These
825functions are <computeroutput>BZ2_bzopen</computeroutput>,
826<computeroutput>BZ2_bzread</computeroutput>,
827<computeroutput>BZ2_bzwrite</computeroutput>,
828<computeroutput>BZ2_bzflush</computeroutput>,
829<computeroutput>BZ2_bzclose</computeroutput>,
830<computeroutput>BZ2_bzerror</computeroutput> and
831<computeroutput>BZ2_bzlibVersion</computeroutput>.  You may find
832these functions more convenient for simple file reading and
833writing, than those in the high-level interface.  These functions
834are not (yet) officially part of the library, and are minimally
835documented here.  If they break, you get to keep all the pieces.
836I hope to document them properly when time permits.</para>
837
838<para>Yoshioka also contributed modifications to allow the
839library to be built as a Windows DLL.</para>
840
841</sect2>
842
843</sect1>
844
845
846<sect1 id="err-handling" xreflabel="Error handling">
847<title>Error handling</title>
848
849<para>The library is designed to recover cleanly in all
850situations, including the worst-case situation of decompressing
851random data.  I'm not 100% sure that it can always do this, so
852you might want to add a signal handler to catch segmentation
853violations during decompression if you are feeling especially
854paranoid.  I would be interested in hearing more about the
855robustness of the library to corrupted compressed data.</para>
856
857<para>Version 1.0.3 more robust in this respect than any
858previous version.  Investigations with Valgrind (a tool for detecting
859problems with memory management) indicate
860that, at least for the few files I tested, all single-bit errors
861in the decompressed data are caught properly, with no
862segmentation faults, no uses of uninitialised data, no out of
863range reads or writes, and no infinite looping in the decompressor.
864So it's certainly pretty robust, although
865I wouldn't claim it to be totally bombproof.</para>
866
867<para>The file <computeroutput>bzlib.h</computeroutput> contains
868all definitions needed to use the library.  In particular, you
869should definitely not include
870<computeroutput>bzlib_private.h</computeroutput>.</para>
871
872<para>In <computeroutput>bzlib.h</computeroutput>, the various
873return values are defined.  The following list is not intended as
874an exhaustive description of the circumstances in which a given
875value may be returned -- those descriptions are given later.
876Rather, it is intended to convey the rough meaning of each return
877value.  The first five actions are normal and not intended to
878denote an error situation.</para>
879
880<variablelist>
881
882 <varlistentry>
883  <term><computeroutput>BZ_OK</computeroutput></term>
884  <listitem><para>The requested action was completed
885   successfully.</para></listitem>
886 </varlistentry>
887
888 <varlistentry>
889  <term><computeroutput>BZ_RUN_OK, BZ_FLUSH_OK,
890    BZ_FINISH_OK</computeroutput></term>
891  <listitem><para>In
892   <computeroutput>BZ2_bzCompress</computeroutput>, the requested
893   flush/finish/nothing-special action was completed
894   successfully.</para></listitem>
895 </varlistentry>
896
897 <varlistentry>
898  <term><computeroutput>BZ_STREAM_END</computeroutput></term>
899  <listitem><para>Compression of data was completed, or the
900   logical stream end was detected during
901   decompression.</para></listitem>
902 </varlistentry>
903
904</variablelist>
905
906<para>The following return values indicate an error of some
907kind.</para>
908
909<variablelist>
910
911 <varlistentry>
912  <term><computeroutput>BZ_CONFIG_ERROR</computeroutput></term>
913  <listitem><para>Indicates that the library has been improperly
914   compiled on your platform -- a major configuration error.
915   Specifically, it means that
916   <computeroutput>sizeof(char)</computeroutput>,
917   <computeroutput>sizeof(short)</computeroutput> and
918   <computeroutput>sizeof(int)</computeroutput> are not 1, 2 and
919   4 respectively, as they should be.  Note that the library
920   should still work properly on 64-bit platforms which follow
921   the LP64 programming model -- that is, where
922   <computeroutput>sizeof(long)</computeroutput> and
923   <computeroutput>sizeof(void*)</computeroutput> are 8.  Under
924   LP64, <computeroutput>sizeof(int)</computeroutput> is still 4,
925   so <computeroutput>libbzip2</computeroutput>, which doesn't
926   use the <computeroutput>long</computeroutput> type, is
927   OK.</para></listitem>
928 </varlistentry>
929
930 <varlistentry>
931  <term><computeroutput>BZ_SEQUENCE_ERROR</computeroutput></term>
932  <listitem><para>When using the library, it is important to call
933   the functions in the correct sequence and with data structures
934   (buffers etc) in the correct states.
935   <computeroutput>libbzip2</computeroutput> checks as much as it
936   can to ensure this is happening, and returns
937   <computeroutput>BZ_SEQUENCE_ERROR</computeroutput> if not.
938   Code which complies precisely with the function semantics, as
939   detailed below, should never receive this value; such an event
940   denotes buggy code which you should
941   investigate.</para></listitem>
942 </varlistentry>
943
944 <varlistentry>
945  <term><computeroutput>BZ_PARAM_ERROR</computeroutput></term>
946  <listitem><para>Returned when a parameter to a function call is
947   out of range or otherwise manifestly incorrect.  As with
948   <computeroutput>BZ_SEQUENCE_ERROR</computeroutput>, this
949   denotes a bug in the client code.  The distinction between
950   <computeroutput>BZ_PARAM_ERROR</computeroutput> and
951   <computeroutput>BZ_SEQUENCE_ERROR</computeroutput> is a bit
952   hazy, but still worth making.</para></listitem>
953 </varlistentry>
954
955 <varlistentry>
956  <term><computeroutput>BZ_MEM_ERROR</computeroutput></term>
957  <listitem><para>Returned when a request to allocate memory
958   failed.  Note that the quantity of memory needed to decompress
959   a stream cannot be determined until the stream's header has
960   been read.  So
961   <computeroutput>BZ2_bzDecompress</computeroutput> and
962   <computeroutput>BZ2_bzRead</computeroutput> may return
963   <computeroutput>BZ_MEM_ERROR</computeroutput> even though some
964   of the compressed data has been read.  The same is not true
965   for compression; once
966   <computeroutput>BZ2_bzCompressInit</computeroutput> or
967   <computeroutput>BZ2_bzWriteOpen</computeroutput> have
968   successfully completed,
969   <computeroutput>BZ_MEM_ERROR</computeroutput> cannot
970   occur.</para></listitem>
971 </varlistentry>
972
973 <varlistentry>
974  <term><computeroutput>BZ_DATA_ERROR</computeroutput></term>
975  <listitem><para>Returned when a data integrity error is
976   detected during decompression.  Most importantly, this means
977   when stored and computed CRCs for the data do not match.  This
978   value is also returned upon detection of any other anomaly in
979   the compressed data.</para></listitem>
980 </varlistentry>
981
982 <varlistentry>
983  <term><computeroutput>BZ_DATA_ERROR_MAGIC</computeroutput></term>
984  <listitem><para>As a special case of
985   <computeroutput>BZ_DATA_ERROR</computeroutput>, it is
986   sometimes useful to know when the compressed stream does not
987   start with the correct magic bytes (<computeroutput>'B' 'Z'
988   'h'</computeroutput>).</para></listitem>
989 </varlistentry>
990
991 <varlistentry>
992  <term><computeroutput>BZ_IO_ERROR</computeroutput></term>
993  <listitem><para>Returned by
994   <computeroutput>BZ2_bzRead</computeroutput> and
995   <computeroutput>BZ2_bzWrite</computeroutput> when there is an
996   error reading or writing in the compressed file, and by
997   <computeroutput>BZ2_bzReadOpen</computeroutput> and
998   <computeroutput>BZ2_bzWriteOpen</computeroutput> for attempts
999   to use a file for which the error indicator (viz,
1000   <computeroutput>ferror(f)</computeroutput>) is set.  On
1001   receipt of <computeroutput>BZ_IO_ERROR</computeroutput>, the
1002   caller should consult <computeroutput>errno</computeroutput>
1003   and/or <computeroutput>perror</computeroutput> to acquire
1004   operating-system specific information about the
1005   problem.</para></listitem>
1006 </varlistentry>
1007
1008 <varlistentry>
1009  <term><computeroutput>BZ_UNEXPECTED_EOF</computeroutput></term>
1010  <listitem><para>Returned by
1011   <computeroutput>BZ2_bzRead</computeroutput> when the
1012   compressed file finishes before the logical end of stream is
1013   detected.</para></listitem>
1014 </varlistentry>
1015
1016 <varlistentry>
1017  <term><computeroutput>BZ_OUTBUFF_FULL</computeroutput></term>
1018  <listitem><para>Returned by
1019   <computeroutput>BZ2_bzBuffToBuffCompress</computeroutput> and
1020   <computeroutput>BZ2_bzBuffToBuffDecompress</computeroutput> to
1021   indicate that the output data will not fit into the output
1022   buffer provided.</para></listitem>
1023 </varlistentry>
1024
1025</variablelist>
1026
1027</sect1>
1028
1029
1030
1031<sect1 id="low-level" xreflabel=">Low-level interface">
1032<title>Low-level interface</title>
1033
1034
1035<sect2 id="bzcompress-init" xreflabel="BZ2_bzCompressInit">
1036<title>BZ2_bzCompressInit</title>
1037
1038<programlisting>
1039typedef struct {
1040  char *next_in;
1041  unsigned int avail_in;
1042  unsigned int total_in_lo32;
1043  unsigned int total_in_hi32;
1044
1045  char *next_out;
1046  unsigned int avail_out;
1047  unsigned int total_out_lo32;
1048  unsigned int total_out_hi32;
1049
1050  void *state;
1051
1052  void *(*bzalloc)(void *,int,int);
1053  void (*bzfree)(void *,void *);
1054  void *opaque;
1055} bz_stream;
1056
1057int BZ2_bzCompressInit ( bz_stream *strm,
1058                         int blockSize100k,
1059                         int verbosity,
1060                         int workFactor );
1061</programlisting>
1062
1063<para>Prepares for compression.  The
1064<computeroutput>bz_stream</computeroutput> structure holds all
1065data pertaining to the compression activity.  A
1066<computeroutput>bz_stream</computeroutput> structure should be
1067allocated and initialised prior to the call.  The fields of
1068<computeroutput>bz_stream</computeroutput> comprise the entirety
1069of the user-visible data.  <computeroutput>state</computeroutput>
1070is a pointer to the private data structures required for
1071compression.</para>
1072
1073<para>Custom memory allocators are supported, via fields
1074<computeroutput>bzalloc</computeroutput>,
1075<computeroutput>bzfree</computeroutput>, and
1076<computeroutput>opaque</computeroutput>.  The value
1077<computeroutput>opaque</computeroutput> is passed to as the first
1078argument to all calls to <computeroutput>bzalloc</computeroutput>
1079and <computeroutput>bzfree</computeroutput>, but is otherwise
1080ignored by the library.  The call <computeroutput>bzalloc (
1081opaque, n, m )</computeroutput> is expected to return a pointer
1082<computeroutput>p</computeroutput> to <computeroutput>n *
1083m</computeroutput> bytes of memory, and <computeroutput>bzfree (
1084opaque, p )</computeroutput> should free that memory.</para>
1085
1086<para>If you don't want to use a custom memory allocator, set
1087<computeroutput>bzalloc</computeroutput>,
1088<computeroutput>bzfree</computeroutput> and
1089<computeroutput>opaque</computeroutput> to
1090<computeroutput>NULL</computeroutput>, and the library will then
1091use the standard <computeroutput>malloc</computeroutput> /
1092<computeroutput>free</computeroutput> routines.</para>
1093
1094<para>Before calling
1095<computeroutput>BZ2_bzCompressInit</computeroutput>, fields
1096<computeroutput>bzalloc</computeroutput>,
1097<computeroutput>bzfree</computeroutput> and
1098<computeroutput>opaque</computeroutput> should be filled
1099appropriately, as just described.  Upon return, the internal
1100state will have been allocated and initialised, and
1101<computeroutput>total_in_lo32</computeroutput>,
1102<computeroutput>total_in_hi32</computeroutput>,
1103<computeroutput>total_out_lo32</computeroutput> and
1104<computeroutput>total_out_hi32</computeroutput> will have been
1105set to zero.  These four fields are used by the library to inform
1106the caller of the total amount of data passed into and out of the
1107library, respectively.  You should not try to change them.  As of
1108version 1.0, 64-bit counts are maintained, even on 32-bit
1109platforms, using the <computeroutput>_hi32</computeroutput>
1110fields to store the upper 32 bits of the count.  So, for example,
1111the total amount of data in is <computeroutput>(total_in_hi32
1112&#60;&#60; 32) + total_in_lo32</computeroutput>.</para>
1113
1114<para>Parameter <computeroutput>blockSize100k</computeroutput>
1115specifies the block size to be used for compression.  It should
1116be a value between 1 and 9 inclusive, and the actual block size
1117used is 100000 x this figure.  9 gives the best compression but
1118takes most memory.</para>
1119
1120<para>Parameter <computeroutput>verbosity</computeroutput> should
1121be set to a number between 0 and 4 inclusive.  0 is silent, and
1122greater numbers give increasingly verbose monitoring/debugging
1123output.  If the library has been compiled with
1124<computeroutput>-DBZ_NO_STDIO</computeroutput>, no such output
1125will appear for any verbosity setting.</para>
1126
1127<para>Parameter <computeroutput>workFactor</computeroutput>
1128controls how the compression phase behaves when presented with
1129worst case, highly repetitive, input data.  If compression runs
1130into difficulties caused by repetitive data, the library switches
1131from the standard sorting algorithm to a fallback algorithm.  The
1132fallback is slower than the standard algorithm by perhaps a
1133factor of three, but always behaves reasonably, no matter how bad
1134the input.</para>
1135
1136<para>Lower values of <computeroutput>workFactor</computeroutput>
1137reduce the amount of effort the standard algorithm will expend
1138before resorting to the fallback.  You should set this parameter
1139carefully; too low, and many inputs will be handled by the
1140fallback algorithm and so compress rather slowly, too high, and
1141your average-to-worst case compression times can become very
1142large.  The default value of 30 gives reasonable behaviour over a
1143wide range of circumstances.</para>
1144
1145<para>Allowable values range from 0 to 250 inclusive.  0 is a
1146special case, equivalent to using the default value of 30.</para>
1147
1148<para>Note that the compressed output generated is the same
1149regardless of whether or not the fallback algorithm is
1150used.</para>
1151
1152<para>Be aware also that this parameter may disappear entirely in
1153future versions of the library.  In principle it should be
1154possible to devise a good way to automatically choose which
1155algorithm to use.  Such a mechanism would render the parameter
1156obsolete.</para>
1157
1158<para>Possible return values:</para>
1159
1160<programlisting>
1161BZ_CONFIG_ERROR
1162  if the library has been mis-compiled
1163BZ_PARAM_ERROR
1164  if strm is NULL
1165  or blockSize < 1 or blockSize > 9
1166  or verbosity < 0 or verbosity > 4
1167  or workFactor < 0 or workFactor > 250
1168BZ_MEM_ERROR
1169  if not enough memory is available
1170BZ_OK
1171  otherwise
1172</programlisting>
1173
1174<para>Allowable next actions:</para>
1175
1176<programlisting>
1177BZ2_bzCompress
1178  if BZ_OK is returned
1179  no specific action needed in case of error
1180</programlisting>
1181
1182</sect2>
1183
1184
1185<sect2 id="bzCompress" xreflabel="BZ2_bzCompress">
1186<title>BZ2_bzCompress</title>
1187
1188<programlisting>
1189int BZ2_bzCompress ( bz_stream *strm, int action );
1190</programlisting>
1191
1192<para>Provides more input and/or output buffer space for the
1193library.  The caller maintains input and output buffers, and
1194calls <computeroutput>BZ2_bzCompress</computeroutput> to transfer
1195data between them.</para>
1196
1197<para>Before each call to
1198<computeroutput>BZ2_bzCompress</computeroutput>,
1199<computeroutput>next_in</computeroutput> should point at the data
1200to be compressed, and <computeroutput>avail_in</computeroutput>
1201should indicate how many bytes the library may read.
1202<computeroutput>BZ2_bzCompress</computeroutput> updates
1203<computeroutput>next_in</computeroutput>,
1204<computeroutput>avail_in</computeroutput> and
1205<computeroutput>total_in</computeroutput> to reflect the number
1206of bytes it has read.</para>
1207
1208<para>Similarly, <computeroutput>next_out</computeroutput> should
1209point to a buffer in which the compressed data is to be placed,
1210with <computeroutput>avail_out</computeroutput> indicating how
1211much output space is available.
1212<computeroutput>BZ2_bzCompress</computeroutput> updates
1213<computeroutput>next_out</computeroutput>,
1214<computeroutput>avail_out</computeroutput> and
1215<computeroutput>total_out</computeroutput> to reflect the number
1216of bytes output.</para>
1217
1218<para>You may provide and remove as little or as much data as you
1219like on each call of
1220<computeroutput>BZ2_bzCompress</computeroutput>.  In the limit,
1221it is acceptable to supply and remove data one byte at a time,
1222although this would be terribly inefficient.  You should always
1223ensure that at least one byte of output space is available at
1224each call.</para>
1225
1226<para>A second purpose of
1227<computeroutput>BZ2_bzCompress</computeroutput> is to request a
1228change of mode of the compressed stream.</para>
1229
1230<para>Conceptually, a compressed stream can be in one of four
1231states: IDLE, RUNNING, FLUSHING and FINISHING.  Before
1232initialisation
1233(<computeroutput>BZ2_bzCompressInit</computeroutput>) and after
1234termination (<computeroutput>BZ2_bzCompressEnd</computeroutput>),
1235a stream is regarded as IDLE.</para>
1236
1237<para>Upon initialisation
1238(<computeroutput>BZ2_bzCompressInit</computeroutput>), the stream
1239is placed in the RUNNING state.  Subsequent calls to
1240<computeroutput>BZ2_bzCompress</computeroutput> should pass
1241<computeroutput>BZ_RUN</computeroutput> as the requested action;
1242other actions are illegal and will result in
1243<computeroutput>BZ_SEQUENCE_ERROR</computeroutput>.</para>
1244
1245<para>At some point, the calling program will have provided all
1246the input data it wants to.  It will then want to finish up -- in
1247effect, asking the library to process any data it might have
1248buffered internally.  In this state,
1249<computeroutput>BZ2_bzCompress</computeroutput> will no longer
1250attempt to read data from
1251<computeroutput>next_in</computeroutput>, but it will want to
1252write data to <computeroutput>next_out</computeroutput>.  Because
1253the output buffer supplied by the user can be arbitrarily small,
1254the finishing-up operation cannot necessarily be done with a
1255single call of
1256<computeroutput>BZ2_bzCompress</computeroutput>.</para>
1257
1258<para>Instead, the calling program passes
1259<computeroutput>BZ_FINISH</computeroutput> as an action to
1260<computeroutput>BZ2_bzCompress</computeroutput>.  This changes
1261the stream's state to FINISHING.  Any remaining input (ie,
1262<computeroutput>next_in[0 .. avail_in-1]</computeroutput>) is
1263compressed and transferred to the output buffer.  To do this,
1264<computeroutput>BZ2_bzCompress</computeroutput> must be called
1265repeatedly until all the output has been consumed.  At that
1266point, <computeroutput>BZ2_bzCompress</computeroutput> returns
1267<computeroutput>BZ_STREAM_END</computeroutput>, and the stream's
1268state is set back to IDLE.
1269<computeroutput>BZ2_bzCompressEnd</computeroutput> should then be
1270called.</para>
1271
1272<para>Just to make sure the calling program does not cheat, the
1273library makes a note of <computeroutput>avail_in</computeroutput>
1274at the time of the first call to
1275<computeroutput>BZ2_bzCompress</computeroutput> which has
1276<computeroutput>BZ_FINISH</computeroutput> as an action (ie, at
1277the time the program has announced its intention to not supply
1278any more input).  By comparing this value with that of
1279<computeroutput>avail_in</computeroutput> over subsequent calls
1280to <computeroutput>BZ2_bzCompress</computeroutput>, the library
1281can detect any attempts to slip in more data to compress.  Any
1282calls for which this is detected will return
1283<computeroutput>BZ_SEQUENCE_ERROR</computeroutput>.  This
1284indicates a programming mistake which should be corrected.</para>
1285
1286<para>Instead of asking to finish, the calling program may ask
1287<computeroutput>BZ2_bzCompress</computeroutput> to take all the
1288remaining input, compress it and terminate the current
1289(Burrows-Wheeler) compression block.  This could be useful for
1290error control purposes.  The mechanism is analogous to that for
1291finishing: call <computeroutput>BZ2_bzCompress</computeroutput>
1292with an action of <computeroutput>BZ_FLUSH</computeroutput>,
1293remove output data, and persist with the
1294<computeroutput>BZ_FLUSH</computeroutput> action until the value
1295<computeroutput>BZ_RUN</computeroutput> is returned.  As with
1296finishing, <computeroutput>BZ2_bzCompress</computeroutput>
1297detects any attempt to provide more input data once the flush has
1298begun.</para>
1299
1300<para>Once the flush is complete, the stream returns to the
1301normal RUNNING state.</para>
1302
1303<para>This all sounds pretty complex, but isn't really.  Here's a
1304table which shows which actions are allowable in each state, what
1305action will be taken, what the next state is, and what the
1306non-error return values are.  Note that you can't explicitly ask
1307what state the stream is in, but nor do you need to -- it can be
1308inferred from the values returned by
1309<computeroutput>BZ2_bzCompress</computeroutput>.</para>
1310
1311<programlisting>
1312IDLE/any
1313  Illegal.  IDLE state only exists after BZ2_bzCompressEnd or
1314  before BZ2_bzCompressInit.
1315  Return value = BZ_SEQUENCE_ERROR
1316
1317RUNNING/BZ_RUN
1318  Compress from next_in to next_out as much as possible.
1319  Next state = RUNNING
1320  Return value = BZ_RUN_OK
1321
1322RUNNING/BZ_FLUSH
1323  Remember current value of next_in. Compress from next_in
1324  to next_out as much as possible, but do not accept any more input.
1325  Next state = FLUSHING
1326  Return value = BZ_FLUSH_OK
1327
1328RUNNING/BZ_FINISH
1329  Remember current value of next_in. Compress from next_in
1330  to next_out as much as possible, but do not accept any more input.
1331  Next state = FINISHING
1332  Return value = BZ_FINISH_OK
1333
1334FLUSHING/BZ_FLUSH
1335  Compress from next_in to next_out as much as possible,
1336  but do not accept any more input.
1337  If all the existing input has been used up and all compressed
1338  output has been removed
1339    Next state = RUNNING; Return value = BZ_RUN_OK
1340  else
1341    Next state = FLUSHING; Return value = BZ_FLUSH_OK
1342
1343FLUSHING/other
1344  Illegal.
1345  Return value = BZ_SEQUENCE_ERROR
1346
1347FINISHING/BZ_FINISH
1348  Compress from next_in to next_out as much as possible,
1349  but to not accept any more input.
1350  If all the existing input has been used up and all compressed
1351  output has been removed
1352    Next state = IDLE; Return value = BZ_STREAM_END
1353  else
1354    Next state = FINISHING; Return value = BZ_FINISH_OK
1355
1356FINISHING/other
1357  Illegal.
1358  Return value = BZ_SEQUENCE_ERROR
1359</programlisting>
1360
1361
1362<para>That still looks complicated?  Well, fair enough.  The
1363usual sequence of calls for compressing a load of data is:</para>
1364
1365<orderedlist>
1366
1367 <listitem><para>Get started with
1368  <computeroutput>BZ2_bzCompressInit</computeroutput>.</para></listitem>
1369
1370 <listitem><para>Shovel data in and shlurp out its compressed form
1371  using zero or more calls of
1372  <computeroutput>BZ2_bzCompress</computeroutput> with action =
1373  <computeroutput>BZ_RUN</computeroutput>.</para></listitem>
1374
1375 <listitem><para>Finish up. Repeatedly call
1376  <computeroutput>BZ2_bzCompress</computeroutput> with action =
1377  <computeroutput>BZ_FINISH</computeroutput>, copying out the
1378  compressed output, until
1379  <computeroutput>BZ_STREAM_END</computeroutput> is
1380  returned.</para></listitem> <listitem><para>Close up and go home.  Call
1381  <computeroutput>BZ2_bzCompressEnd</computeroutput>.</para></listitem>
1382
1383</orderedlist>
1384
1385<para>If the data you want to compress fits into your input
1386buffer all at once, you can skip the calls of
1387<computeroutput>BZ2_bzCompress ( ..., BZ_RUN )</computeroutput>
1388and just do the <computeroutput>BZ2_bzCompress ( ..., BZ_FINISH
1389)</computeroutput> calls.</para>
1390
1391<para>All required memory is allocated by
1392<computeroutput>BZ2_bzCompressInit</computeroutput>.  The
1393compression library can accept any data at all (obviously).  So
1394you shouldn't get any error return values from the
1395<computeroutput>BZ2_bzCompress</computeroutput> calls.  If you
1396do, they will be
1397<computeroutput>BZ_SEQUENCE_ERROR</computeroutput>, and indicate
1398a bug in your programming.</para>
1399
1400<para>Trivial other possible return values:</para>
1401
1402<programlisting>
1403BZ_PARAM_ERROR
1404  if strm is NULL, or strm->s is NULL
1405</programlisting>
1406
1407</sect2>
1408
1409
1410<sect2 id="bzCompress-end" xreflabel="BZ2_bzCompressEnd">
1411<title>BZ2_bzCompressEnd</title>
1412
1413<programlisting>
1414int BZ2_bzCompressEnd ( bz_stream *strm );
1415</programlisting>
1416
1417<para>Releases all memory associated with a compression
1418stream.</para>
1419
1420<para>Possible return values:</para>
1421
1422<programlisting>
1423BZ_PARAM_ERROR  if strm is NULL or strm->s is NULL
1424BZ_OK           otherwise
1425</programlisting>
1426
1427</sect2>
1428
1429
1430<sect2 id="bzDecompress-init" xreflabel="BZ2_bzDecompressInit">
1431<title>BZ2_bzDecompressInit</title>
1432
1433<programlisting>
1434int BZ2_bzDecompressInit ( bz_stream *strm, int verbosity, int small );
1435</programlisting>
1436
1437<para>Prepares for decompression.  As with
1438<computeroutput>BZ2_bzCompressInit</computeroutput>, a
1439<computeroutput>bz_stream</computeroutput> record should be
1440allocated and initialised before the call.  Fields
1441<computeroutput>bzalloc</computeroutput>,
1442<computeroutput>bzfree</computeroutput> and
1443<computeroutput>opaque</computeroutput> should be set if a custom
1444memory allocator is required, or made
1445<computeroutput>NULL</computeroutput> for the normal
1446<computeroutput>malloc</computeroutput> /
1447<computeroutput>free</computeroutput> routines.  Upon return, the
1448internal state will have been initialised, and
1449<computeroutput>total_in</computeroutput> and
1450<computeroutput>total_out</computeroutput> will be zero.</para>
1451
1452<para>For the meaning of parameter
1453<computeroutput>verbosity</computeroutput>, see
1454<computeroutput>BZ2_bzCompressInit</computeroutput>.</para>
1455
1456<para>If <computeroutput>small</computeroutput> is nonzero, the
1457library will use an alternative decompression algorithm which
1458uses less memory but at the cost of decompressing more slowly
1459(roughly speaking, half the speed, but the maximum memory
1460requirement drops to around 2300k).  See <xref linkend="using"/>
1461for more information on memory management.</para>
1462
1463<para>Note that the amount of memory needed to decompress a
1464stream cannot be determined until the stream's header has been
1465read, so even if
1466<computeroutput>BZ2_bzDecompressInit</computeroutput> succeeds, a
1467subsequent <computeroutput>BZ2_bzDecompress</computeroutput>
1468could fail with
1469<computeroutput>BZ_MEM_ERROR</computeroutput>.</para>
1470
1471<para>Possible return values:</para>
1472
1473<programlisting>
1474BZ_CONFIG_ERROR
1475  if the library has been mis-compiled
1476BZ_PARAM_ERROR
1477  if ( small != 0 && small != 1 )
1478  or (verbosity <; 0 || verbosity > 4)
1479BZ_MEM_ERROR
1480  if insufficient memory is available
1481</programlisting>
1482
1483<para>Allowable next actions:</para>
1484
1485<programlisting>
1486BZ2_bzDecompress
1487  if BZ_OK was returned
1488  no specific action required in case of error
1489</programlisting>
1490
1491</sect2>
1492
1493
1494<sect2 id="bzDecompress" xreflabel="BZ2_bzDecompress">
1495<title>BZ2_bzDecompress</title>
1496
1497<programlisting>
1498int BZ2_bzDecompress ( bz_stream *strm );
1499</programlisting>
1500
1501<para>Provides more input and/out output buffer space for the
1502library.  The caller maintains input and output buffers, and uses
1503<computeroutput>BZ2_bzDecompress</computeroutput> to transfer
1504data between them.</para>
1505
1506<para>Before each call to
1507<computeroutput>BZ2_bzDecompress</computeroutput>,
1508<computeroutput>next_in</computeroutput> should point at the
1509compressed data, and <computeroutput>avail_in</computeroutput>
1510should indicate how many bytes the library may read.
1511<computeroutput>BZ2_bzDecompress</computeroutput> updates
1512<computeroutput>next_in</computeroutput>,
1513<computeroutput>avail_in</computeroutput> and
1514<computeroutput>total_in</computeroutput> to reflect the number
1515of bytes it has read.</para>
1516
1517<para>Similarly, <computeroutput>next_out</computeroutput> should
1518point to a buffer in which the uncompressed output is to be
1519placed, with <computeroutput>avail_out</computeroutput>
1520indicating how much output space is available.
1521<computeroutput>BZ2_bzCompress</computeroutput> updates
1522<computeroutput>next_out</computeroutput>,
1523<computeroutput>avail_out</computeroutput> and
1524<computeroutput>total_out</computeroutput> to reflect the number
1525of bytes output.</para>
1526
1527<para>You may provide and remove as little or as much data as you
1528like on each call of
1529<computeroutput>BZ2_bzDecompress</computeroutput>.  In the limit,
1530it is acceptable to supply and remove data one byte at a time,
1531although this would be terribly inefficient.  You should always
1532ensure that at least one byte of output space is available at
1533each call.</para>
1534
1535<para>Use of <computeroutput>BZ2_bzDecompress</computeroutput> is
1536simpler than
1537<computeroutput>BZ2_bzCompress</computeroutput>.</para>
1538
1539<para>You should provide input and remove output as described
1540above, and repeatedly call
1541<computeroutput>BZ2_bzDecompress</computeroutput> until
1542<computeroutput>BZ_STREAM_END</computeroutput> is returned.
1543Appearance of <computeroutput>BZ_STREAM_END</computeroutput>
1544denotes that <computeroutput>BZ2_bzDecompress</computeroutput>
1545has detected the logical end of the compressed stream.
1546<computeroutput>BZ2_bzDecompress</computeroutput> will not
1547produce <computeroutput>BZ_STREAM_END</computeroutput> until all
1548output data has been placed into the output buffer, so once
1549<computeroutput>BZ_STREAM_END</computeroutput> appears, you are
1550guaranteed to have available all the decompressed output, and
1551<computeroutput>BZ2_bzDecompressEnd</computeroutput> can safely
1552be called.</para>
1553
1554<para>If case of an error return value, you should call
1555<computeroutput>BZ2_bzDecompressEnd</computeroutput> to clean up
1556and release memory.</para>
1557
1558<para>Possible return values:</para>
1559
1560<programlisting>
1561BZ_PARAM_ERROR
1562  if strm is NULL or strm->s is NULL
1563  or strm->avail_out < 1
1564BZ_DATA_ERROR
1565  if a data integrity error is detected in the compressed stream
1566BZ_DATA_ERROR_MAGIC
1567  if the compressed stream doesn't begin with the right magic bytes
1568BZ_MEM_ERROR
1569  if there wasn't enough memory available
1570BZ_STREAM_END
1571  if the logical end of the data stream was detected and all
1572  output in has been consumed, eg s-->avail_out > 0
1573BZ_OK
1574  otherwise
1575</programlisting>
1576
1577<para>Allowable next actions:</para>
1578
1579<programlisting>
1580BZ2_bzDecompress
1581  if BZ_OK was returned
1582BZ2_bzDecompressEnd
1583  otherwise
1584</programlisting>
1585
1586</sect2>
1587
1588
1589<sect2 id="bzDecompress-end" xreflabel="BZ2_bzDecompressEnd">
1590<title>BZ2_bzDecompressEnd</title>
1591
1592<programlisting>
1593int BZ2_bzDecompressEnd ( bz_stream *strm );
1594</programlisting>
1595
1596<para>Releases all memory associated with a decompression
1597stream.</para>
1598
1599<para>Possible return values:</para>
1600
1601<programlisting>
1602BZ_PARAM_ERROR
1603  if strm is NULL or strm->s is NULL
1604BZ_OK
1605  otherwise
1606</programlisting>
1607
1608<para>Allowable next actions:</para>
1609
1610<programlisting>
1611  None.
1612</programlisting>
1613
1614</sect2>
1615
1616</sect1>
1617
1618
1619<sect1 id="hl-interface" xreflabel="High-level interface">
1620<title>High-level interface</title>
1621
1622<para>This interface provides functions for reading and writing
1623<computeroutput>bzip2</computeroutput> format files.  First, some
1624general points.</para>
1625
1626<itemizedlist mark='bullet'>
1627
1628 <listitem><para>All of the functions take an
1629  <computeroutput>int*</computeroutput> first argument,
1630  <computeroutput>bzerror</computeroutput>.  After each call,
1631  <computeroutput>bzerror</computeroutput> should be consulted
1632  first to determine the outcome of the call.  If
1633  <computeroutput>bzerror</computeroutput> is
1634  <computeroutput>BZ_OK</computeroutput>, the call completed
1635  successfully, and only then should the return value of the
1636  function (if any) be consulted.  If
1637  <computeroutput>bzerror</computeroutput> is
1638  <computeroutput>BZ_IO_ERROR</computeroutput>, there was an
1639  error reading/writing the underlying compressed file, and you
1640  should then consult <computeroutput>errno</computeroutput> /
1641  <computeroutput>perror</computeroutput> to determine the cause
1642  of the difficulty.  <computeroutput>bzerror</computeroutput>
1643  may also be set to various other values; precise details are
1644  given on a per-function basis below.</para></listitem>
1645
1646 <listitem><para>If <computeroutput>bzerror</computeroutput> indicates
1647  an error (ie, anything except
1648  <computeroutput>BZ_OK</computeroutput> and
1649  <computeroutput>BZ_STREAM_END</computeroutput>), you should
1650  immediately call
1651  <computeroutput>BZ2_bzReadClose</computeroutput> (or
1652  <computeroutput>BZ2_bzWriteClose</computeroutput>, depending on
1653  whether you are attempting to read or to write) to free up all
1654  resources associated with the stream.  Once an error has been
1655  indicated, behaviour of all calls except
1656  <computeroutput>BZ2_bzReadClose</computeroutput>
1657  (<computeroutput>BZ2_bzWriteClose</computeroutput>) is
1658  undefined.  The implication is that (1)
1659  <computeroutput>bzerror</computeroutput> should be checked
1660  after each call, and (2) if
1661  <computeroutput>bzerror</computeroutput> indicates an error,
1662  <computeroutput>BZ2_bzReadClose</computeroutput>
1663  (<computeroutput>BZ2_bzWriteClose</computeroutput>) should then
1664  be called to clean up.</para></listitem>
1665
1666 <listitem><para>The <computeroutput>FILE*</computeroutput> arguments
1667  passed to <computeroutput>BZ2_bzReadOpen</computeroutput> /
1668  <computeroutput>BZ2_bzWriteOpen</computeroutput> should be set
1669  to binary mode.  Most Unix systems will do this by default, but
1670  other platforms, including Windows and Mac, will not.  If you
1671  omit this, you may encounter problems when moving code to new
1672  platforms.</para></listitem>
1673
1674 <listitem><para>Memory allocation requests are handled by
1675  <computeroutput>malloc</computeroutput> /
1676  <computeroutput>free</computeroutput>.  At present there is no
1677  facility for user-defined memory allocators in the file I/O
1678  functions (could easily be added, though).</para></listitem>
1679
1680</itemizedlist>
1681
1682
1683
1684<sect2 id="bzreadopen" xreflabel="BZ2_bzReadOpen">
1685<title>BZ2_bzReadOpen</title>
1686
1687<programlisting>
1688typedef void BZFILE;
1689
1690BZFILE *BZ2_bzReadOpen( int *bzerror, FILE *f,
1691                        int verbosity, int small,
1692                        void *unused, int nUnused );
1693</programlisting>
1694
1695<para>Prepare to read compressed data from file handle
1696<computeroutput>f</computeroutput>.
1697<computeroutput>f</computeroutput> should refer to a file which
1698has been opened for reading, and for which the error indicator
1699(<computeroutput>ferror(f)</computeroutput>)is not set.  If
1700<computeroutput>small</computeroutput> is 1, the library will try
1701to decompress using less memory, at the expense of speed.</para>
1702
1703<para>For reasons explained below,
1704<computeroutput>BZ2_bzRead</computeroutput> will decompress the
1705<computeroutput>nUnused</computeroutput> bytes starting at
1706<computeroutput>unused</computeroutput>, before starting to read
1707from the file <computeroutput>f</computeroutput>.  At most
1708<computeroutput>BZ_MAX_UNUSED</computeroutput> bytes may be
1709supplied like this.  If this facility is not required, you should
1710pass <computeroutput>NULL</computeroutput> and
1711<computeroutput>0</computeroutput> for
1712<computeroutput>unused</computeroutput> and
1713n<computeroutput>Unused</computeroutput> respectively.</para>
1714
1715<para>For the meaning of parameters
1716<computeroutput>small</computeroutput> and
1717<computeroutput>verbosity</computeroutput>, see
1718<computeroutput>BZ2_bzDecompressInit</computeroutput>.</para>
1719
1720<para>The amount of memory needed to decompress a file cannot be
1721determined until the file's header has been read.  So it is
1722possible that <computeroutput>BZ2_bzReadOpen</computeroutput>
1723returns <computeroutput>BZ_OK</computeroutput> but a subsequent
1724call of <computeroutput>BZ2_bzRead</computeroutput> will return
1725<computeroutput>BZ_MEM_ERROR</computeroutput>.</para>
1726
1727<para>Possible assignments to
1728<computeroutput>bzerror</computeroutput>:</para>
1729
1730<programlisting>
1731BZ_CONFIG_ERROR
1732  if the library has been mis-compiled
1733BZ_PARAM_ERROR
1734  if f is NULL
1735  or small is neither 0 nor 1
1736  or ( unused == NULL && nUnused != 0 )
1737  or ( unused != NULL && !(0 <= nUnused <= BZ_MAX_UNUSED) )
1738BZ_IO_ERROR
1739  if ferror(f) is nonzero
1740BZ_MEM_ERROR
1741  if insufficient memory is available
1742BZ_OK
1743  otherwise.
1744</programlisting>
1745
1746<para>Possible return values:</para>
1747
1748<programlisting>
1749Pointer to an abstract BZFILE
1750  if bzerror is BZ_OK
1751NULL
1752  otherwise
1753</programlisting>
1754
1755<para>Allowable next actions:</para>
1756
1757<programlisting>
1758BZ2_bzRead
1759  if bzerror is BZ_OK
1760BZ2_bzClose
1761  otherwise
1762</programlisting>
1763
1764</sect2>
1765
1766
1767<sect2 id="bzread" xreflabel="BZ2_bzRead">
1768<title>BZ2_bzRead</title>
1769
1770<programlisting>
1771int BZ2_bzRead ( int *bzerror, BZFILE *b, void *buf, int len );
1772</programlisting>
1773
1774<para>Reads up to <computeroutput>len</computeroutput>
1775(uncompressed) bytes from the compressed file
1776<computeroutput>b</computeroutput> into the buffer
1777<computeroutput>buf</computeroutput>.  If the read was
1778successful, <computeroutput>bzerror</computeroutput> is set to
1779<computeroutput>BZ_OK</computeroutput> and the number of bytes
1780read is returned.  If the logical end-of-stream was detected,
1781<computeroutput>bzerror</computeroutput> will be set to
1782<computeroutput>BZ_STREAM_END</computeroutput>, and the number of
1783bytes read is returned.  All other
1784<computeroutput>bzerror</computeroutput> values denote an
1785error.</para>
1786
1787<para><computeroutput>BZ2_bzRead</computeroutput> will supply
1788<computeroutput>len</computeroutput> bytes, unless the logical
1789stream end is detected or an error occurs.  Because of this, it
1790is possible to detect the stream end by observing when the number
1791of bytes returned is less than the number requested.
1792Nevertheless, this is regarded as inadvisable; you should instead
1793check <computeroutput>bzerror</computeroutput> after every call
1794and watch out for
1795<computeroutput>BZ_STREAM_END</computeroutput>.</para>
1796
1797<para>Internally, <computeroutput>BZ2_bzRead</computeroutput>
1798copies data from the compressed file in chunks of size
1799<computeroutput>BZ_MAX_UNUSED</computeroutput> bytes before
1800decompressing it.  If the file contains more bytes than strictly
1801needed to reach the logical end-of-stream,
1802<computeroutput>BZ2_bzRead</computeroutput> will almost certainly
1803read some of the trailing data before signalling
1804<computeroutput>BZ_SEQUENCE_END</computeroutput>.  To collect the
1805read but unused data once
1806<computeroutput>BZ_SEQUENCE_END</computeroutput> has appeared,
1807call <computeroutput>BZ2_bzReadGetUnused</computeroutput>
1808immediately before
1809<computeroutput>BZ2_bzReadClose</computeroutput>.</para>
1810
1811<para>Possible assignments to
1812<computeroutput>bzerror</computeroutput>:</para>
1813
1814<programlisting>
1815BZ_PARAM_ERROR
1816  if b is NULL or buf is NULL or len < 0
1817BZ_SEQUENCE_ERROR
1818  if b was opened with BZ2_bzWriteOpen
1819BZ_IO_ERROR
1820  if there is an error reading from the compressed file
1821BZ_UNEXPECTED_EOF
1822  if the compressed file ended before
1823  the logical end-of-stream was detected
1824BZ_DATA_ERROR
1825  if a data integrity error was detected in the compressed stream
1826BZ_DATA_ERROR_MAGIC
1827  if the stream does not begin with the requisite header bytes
1828  (ie, is not a bzip2 data file).  This is really
1829  a special case of BZ_DATA_ERROR.
1830BZ_MEM_ERROR
1831  if insufficient memory was available
1832BZ_STREAM_END
1833  if the logical end of stream was detected.
1834BZ_OK
1835  otherwise.
1836</programlisting>
1837
1838<para>Possible return values:</para>
1839
1840<programlisting>
1841number of bytes read
1842  if bzerror is BZ_OK or BZ_STREAM_END
1843undefined
1844  otherwise
1845</programlisting>
1846
1847<para>Allowable next actions:</para>
1848
1849<programlisting>
1850collect data from buf, then BZ2_bzRead or BZ2_bzReadClose
1851  if bzerror is BZ_OK
1852collect data from buf, then BZ2_bzReadClose or BZ2_bzReadGetUnused
1853  if bzerror is BZ_SEQUENCE_END
1854BZ2_bzReadClose
1855  otherwise
1856</programlisting>
1857
1858</sect2>
1859
1860
1861<sect2 id="bzreadgetunused" xreflabel="BZ2_bzReadGetUnused">
1862<title>BZ2_bzReadGetUnused</title>
1863
1864<programlisting>
1865void BZ2_bzReadGetUnused( int* bzerror, BZFILE *b,
1866                          void** unused, int* nUnused );
1867</programlisting>
1868
1869<para>Returns data which was read from the compressed file but
1870was not needed to get to the logical end-of-stream.
1871<computeroutput>*unused</computeroutput> is set to the address of
1872the data, and <computeroutput>*nUnused</computeroutput> to the
1873number of bytes.  <computeroutput>*nUnused</computeroutput> will
1874be set to a value between <computeroutput>0</computeroutput> and
1875<computeroutput>BZ_MAX_UNUSED</computeroutput> inclusive.</para>
1876
1877<para>This function may only be called once
1878<computeroutput>BZ2_bzRead</computeroutput> has signalled
1879<computeroutput>BZ_STREAM_END</computeroutput> but before
1880<computeroutput>BZ2_bzReadClose</computeroutput>.</para>
1881
1882<para>Possible assignments to
1883<computeroutput>bzerror</computeroutput>:</para>
1884
1885<programlisting>
1886BZ_PARAM_ERROR
1887  if b is NULL
1888  or unused is NULL or nUnused is NULL
1889BZ_SEQUENCE_ERROR
1890  if BZ_STREAM_END has not been signalled
1891  or if b was opened with BZ2_bzWriteOpen
1892BZ_OK
1893  otherwise
1894</programlisting>
1895
1896<para>Allowable next actions:</para>
1897
1898<programlisting>
1899BZ2_bzReadClose
1900</programlisting>
1901
1902</sect2>
1903
1904
1905<sect2 id="bzreadclose" xreflabel="BZ2_bzReadClose">
1906<title>BZ2_bzReadClose</title>
1907
1908<programlisting>
1909void BZ2_bzReadClose ( int *bzerror, BZFILE *b );
1910</programlisting>
1911
1912<para>Releases all memory pertaining to the compressed file
1913<computeroutput>b</computeroutput>.
1914<computeroutput>BZ2_bzReadClose</computeroutput> does not call
1915<computeroutput>fclose</computeroutput> on the underlying file
1916handle, so you should do that yourself if appropriate.
1917<computeroutput>BZ2_bzReadClose</computeroutput> should be called
1918to clean up after all error situations.</para>
1919
1920<para>Possible assignments to
1921<computeroutput>bzerror</computeroutput>:</para>
1922
1923<programlisting>
1924BZ_SEQUENCE_ERROR
1925  if b was opened with BZ2_bzOpenWrite
1926BZ_OK
1927  otherwise
1928</programlisting>
1929
1930<para>Allowable next actions:</para>
1931
1932<programlisting>
1933none
1934</programlisting>
1935
1936</sect2>
1937
1938
1939<sect2 id="bzwriteopen" xreflabel="BZ2_bzWriteOpen">
1940<title>BZ2_bzWriteOpen</title>
1941
1942<programlisting>
1943BZFILE *BZ2_bzWriteOpen( int *bzerror, FILE *f,
1944                         int blockSize100k, int verbosity,
1945                         int workFactor );
1946</programlisting>
1947
1948<para>Prepare to write compressed data to file handle
1949<computeroutput>f</computeroutput>.
1950<computeroutput>f</computeroutput> should refer to a file which
1951has been opened for writing, and for which the error indicator
1952(<computeroutput>ferror(f)</computeroutput>)is not set.</para>
1953
1954<para>For the meaning of parameters
1955<computeroutput>blockSize100k</computeroutput>,
1956<computeroutput>verbosity</computeroutput> and
1957<computeroutput>workFactor</computeroutput>, see
1958<computeroutput>BZ2_bzCompressInit</computeroutput>.</para>
1959
1960<para>All required memory is allocated at this stage, so if the
1961call completes successfully,
1962<computeroutput>BZ_MEM_ERROR</computeroutput> cannot be signalled
1963by a subsequent call to
1964<computeroutput>BZ2_bzWrite</computeroutput>.</para>
1965
1966<para>Possible assignments to
1967<computeroutput>bzerror</computeroutput>:</para>
1968
1969<programlisting>
1970BZ_CONFIG_ERROR
1971  if the library has been mis-compiled
1972BZ_PARAM_ERROR
1973  if f is NULL
1974  or blockSize100k < 1 or blockSize100k > 9
1975BZ_IO_ERROR
1976  if ferror(f) is nonzero
1977BZ_MEM_ERROR
1978  if insufficient memory is available
1979BZ_OK
1980  otherwise
1981</programlisting>
1982
1983<para>Possible return values:</para>
1984
1985<programlisting>
1986Pointer to an abstract BZFILE
1987  if bzerror is BZ_OK
1988NULL
1989  otherwise
1990</programlisting>
1991
1992<para>Allowable next actions:</para>
1993
1994<programlisting>
1995BZ2_bzWrite
1996  if bzerror is BZ_OK
1997  (you could go directly to BZ2_bzWriteClose, but this would be pretty pointless)
1998BZ2_bzWriteClose
1999  otherwise
2000</programlisting>
2001
2002</sect2>
2003
2004
2005<sect2 id="bzwrite" xreflabel="BZ2_bzWrite">
2006<title>BZ2_bzWrite</title>
2007
2008<programlisting>
2009void BZ2_bzWrite ( int *bzerror, BZFILE *b, void *buf, int len );
2010</programlisting>
2011
2012<para>Absorbs <computeroutput>len</computeroutput> bytes from the
2013buffer <computeroutput>buf</computeroutput>, eventually to be
2014compressed and written to the file.</para>
2015
2016<para>Possible assignments to
2017<computeroutput>bzerror</computeroutput>:</para>
2018
2019<programlisting>
2020BZ_PARAM_ERROR
2021  if b is NULL or buf is NULL or len < 0
2022BZ_SEQUENCE_ERROR
2023  if b was opened with BZ2_bzReadOpen
2024BZ_IO_ERROR
2025  if there is an error writing the compressed file.
2026BZ_OK
2027  otherwise
2028</programlisting>
2029
2030</sect2>
2031
2032
2033<sect2 id="bzwriteclose" xreflabel="BZ2_bzWriteClose">
2034<title>BZ2_bzWriteClose</title>
2035
2036<programlisting>
2037void BZ2_bzWriteClose( int *bzerror, BZFILE* f,
2038                       int abandon,
2039                       unsigned int* nbytes_in,
2040                       unsigned int* nbytes_out );
2041
2042void BZ2_bzWriteClose64( int *bzerror, BZFILE* f,
2043                         int abandon,
2044                         unsigned int* nbytes_in_lo32,
2045                         unsigned int* nbytes_in_hi32,
2046                         unsigned int* nbytes_out_lo32,
2047                         unsigned int* nbytes_out_hi32 );
2048</programlisting>
2049
2050<para>Compresses and flushes to the compressed file all data so
2051far supplied by <computeroutput>BZ2_bzWrite</computeroutput>.
2052The logical end-of-stream markers are also written, so subsequent
2053calls to <computeroutput>BZ2_bzWrite</computeroutput> are
2054illegal.  All memory associated with the compressed file
2055<computeroutput>b</computeroutput> is released.
2056<computeroutput>fflush</computeroutput> is called on the
2057compressed file, but it is not
2058<computeroutput>fclose</computeroutput>'d.</para>
2059
2060<para>If <computeroutput>BZ2_bzWriteClose</computeroutput> is
2061called to clean up after an error, the only action is to release
2062the memory.  The library records the error codes issued by
2063previous calls, so this situation will be detected automatically.
2064There is no attempt to complete the compression operation, nor to
2065<computeroutput>fflush</computeroutput> the compressed file.  You
2066can force this behaviour to happen even in the case of no error,
2067by passing a nonzero value to
2068<computeroutput>abandon</computeroutput>.</para>
2069
2070<para>If <computeroutput>nbytes_in</computeroutput> is non-null,
2071<computeroutput>*nbytes_in</computeroutput> will be set to be the
2072total volume of uncompressed data handled.  Similarly,
2073<computeroutput>nbytes_out</computeroutput> will be set to the
2074total volume of compressed data written.  For compatibility with
2075older versions of the library,
2076<computeroutput>BZ2_bzWriteClose</computeroutput> only yields the
2077lower 32 bits of these counts.  Use
2078<computeroutput>BZ2_bzWriteClose64</computeroutput> if you want
2079the full 64 bit counts.  These two functions are otherwise
2080absolutely identical.</para>
2081
2082<para>Possible assignments to
2083<computeroutput>bzerror</computeroutput>:</para>
2084
2085<programlisting>
2086BZ_SEQUENCE_ERROR
2087  if b was opened with BZ2_bzReadOpen
2088BZ_IO_ERROR
2089  if there is an error writing the compressed file
2090BZ_OK
2091  otherwise
2092</programlisting>
2093
2094</sect2>
2095
2096
2097<sect2 id="embed" xreflabel="Handling embedded compressed data streams">
2098<title>Handling embedded compressed data streams</title>
2099
2100<para>The high-level library facilitates use of
2101<computeroutput>bzip2</computeroutput> data streams which form
2102some part of a surrounding, larger data stream.</para>
2103
2104<itemizedlist mark='bullet'>
2105
2106 <listitem><para>For writing, the library takes an open file handle,
2107  writes compressed data to it,
2108  <computeroutput>fflush</computeroutput>es it but does not
2109  <computeroutput>fclose</computeroutput> it.  The calling
2110  application can write its own data before and after the
2111  compressed data stream, using that same file handle.</para></listitem>
2112
2113 <listitem><para>Reading is more complex, and the facilities are not as
2114  general as they could be since generality is hard to reconcile
2115  with efficiency.  <computeroutput>BZ2_bzRead</computeroutput>
2116  reads from the compressed file in blocks of size
2117  <computeroutput>BZ_MAX_UNUSED</computeroutput> bytes, and in
2118  doing so probably will overshoot the logical end of compressed
2119  stream.  To recover this data once decompression has ended,
2120  call <computeroutput>BZ2_bzReadGetUnused</computeroutput> after
2121  the last call of <computeroutput>BZ2_bzRead</computeroutput>
2122  (the one returning
2123  <computeroutput>BZ_STREAM_END</computeroutput>) but before
2124  calling
2125  <computeroutput>BZ2_bzReadClose</computeroutput>.</para></listitem>
2126
2127</itemizedlist>
2128
2129<para>This mechanism makes it easy to decompress multiple
2130<computeroutput>bzip2</computeroutput> streams placed end-to-end.
2131As the end of one stream, when
2132<computeroutput>BZ2_bzRead</computeroutput> returns
2133<computeroutput>BZ_STREAM_END</computeroutput>, call
2134<computeroutput>BZ2_bzReadGetUnused</computeroutput> to collect
2135the unused data (copy it into your own buffer somewhere).  That
2136data forms the start of the next compressed stream.  To start
2137uncompressing that next stream, call
2138<computeroutput>BZ2_bzReadOpen</computeroutput> again, feeding in
2139the unused data via the <computeroutput>unused</computeroutput> /
2140<computeroutput>nUnused</computeroutput> parameters.  Keep doing
2141this until <computeroutput>BZ_STREAM_END</computeroutput> return
2142coincides with the physical end of file
2143(<computeroutput>feof(f)</computeroutput>).  In this situation
2144<computeroutput>BZ2_bzReadGetUnused</computeroutput> will of
2145course return no data.</para>
2146
2147<para>This should give some feel for how the high-level interface
2148can be used.  If you require extra flexibility, you'll have to
2149bite the bullet and get to grips with the low-level
2150interface.</para>
2151
2152</sect2>
2153
2154
2155<sect2 id="std-rdwr" xreflabel="Standard file-reading/writing code">
2156<title>Standard file-reading/writing code</title>
2157
2158<para>Here's how you'd write data to a compressed file:</para>
2159
2160<programlisting>
2161FILE*   f;
2162BZFILE* b;
2163int     nBuf;
2164char    buf[ /* whatever size you like */ ];
2165int     bzerror;
2166int     nWritten;
2167
2168f = fopen ( "myfile.bz2", "w" );
2169if ( !f ) {
2170 /* handle error */
2171}
2172b = BZ2_bzWriteOpen( &bzerror, f, 9 );
2173if (bzerror != BZ_OK) {
2174 BZ2_bzWriteClose ( b );
2175 /* handle error */
2176}
2177
2178while ( /* condition */ ) {
2179 /* get data to write into buf, and set nBuf appropriately */
2180 nWritten = BZ2_bzWrite ( &bzerror, b, buf, nBuf );
2181 if (bzerror == BZ_IO_ERROR) {
2182   BZ2_bzWriteClose ( &bzerror, b );
2183   /* handle error */
2184 }
2185}
2186
2187BZ2_bzWriteClose( &bzerror, b );
2188if (bzerror == BZ_IO_ERROR) {
2189 /* handle error */
2190}
2191</programlisting>
2192
2193<para>And to read from a compressed file:</para>
2194
2195<programlisting>
2196FILE*   f;
2197BZFILE* b;
2198int     nBuf;
2199char    buf[ /* whatever size you like */ ];
2200int     bzerror;
2201int     nWritten;
2202
2203f = fopen ( "myfile.bz2", "r" );
2204if ( !f ) {
2205  /* handle error */
2206}
2207b = BZ2_bzReadOpen ( &bzerror, f, 0, NULL, 0 );
2208if ( bzerror != BZ_OK ) {
2209  BZ2_bzReadClose ( &bzerror, b );
2210  /* handle error */
2211}
2212
2213bzerror = BZ_OK;
2214while ( bzerror == BZ_OK && /* arbitrary other conditions */) {
2215  nBuf = BZ2_bzRead ( &bzerror, b, buf, /* size of buf */ );
2216  if ( bzerror == BZ_OK ) {
2217    /* do something with buf[0 .. nBuf-1] */
2218  }
2219}
2220if ( bzerror != BZ_STREAM_END ) {
2221   BZ2_bzReadClose ( &bzerror, b );
2222   /* handle error */
2223} else {
2224   BZ2_bzReadClose ( &bzerror, b );
2225}
2226</programlisting>
2227
2228</sect2>
2229
2230</sect1>
2231
2232
2233<sect1 id="util-fns" xreflabel="Utility functions">
2234<title>Utility functions</title>
2235
2236
2237<sect2 id="bzbufftobuffcompress" xreflabel="BZ2_bzBuffToBuffCompress">
2238<title>BZ2_bzBuffToBuffCompress</title>
2239
2240<programlisting>
2241int BZ2_bzBuffToBuffCompress( char*         dest,
2242                              unsigned int* destLen,
2243                              char*         source,
2244                              unsigned int  sourceLen,
2245                              int           blockSize100k,
2246                              int           verbosity,
2247                              int           workFactor );
2248</programlisting>
2249
2250<para>Attempts to compress the data in <computeroutput>source[0
2251.. sourceLen-1]</computeroutput> into the destination buffer,
2252<computeroutput>dest[0 .. *destLen-1]</computeroutput>.  If the
2253destination buffer is big enough,
2254<computeroutput>*destLen</computeroutput> is set to the size of
2255the compressed data, and <computeroutput>BZ_OK</computeroutput>
2256is returned.  If the compressed data won't fit,
2257<computeroutput>*destLen</computeroutput> is unchanged, and
2258<computeroutput>BZ_OUTBUFF_FULL</computeroutput> is
2259returned.</para>
2260
2261<para>Compression in this manner is a one-shot event, done with a
2262single call to this function.  The resulting compressed data is a
2263complete <computeroutput>bzip2</computeroutput> format data
2264stream.  There is no mechanism for making additional calls to
2265provide extra input data.  If you want that kind of mechanism,
2266use the low-level interface.</para>
2267
2268<para>For the meaning of parameters
2269<computeroutput>blockSize100k</computeroutput>,
2270<computeroutput>verbosity</computeroutput> and
2271<computeroutput>workFactor</computeroutput>, see
2272<computeroutput>BZ2_bzCompressInit</computeroutput>.</para>
2273
2274<para>To guarantee that the compressed data will fit in its
2275buffer, allocate an output buffer of size 1% larger than the
2276uncompressed data, plus six hundred extra bytes.</para>
2277
2278<para><computeroutput>BZ2_bzBuffToBuffDecompress</computeroutput>
2279will not write data at or beyond
2280<computeroutput>dest[*destLen]</computeroutput>, even in case of
2281buffer overflow.</para>
2282
2283<para>Possible return values:</para>
2284
2285<programlisting>
2286BZ_CONFIG_ERROR
2287  if the library has been mis-compiled
2288BZ_PARAM_ERROR
2289  if dest is NULL or destLen is NULL
2290  or blockSize100k < 1 or blockSize100k > 9
2291  or verbosity < 0 or verbosity > 4
2292  or workFactor < 0 or workFactor > 250
2293BZ_MEM_ERROR
2294  if insufficient memory is available
2295BZ_OUTBUFF_FULL
2296  if the size of the compressed data exceeds *destLen
2297BZ_OK
2298  otherwise
2299</programlisting>
2300
2301</sect2>
2302
2303
2304<sect2 id="bzbufftobuffdecompress" xreflabel="BZ2_bzBuffToBuffDecompress">
2305<title>BZ2_bzBuffToBuffDecompress</title>
2306
2307<programlisting>
2308int BZ2_bzBuffToBuffDecompress( char*         dest,
2309                                unsigned int* destLen,
2310                                char*         source,
2311                                unsigned int  sourceLen,
2312                                int           small,
2313                                int           verbosity );
2314</programlisting>
2315
2316<para>Attempts to decompress the data in <computeroutput>source[0
2317.. sourceLen-1]</computeroutput> into the destination buffer,
2318<computeroutput>dest[0 .. *destLen-1]</computeroutput>.  If the
2319destination buffer is big enough,
2320<computeroutput>*destLen</computeroutput> is set to the size of
2321the uncompressed data, and <computeroutput>BZ_OK</computeroutput>
2322is returned.  If the compressed data won't fit,
2323<computeroutput>*destLen</computeroutput> is unchanged, and
2324<computeroutput>BZ_OUTBUFF_FULL</computeroutput> is
2325returned.</para>
2326
2327<para><computeroutput>source</computeroutput> is assumed to hold
2328a complete <computeroutput>bzip2</computeroutput> format data
2329stream.
2330<computeroutput>BZ2_bzBuffToBuffDecompress</computeroutput> tries
2331to decompress the entirety of the stream into the output
2332buffer.</para>
2333
2334<para>For the meaning of parameters
2335<computeroutput>small</computeroutput> and
2336<computeroutput>verbosity</computeroutput>, see
2337<computeroutput>BZ2_bzDecompressInit</computeroutput>.</para>
2338
2339<para>Because the compression ratio of the compressed data cannot
2340be known in advance, there is no easy way to guarantee that the
2341output buffer will be big enough.  You may of course make
2342arrangements in your code to record the size of the uncompressed
2343data, but such a mechanism is beyond the scope of this
2344library.</para>
2345
2346<para><computeroutput>BZ2_bzBuffToBuffDecompress</computeroutput>
2347will not write data at or beyond
2348<computeroutput>dest[*destLen]</computeroutput>, even in case of
2349buffer overflow.</para>
2350
2351<para>Possible return values:</para>
2352
2353<programlisting>
2354BZ_CONFIG_ERROR
2355  if the library has been mis-compiled
2356BZ_PARAM_ERROR
2357  if dest is NULL or destLen is NULL
2358  or small != 0 && small != 1
2359  or verbosity < 0 or verbosity > 4
2360BZ_MEM_ERROR
2361  if insufficient memory is available
2362BZ_OUTBUFF_FULL
2363  if the size of the compressed data exceeds *destLen
2364BZ_DATA_ERROR
2365  if a data integrity error was detected in the compressed data
2366BZ_DATA_ERROR_MAGIC
2367  if the compressed data doesn't begin with the right magic bytes
2368BZ_UNEXPECTED_EOF
2369  if the compressed data ends unexpectedly
2370BZ_OK
2371  otherwise
2372</programlisting>
2373
2374</sect2>
2375
2376</sect1>
2377
2378
2379<sect1 id="zlib-compat" xreflabel="zlib compatibility functions">
2380<title>zlib compatibility functions</title>
2381
2382<para>Yoshioka Tsuneo has contributed some functions to give
2383better <computeroutput>zlib</computeroutput> compatibility.
2384These functions are <computeroutput>BZ2_bzopen</computeroutput>,
2385<computeroutput>BZ2_bzread</computeroutput>,
2386<computeroutput>BZ2_bzwrite</computeroutput>,
2387<computeroutput>BZ2_bzflush</computeroutput>,
2388<computeroutput>BZ2_bzclose</computeroutput>,
2389<computeroutput>BZ2_bzerror</computeroutput> and
2390<computeroutput>BZ2_bzlibVersion</computeroutput>.  These
2391functions are not (yet) officially part of the library.  If they
2392break, you get to keep all the pieces.  Nevertheless, I think
2393they work ok.</para>
2394
2395<programlisting>
2396typedef void BZFILE;
2397
2398const char * BZ2_bzlibVersion ( void );
2399</programlisting>
2400
2401<para>Returns a string indicating the library version.</para>
2402
2403<programlisting>
2404BZFILE * BZ2_bzopen  ( const char *path, const char *mode );
2405BZFILE * BZ2_bzdopen ( int        fd,    const char *mode );
2406</programlisting>
2407
2408<para>Opens a <computeroutput>.bz2</computeroutput> file for
2409reading or writing, using either its name or a pre-existing file
2410descriptor.  Analogous to <computeroutput>fopen</computeroutput>
2411and <computeroutput>fdopen</computeroutput>.</para>
2412
2413<programlisting>
2414int BZ2_bzread  ( BZFILE* b, void* buf, int len );
2415int BZ2_bzwrite ( BZFILE* b, void* buf, int len );
2416</programlisting>
2417
2418<para>Reads/writes data from/to a previously opened
2419<computeroutput>BZFILE</computeroutput>.  Analogous to
2420<computeroutput>fread</computeroutput> and
2421<computeroutput>fwrite</computeroutput>.</para>
2422
2423<programlisting>
2424int  BZ2_bzflush ( BZFILE* b );
2425void BZ2_bzclose ( BZFILE* b );
2426</programlisting>
2427
2428<para>Flushes/closes a <computeroutput>BZFILE</computeroutput>.
2429<computeroutput>BZ2_bzflush</computeroutput> doesn't actually do
2430anything.  Analogous to <computeroutput>fflush</computeroutput>
2431and <computeroutput>fclose</computeroutput>.</para>
2432
2433<programlisting>
2434const char * BZ2_bzerror ( BZFILE *b, int *errnum )
2435</programlisting>
2436
2437<para>Returns a string describing the more recent error status of
2438<computeroutput>b</computeroutput>, and also sets
2439<computeroutput>*errnum</computeroutput> to its numerical
2440value.</para>
2441
2442</sect1>
2443
2444
2445<sect1 id="stdio-free"
2446       xreflabel="Using the library in a stdio-free environment">
2447<title>Using the library in a stdio-free environment</title>
2448
2449
2450<sect2 id="stdio-bye" xreflabel="Getting rid of stdio">
2451<title>Getting rid of stdio</title>
2452
2453<para>In a deeply embedded application, you might want to use
2454just the memory-to-memory functions.  You can do this
2455conveniently by compiling the library with preprocessor symbol
2456<computeroutput>BZ_NO_STDIO</computeroutput> defined.  Doing this
2457gives you a library containing only the following eight
2458functions:</para>
2459
2460<para><computeroutput>BZ2_bzCompressInit</computeroutput>,
2461<computeroutput>BZ2_bzCompress</computeroutput>,
2462<computeroutput>BZ2_bzCompressEnd</computeroutput>
2463<computeroutput>BZ2_bzDecompressInit</computeroutput>,
2464<computeroutput>BZ2_bzDecompress</computeroutput>,
2465<computeroutput>BZ2_bzDecompressEnd</computeroutput>
2466<computeroutput>BZ2_bzBuffToBuffCompress</computeroutput>,
2467<computeroutput>BZ2_bzBuffToBuffDecompress</computeroutput></para>
2468
2469<para>When compiled like this, all functions will ignore
2470<computeroutput>verbosity</computeroutput> settings.</para>
2471
2472</sect2>
2473
2474
2475<sect2 id="critical-error" xreflabel="Critical error handling">
2476<title>Critical error handling</title>
2477
2478<para><computeroutput>libbzip2</computeroutput> contains a number
2479of internal assertion checks which should, needless to say, never
2480be activated.  Nevertheless, if an assertion should fail,
2481behaviour depends on whether or not the library was compiled with
2482<computeroutput>BZ_NO_STDIO</computeroutput> set.</para>
2483
2484<para>For a normal compile, an assertion failure yields the
2485message:</para>
2486
2487<blockquote>
2488<para>bzip2/libbzip2: internal error number N.</para>
2489<para>This is a bug in bzip2/libbzip2, &bz-version; of &bz-date;.
2490Please report it to me at: &bz-email;.  If this happened
2491when you were using some program which uses libbzip2 as a
2492component, you should also report this bug to the author(s)
2493of that program.  Please make an effort to report this bug;
2494timely and accurate bug reports eventually lead to higher
2495quality software.  Thanks.  Julian Seward, &bz-date;.
2496</para></blockquote>
2497
2498<para>where <computeroutput>N</computeroutput> is some error code
2499number.  If <computeroutput>N == 1007</computeroutput>, it also
2500prints some extra text advising the reader that unreliable memory
2501is often associated with internal error 1007. (This is a
2502frequently-observed-phenomenon with versions 1.0.0/1.0.1).</para>
2503
2504<para><computeroutput>exit(3)</computeroutput> is then
2505called.</para>
2506
2507<para>For a <computeroutput>stdio</computeroutput>-free library,
2508assertion failures result in a call to a function declared
2509as:</para>
2510
2511<programlisting>
2512extern void bz_internal_error ( int errcode );
2513</programlisting>
2514
2515<para>The relevant code is passed as a parameter.  You should
2516supply such a function.</para>
2517
2518<para>In either case, once an assertion failure has occurred, any
2519<computeroutput>bz_stream</computeroutput> records involved can
2520be regarded as invalid.  You should not attempt to resume normal
2521operation with them.</para>
2522
2523<para>You may, of course, change critical error handling to suit
2524your needs.  As I said above, critical errors indicate bugs in
2525the library and should not occur.  All "normal" error situations
2526are indicated via error return codes from functions, and can be
2527recovered from.</para>
2528
2529</sect2>
2530
2531</sect1>
2532
2533
2534<sect1 id="win-dll" xreflabel="Making a Windows DLL">
2535<title>Making a Windows DLL</title>
2536
2537<para>Everything related to Windows has been contributed by
2538Yoshioka Tsuneo
2539(<computeroutput>tsuneo@rr.iij4u.or.jp</computeroutput>), so
2540you should send your queries to him (but perhaps Cc: me,
2541<computeroutput>&bz-email;</computeroutput>).</para>
2542
2543<para>My vague understanding of what to do is: using Visual C++
25445.0, open the project file
2545<computeroutput>libbz2.dsp</computeroutput>, and build.  That's
2546all.</para>
2547
2548<para>If you can't open the project file for some reason, make a
2549new one, naming these files:
2550<computeroutput>blocksort.c</computeroutput>,
2551<computeroutput>bzlib.c</computeroutput>,
2552<computeroutput>compress.c</computeroutput>,
2553<computeroutput>crctable.c</computeroutput>,
2554<computeroutput>decompress.c</computeroutput>,
2555<computeroutput>huffman.c</computeroutput>,
2556<computeroutput>randtable.c</computeroutput> and
2557<computeroutput>libbz2.def</computeroutput>.  You will also need
2558to name the header files <computeroutput>bzlib.h</computeroutput>
2559and <computeroutput>bzlib_private.h</computeroutput>.</para>
2560
2561<para>If you don't use VC++, you may need to define the
2562proprocessor symbol
2563<computeroutput>_WIN32</computeroutput>.</para>
2564
2565<para>Finally, <computeroutput>dlltest.c</computeroutput> is a
2566sample program using the DLL.  It has a project file,
2567<computeroutput>dlltest.dsp</computeroutput>.</para>
2568
2569<para>If you just want a makefile for Visual C, have a look at
2570<computeroutput>makefile.msc</computeroutput>.</para>
2571
2572<para>Be aware that if you compile
2573<computeroutput>bzip2</computeroutput> itself on Win32, you must
2574set <computeroutput>BZ_UNIX</computeroutput> to 0 and
2575<computeroutput>BZ_LCCWIN32</computeroutput> to 1, in the file
2576<computeroutput>bzip2.c</computeroutput>, before compiling.
2577Otherwise the resulting binary won't work correctly.</para>
2578
2579<para>I haven't tried any of this stuff myself, but it all looks
2580plausible.</para>
2581
2582</sect1>
2583
2584</chapter>
2585
2586
2587
2588<chapter id="misc" xreflabel="Miscellanea">
2589<title>Miscellanea</title>
2590
2591<para>These are just some random thoughts of mine.  Your mileage
2592may vary.</para>
2593
2594
2595<sect1 id="limits" xreflabel="Limitations of the compressed file format">
2596<title>Limitations of the compressed file format</title>
2597
2598<para><computeroutput>bzip2-1.0.X</computeroutput>,
2599<computeroutput>0.9.5</computeroutput> and
2600<computeroutput>0.9.0</computeroutput> use exactly the same file
2601format as the original version,
2602<computeroutput>bzip2-0.1</computeroutput>.  This decision was
2603made in the interests of stability.  Creating yet another
2604incompatible compressed file format would create further
2605confusion and disruption for users.</para>
2606
2607<para>Nevertheless, this is not a painless decision.  Development
2608work since the release of
2609<computeroutput>bzip2-0.1</computeroutput> in August 1997 has
2610shown complexities in the file format which slow down
2611decompression and, in retrospect, are unnecessary.  These
2612are:</para>
2613
2614<itemizedlist mark='bullet'>
2615
2616 <listitem><para>The run-length encoder, which is the first of the
2617   compression transformations, is entirely irrelevant.  The
2618   original purpose was to protect the sorting algorithm from the
2619   very worst case input: a string of repeated symbols.  But
2620   algorithm steps Q6a and Q6b in the original Burrows-Wheeler
2621   technical report (SRC-124) show how repeats can be handled
2622   without difficulty in block sorting.</para></listitem>
2623
2624 <listitem><para>The randomisation mechanism doesn't really need to be
2625   there.  Udi Manber and Gene Myers published a suffix array
2626   construction algorithm a few years back, which can be employed
2627   to sort any block, no matter how repetitive, in O(N log N)
2628   time.  Subsequent work by Kunihiko Sadakane has produced a
2629   derivative O(N (log N)^2) algorithm which usually outperforms
2630   the Manber-Myers algorithm.</para>
2631
2632   <para>I could have changed to Sadakane's algorithm, but I find
2633   it to be slower than <computeroutput>bzip2</computeroutput>'s
2634   existing algorithm for most inputs, and the randomisation
2635   mechanism protects adequately against bad cases.  I didn't
2636   think it was a good tradeoff to make.  Partly this is due to
2637   the fact that I was not flooded with email complaints about
2638   <computeroutput>bzip2-0.1</computeroutput>'s performance on
2639   repetitive data, so perhaps it isn't a problem for real
2640   inputs.</para>
2641
2642   <para>Probably the best long-term solution, and the one I have
2643   incorporated into 0.9.5 and above, is to use the existing
2644   sorting algorithm initially, and fall back to a O(N (log N)^2)
2645   algorithm if the standard algorithm gets into
2646   difficulties.</para></listitem>
2647
2648  <listitem><para>The compressed file format was never designed to be
2649   handled by a library, and I have had to jump though some hoops
2650   to produce an efficient implementation of decompression.  It's
2651   a bit hairy.  Try passing
2652   <computeroutput>decompress.c</computeroutput> through the C
2653   preprocessor and you'll see what I mean.  Much of this
2654   complexity could have been avoided if the compressed size of
2655   each block of data was recorded in the data stream.</para></listitem>
2656
2657 <listitem><para>An Adler-32 checksum, rather than a CRC32 checksum,
2658   would be faster to compute.</para></listitem>
2659
2660</itemizedlist>
2661
2662<para>It would be fair to say that the
2663<computeroutput>bzip2</computeroutput> format was frozen before I
2664properly and fully understood the performance consequences of
2665doing so.</para>
2666
2667<para>Improvements which I was able to incorporate into 0.9.0,
2668despite using the same file format, are:</para>
2669
2670<itemizedlist mark='bullet'>
2671
2672 <listitem><para>Single array implementation of the inverse BWT.  This
2673  significantly speeds up decompression, presumably because it
2674  reduces the number of cache misses.</para></listitem>
2675
2676 <listitem><para>Faster inverse MTF transform for large MTF values.
2677  The new implementation is based on the notion of sliding blocks
2678  of values.</para></listitem>
2679
2680 <listitem><para><computeroutput>bzip2-0.9.0</computeroutput> now reads
2681  and writes files with <computeroutput>fread</computeroutput>
2682  and <computeroutput>fwrite</computeroutput>; version 0.1 used
2683  <computeroutput>putc</computeroutput> and
2684  <computeroutput>getc</computeroutput>.  Duh!  Well, you live
2685  and learn.</para></listitem>
2686
2687</itemizedlist>
2688
2689<para>Further ahead, it would be nice to be able to do random
2690access into files.  This will require some careful design of
2691compressed file formats.</para>
2692
2693</sect1>
2694
2695
2696<sect1 id="port-issues" xreflabel="Portability issues">
2697<title>Portability issues</title>
2698
2699<para>After some consideration, I have decided not to use GNU
2700<computeroutput>autoconf</computeroutput> to configure 0.9.5 or
27011.0.</para>
2702
2703<para><computeroutput>autoconf</computeroutput>, admirable and
2704wonderful though it is, mainly assists with portability problems
2705between Unix-like platforms.  But
2706<computeroutput>bzip2</computeroutput> doesn't have much in the
2707way of portability problems on Unix; most of the difficulties
2708appear when porting to the Mac, or to Microsoft's operating
2709systems.  <computeroutput>autoconf</computeroutput> doesn't help
2710in those cases, and brings in a whole load of new
2711complexity.</para>
2712
2713<para>Most people should be able to compile the library and
2714program under Unix straight out-of-the-box, so to speak,
2715especially if you have a version of GNU C available.</para>
2716
2717<para>There are a couple of
2718<computeroutput>__inline__</computeroutput> directives in the
2719code.  GNU C (<computeroutput>gcc</computeroutput>) should be
2720able to handle them.  If you're not using GNU C, your C compiler
2721shouldn't see them at all.  If your compiler does, for some
2722reason, see them and doesn't like them, just
2723<computeroutput>#define</computeroutput>
2724<computeroutput>__inline__</computeroutput> to be
2725<computeroutput>/* */</computeroutput>.  One easy way to do this
2726is to compile with the flag
2727<computeroutput>-D__inline__=</computeroutput>, which should be
2728understood by most Unix compilers.</para>
2729
2730<para>If you still have difficulties, try compiling with the
2731macro <computeroutput>BZ_STRICT_ANSI</computeroutput> defined.
2732This should enable you to build the library in a strictly ANSI
2733compliant environment.  Building the program itself like this is
2734dangerous and not supported, since you remove
2735<computeroutput>bzip2</computeroutput>'s checks against
2736compressing directories, symbolic links, devices, and other
2737not-really-a-file entities.  This could cause filesystem
2738corruption!</para>
2739
2740<para>One other thing: if you create a
2741<computeroutput>bzip2</computeroutput> binary for public distribution,
2742please consider linking it statically (<computeroutput>gcc
2743-static</computeroutput>).  This avoids all sorts of library-version
2744issues that others may encounter later on.</para>
2745
2746<para>If you build <computeroutput>bzip2</computeroutput> on
2747Win32, you must set <computeroutput>BZ_UNIX</computeroutput> to 0
2748and <computeroutput>BZ_LCCWIN32</computeroutput> to 1, in the
2749file <computeroutput>bzip2.c</computeroutput>, before compiling.
2750Otherwise the resulting binary won't work correctly.</para>
2751
2752</sect1>
2753
2754
2755<sect1 id="bugs" xreflabel="Reporting bugs">
2756<title>Reporting bugs</title>
2757
2758<para>I tried pretty hard to make sure
2759<computeroutput>bzip2</computeroutput> is bug free, both by
2760design and by testing.  Hopefully you'll never need to read this
2761section for real.</para>
2762
2763<para>Nevertheless, if <computeroutput>bzip2</computeroutput> dies
2764with a segmentation fault, a bus error or an internal assertion
2765failure, it will ask you to email me a bug report.  Experience from
2766years of feedback of bzip2 users indicates that almost all these
2767problems can be traced to either compiler bugs or hardware
2768problems.</para>
2769
2770<itemizedlist mark='bullet'>
2771
2772 <listitem><para>Recompile the program with no optimisation, and
2773  see if it works.  And/or try a different compiler.  I heard all
2774  sorts of stories about various flavours of GNU C (and other
2775  compilers) generating bad code for
2776  <computeroutput>bzip2</computeroutput>, and I've run across two
2777  such examples myself.</para>
2778
2779  <para>2.7.X versions of GNU C are known to generate bad code
2780  from time to time, at high optimisation levels.  If you get
2781  problems, try using the flags
2782  <computeroutput>-O2</computeroutput>
2783  <computeroutput>-fomit-frame-pointer</computeroutput>
2784  <computeroutput>-fno-strength-reduce</computeroutput>.  You
2785  should specifically <emphasis>not</emphasis> use
2786  <computeroutput>-funroll-loops</computeroutput>.</para>
2787
2788  <para>You may notice that the Makefile runs six tests as part
2789  of the build process.  If the program passes all of these, it's
2790  a pretty good (but not 100%) indication that the compiler has
2791  done its job correctly.</para></listitem>
2792
2793 <listitem><para>If <computeroutput>bzip2</computeroutput>
2794  crashes randomly, and the crashes are not repeatable, you may
2795  have a flaky memory subsystem.
2796  <computeroutput>bzip2</computeroutput> really hammers your
2797  memory hierarchy, and if it's a bit marginal, you may get these
2798  problems.  Ditto if your disk or I/O subsystem is slowly
2799  failing.  Yup, this really does happen.</para>
2800
2801  <para>Try using a different machine of the same type, and see
2802  if you can repeat the problem.</para></listitem>
2803
2804  <listitem><para>This isn't really a bug, but ... If
2805  <computeroutput>bzip2</computeroutput> tells you your file is
2806  corrupted on decompression, and you obtained the file via FTP,
2807  there is a possibility that you forgot to tell FTP to do a
2808  binary mode transfer.  That absolutely will cause the file to
2809  be non-decompressible.  You'll have to transfer it
2810  again.</para></listitem>
2811
2812</itemizedlist>
2813
2814<para>If you've incorporated
2815<computeroutput>libbzip2</computeroutput> into your own program
2816and are getting problems, please, please, please, check that the
2817parameters you are passing in calls to the library, are correct,
2818and in accordance with what the documentation says is allowable.
2819I have tried to make the library robust against such problems,
2820but I'm sure I haven't succeeded.</para>
2821
2822<para>Finally, if the above comments don't help, you'll have to
2823send me a bug report.  Now, it's just amazing how many people
2824will send me a bug report saying something like:</para>
2825
2826<programlisting>
2827bzip2 crashed with segmentation fault on my machine
2828</programlisting>
2829
2830<para>and absolutely nothing else.  Needless to say, a such a
2831report is <emphasis>totally, utterly, completely and
2832comprehensively 100% useless; a waste of your time, my time, and
2833net bandwidth</emphasis>.  With no details at all, there's no way
2834I can possibly begin to figure out what the problem is.</para>
2835
2836<para>The rules of the game are: facts, facts, facts.  Don't omit
2837them because "oh, they won't be relevant".  At the bare
2838minimum:</para>
2839
2840<programlisting>
2841Machine type.  Operating system version.
2842Exact version of bzip2 (do bzip2 -V).
2843Exact version of the compiler used.
2844Flags passed to the compiler.
2845</programlisting>
2846
2847<para>However, the most important single thing that will help me
2848is the file that you were trying to compress or decompress at the
2849time the problem happened.  Without that, my ability to do
2850anything more than speculate about the cause, is limited.</para>
2851
2852</sect1>
2853
2854
2855<sect1 id="package" xreflabel="Did you get the right package?">
2856<title>Did you get the right package?</title>
2857
2858<para><computeroutput>bzip2</computeroutput> is a resource hog.
2859It soaks up large amounts of CPU cycles and memory.  Also, it
2860gives very large latencies.  In the worst case, you can feed many
2861megabytes of uncompressed data into the library before getting
2862any compressed output, so this probably rules out applications
2863requiring interactive behaviour.</para>
2864
2865<para>These aren't faults of my implementation, I hope, but more
2866an intrinsic property of the Burrows-Wheeler transform
2867(unfortunately).  Maybe this isn't what you want.</para>
2868
2869<para>If you want a compressor and/or library which is faster,
2870uses less memory but gets pretty good compression, and has
2871minimal latency, consider Jean-loup Gailly's and Mark Adler's
2872work, <computeroutput>zlib-1.2.1</computeroutput> and
2873<computeroutput>gzip-1.2.4</computeroutput>.  Look for them at
2874<ulink url="http://www.zlib.org">http://www.zlib.org</ulink> and
2875<ulink url="http://www.gzip.org">http://www.gzip.org</ulink>
2876respectively.</para>
2877
2878<para>For something faster and lighter still, you might try Markus F
2879X J Oberhumer's <computeroutput>LZO</computeroutput> real-time
2880compression/decompression library, at
2881<ulink url="http://www.oberhumer.com/opensource">http://www.oberhumer.com/opensource</ulink>.</para>
2882
2883</sect1>
2884
2885
2886
2887<sect1 id="reading" xreflabel="Further Reading">
2888<title>Further Reading</title>
2889
2890<para><computeroutput>bzip2</computeroutput> is not research
2891work, in the sense that it doesn't present any new ideas.
2892Rather, it's an engineering exercise based on existing
2893ideas.</para>
2894
2895<para>Four documents describe essentially all the ideas behind
2896<computeroutput>bzip2</computeroutput>:</para>
2897
2898<literallayout>Michael Burrows and D. J. Wheeler:
2899  "A block-sorting lossless data compression algorithm"
2900   10th May 1994.
2901   Digital SRC Research Report 124.
2902   ftp://ftp.digital.com/pub/DEC/SRC/research-reports/SRC-124.ps.gz
2903   If you have trouble finding it, try searching at the
2904   New Zealand Digital Library, http://www.nzdl.org.
2905
2906Daniel S. Hirschberg and Debra A. LeLewer
2907  "Efficient Decoding of Prefix Codes"
2908   Communications of the ACM, April 1990, Vol 33, Number 4.
2909   You might be able to get an electronic copy of this
2910   from the ACM Digital Library.
2911
2912David J. Wheeler
2913   Program bred3.c and accompanying document bred3.ps.
2914   This contains the idea behind the multi-table Huffman coding scheme.
2915   ftp://ftp.cl.cam.ac.uk/users/djw3/
2916
2917Jon L. Bentley and Robert Sedgewick
2918  "Fast Algorithms for Sorting and Searching Strings"
2919   Available from Sedgewick's web page,
2920   www.cs.princeton.edu/~rs
2921</literallayout>
2922
2923<para>The following paper gives valuable additional insights into
2924the algorithm, but is not immediately the basis of any code used
2925in bzip2.</para>
2926
2927<literallayout>Peter Fenwick:
2928   Block Sorting Text Compression
2929   Proceedings of the 19th Australasian Computer Science Conference,
2930     Melbourne, Australia.  Jan 31 - Feb 2, 1996.
2931   ftp://ftp.cs.auckland.ac.nz/pub/peter-f/ACSC96paper.ps</literallayout>
2932
2933<para>Kunihiko Sadakane's sorting algorithm, mentioned above, is
2934available from:</para>
2935
2936<literallayout>http://naomi.is.s.u-tokyo.ac.jp/~sada/papers/Sada98b.ps.gz
2937</literallayout>
2938
2939<para>The Manber-Myers suffix array construction algorithm is
2940described in a paper available from:</para>
2941
2942<literallayout>http://www.cs.arizona.edu/people/gene/PAPERS/suffix.ps
2943</literallayout>
2944
2945<para>Finally, the following papers document some
2946investigations I made into the performance of sorting
2947and decompression algorithms:</para>
2948
2949<literallayout>Julian Seward
2950   On the Performance of BWT Sorting Algorithms
2951   Proceedings of the IEEE Data Compression Conference 2000
2952     Snowbird, Utah.  28-30 March 2000.
2953
2954Julian Seward
2955   Space-time Tradeoffs in the Inverse B-W Transform
2956   Proceedings of the IEEE Data Compression Conference 2001
2957     Snowbird, Utah.  27-29 March 2001.
2958</literallayout>
2959
2960</sect1>
2961
2962</chapter>
2963
2964</book>
2965