NEWS

Bowtie: an Ultrafast, Lightweight Short Read Aligner

Bowtie NEWS
===========

   Bowtie is now available for download.  0.9.0 is the first version to
be released under the OSI Artistic License (see `LICENSE') and freely
available to the public for download.  The current version is 1.3.1.

Reporting Issues
================

Please report any issues on GitHub:

    * https://github.com/BenLangmead/bowtie/issues

Announcements
=============

Keep up with our releases on the GitHub releases page:

    * https://github.com/BenLangmead/bowtie/releases

Version Release History
=======================
Version 1.3.1 - Sep 13, 2021

    * Fixed an overflow issue in `bowtie-build` that would sometimes
      yield corrupt "large" (64-bit) indexes; the resulting index
      would sometimes cause `bowtie` to hang. Note: `bowtie2-build`
      does not have this issue.
    * Fixed an issue in `bowtie` causing XM:i SAM optional field to
      sometimes be off by 1 when using the `-m/-M` flags.
    * Fixed an issue that would sometimes cause deadlocks in `bowtie`
      when running multithreaded.
    * Fixed an issue causing build errors when compiling against
      a pre-C++11 standard.

Version 1.3.0 - Jul 22, 2020

    * Fixed an issue causing `bowtie` to report incorrect results
      when using a Bowtie 2 index.
    * New, more efficient implementation of `--reorder` for keeping
      SAM output lines in same order as input reads
    * Added `-x` parameter for specifying index. `bowtie` still
      supports specifying an index via positional parameter, but this
      behavior will be deprecated.
    * Migrated `python` scripts to `python3`.
    * Fully removed colorspace functionality.
    * Added support for compiling on ARM architectures.
    * Fixed an issue preventing `bowtie` from outputting newlines in
      `--max` and `--un` output files.
    * Fixed an issue causing alignment results to vary based on read
      names.
    * Fixed an issue preventing `--no-unal` from suppressing unmapped
      reads.
    * Removed dependence on some third-party libraries, simplifying
      the code and improving portability.
    * Fix an issue preventing `bowtie` from running with many threads
      on big-endian machines.

Version 1.2.3 - Jul 5, 2019

    * Added support for reading and inspecting Bowtie 2 indexes.
      Bowtie 2 indexes can now be used with either Bowtie or Bowtie 2.
    * Added support for building an index from a gzipped-compressed
      FASTA.
    * Fixed issue preventing bowtie from reporting repeated alignments
      when -M is specified.
    * Fixed issue with -F mode omitting final base of each read.
    * Fixed clipping of first letter of first read in batches after first.
    * Fixed an issue preventing bowtie wrapper script from finding indexes.

Version 1.2.2 - Dec 11, 2017
Update (12/12/2017): We have had to re-release this version of bowtie to address an issue when compiling with pthreads (make NO_TBB=1).

    * Fixed major issue causing corrupt SAM output when using many threads (-p/--threads) on certain systems
    * Fixed major issue with incorrect alignment offsets being reported in --large-index mode
    * Fixed major issue with reads files being skipped when multiple inputs were specified together with -p/--threads
    * The official LICENSE of Bowtie was changed to Artistic License 2.0.  This fixes an issue with the previous LICENSE, which mistakenly combined elements of different open-source licenses.
    * Fixed issue where bowtie would still run for a long time even when -u was set to a small number.
    * Fixed spurious "Reads file contained a pattern with more than 1024 quality values" error for some colorspace inputs.
    * Fixed issue with --strata sometimes failing to suppress alignments at lower strata.
    * Fixed issue with ends of paired-end reads sometimes appearing in non-adjacent lines of the SAM output with -p/--threads >1
    * Fixed issue whereby the read name of end #2 was not always truncated at the first whitespace character
    * Code simplifications

Version 1.2.1.1 - Jun 13, 2017
    * Fixed an issue causing Bowtie to segfault when processing reads from stdin

Version 1.2.1 - Jun 12, 2017
- Please note that Bowtie will be switching to the Artistic 2.0 license in the next release.
- Pre-build binaries now include statically linked TBB and zlib libraries no longer requiring
  these libraries to be pre-installed

    * Fixed an issue which caused Bowtie to hang during parallell index
      building when running an optimized binary
    * Deprecated --refout option.  It will be fully removed in the next
      release
    * Added parallel index building with the bowtie2-build --threads
      option (credit to Aidan Reilly - https://github.com/aidanreilly8)
    * Added native support for gzipped read files.  The wrapper
      script is no longer responsible for this, which simplifies the wrapper
      and improves speed and thread scaling.
    * Added support for interleaved paired-end FASTQ inputs
      (--interleaved)
    * Fixed issue where first character of some read names was
      omitted from SAM output when using tabbed input formats
    * Fixed issue that caused Bowtie to hang when aligning FASTA inputs
      with more than one thread
    * Bowtie wrapper now works even when invoked via a symlink in a
      different directory from the executables
    * Fixed issue preventing reading --tab5 input on stdin
    * Added --no-unal option for suppressing unmapped reads in SAM output

Version 1.2.0 - Dec 12, 2016
    * Code related to read parsing was completely rewritten to improve
      scalability to many threads.  In short, the critical section is
      simpler and parses input reads in batches rather than one at a time.
      The improvement applies to all read formats.
    * --reads-per-batch command line parameter added to specify the number
      of reads to read from input file at once
    * TBB is now the default threading library.  We consistently found TBB to give
      superior thread scaling.  It is widely available and widely installed.
      That said, we are also preserving a "legacy" version of Bowtie that,
      like previous releases, does not use TBB.  To compile Bowtie source
      in legacy mode use `NO_TBB` (TODO: insert actual flag here).  To
      use legacy binaries, download the appropriate binary archive with
      "legacy" in the name.
    * Bowtie now uses a queue-based lock rather
      than a spin or heavyweight lock.  We find this gives superior thread
      scaling; we saw an order-of-magnitude throughput improvements at
      120 threads in one experiment, for example.
    * Unnecessary thread synchronization removed
    * Fixed colorspace parsing when primer base is present
    * Fixed bugs related to --skip command line option

Version 1.1.2 - Jun 23, 2015
   * Fixed the building process for OSX Yosemite.
   * Added the install target for linux to better aid package building process
     and the overall installation process.
   * Added the Intel TBB option which provides in most situations a better performance
     output. TBB is not present by default in the current build but can be added
     by compiling the source code with WITH_TBB=1 option.
   * Fixed a minor issue related with managing the number of threads spawned.
   * Fixed a minor issue which may have caused a memory leak after an exception
     was thrown.
   * Fixed a bug that caused bowtie to crash if a read was trimmed more than the
     read's lenght on 5 prime end.
   * Added some minor corrections/addition to the manual.
   * Fixed a bug that caused the wrapper to incorrectly identify the bowtie binary.


Version 1.1.1 - Oct 1, 2014
   * Fixed a compiling linkage problem related with Mavericks OSX.
   * Improved performance for cases where the reference contains ambiguous
     or masked nucleobases represented by Ns.
   * Some minor automatic tests updates.


Version 1.1.0 - July 19, 2014
   * Added support for large and small indexes, removing 4-billion-nucleotide
     barrier.  Bowtie can now be used with reference genomes of any size.
   * No longer releasing 32-bit binaries.  Simplified manual and Makefile
     accordingly.
   * Phased out CygWin support.
   * Improved efficiency of index files loading.
   * Fixed a bug that made bowtie-inspect fail in some situations.
   * (This release was briefly given version number 2.0.0, but we changed it
     to 1.1.0 to avoid confusion with Bowtie 2.)

Version 1.0.1 - March 14, 2014
   * Improved index querying efficiency using "population count" instructions
     available since SSE4.2.
   * Credits to the Intel(r) enabling team for performance optimizations
     included in this release.  Thank you!

Version 1.0.0 - April 09, 2013
   * Due to general perfromance improvement spinlocking is now used by default.
   * Fixed some racing conditions.
   * CygWin builds are using now pthreah lib.
   * Changed MinGW to optionally use pthread lib on Win32 platforms.
   * Changed the Windows build to use native Windows threads by default.
   * Renamed COPYING to LICENSE to be more GitHub-friendly.
   * Tokenizer no longer has limit of 10,000 tokens, which was a
     problem for users trying to index a very large number of FASTA
     files.
   * Removed references to no-longer-implemented --cutoff option for
     bowtie-build

Version 0.12.9 - December 14, 2012
   * Fixed a bug whereby read names would not be truncated at first
     whitespace character in unmapped or maxed-out SAM records.
   * Fixed errors and warnings when compiling with clang++.
   * Fixed most errors and warnings when compiling with recent versions
     of g++, though you may need to add EXTRA_FLAGS=-Wno-enum-compare
     to avoid all warnings.

Version 0.12.8 - May 6, 2012
   * Fixed a bug that would sometimes cause an immediate segmentation
     fault in --sam mode
   * Fixed make_galGal3.sh script to not omit chromosome 25.
   * Removed -B option from usage message for bowtie-build; that option
     is not implemented
   * Fixed issue that could cause Bowtie not to compile when
     BOWTIE_PTHREADS is left undefined and pthreads.h is not present
   * Elaborated documentation for -B/--offbase option to indicate that
     it only affects offsets in the default output mode, not in SAM
     mode.

Version 0.12.7 - September 7, 2010
   * Fixes the all-gap reference sequence issue that was present in
     Bowtie 0.12.6.  Index files produced by bowtie-build 0.12.5 and
     earlier (back to 0.10.*) are compatible with bowtie 0.12.7.  Index
     files produced by bowtie-build 0.12.7 are backward compatible with
     bowtie 0.12.5 and earlier as long as the first reference sequence
     is not all-gaps (or, when colorspace indexes, as long as the first
     reference sequence has two consecutive ACGT characters in it
     somewhere).
   * Indexes where the first sequence consists of all gaps or other
     non-ACGT characters are not handled properly by Bowtie versions
     0.12.5 and older, but are handled properly by Bowtie 0.12.7.
   * REMOVED: bowtie-build's --old-reverse option; the old reverse-
     index scheme is again the default.  The new scheme is disabled
     pending further refinement.

September 3, 2010
   * The version of bowtie-build distributed in Bowtie 0.12.6 IS
     BROKEN.  It could not handle stretches of ambiguous reference
     characters compatibly with prior versions of Bowtie.  Please use
     Bowtie 0.12.5's bowtie-build instead.  Bowtie 0.12.7, due out
     soon, will contain a complete set of working tools.  Bowtie 0.12.7
     will revert to old way of building the reverse index.  Sorry for
     the inconvenience.

Version 0.12.6 - August 29, 2010
   * Modified bowtie-inspect's default mode to use the bit-encoded
     reference portion of the index to reconstruct the reference
     sequence, rather than the ebwt portion.  This makes bowtie-inspect
     much faster and uses less memory, and the output for a colorspace
     index will now be in nucleotide space.  To get the behavior of the
     old default, use the new -e/--ebwt-ref option.
   * Fixed bug whereby SOLiD QV strings would fail to parse.
   * Moved to a new default way of building the reverse index.  Revert
     to the old behavior with bowtie-build's new --old-reverse option.
     The new reverse index format is forward and backward compatible
     with `bowtie`, unless otherwise noted in a future version.
   * Fixed issue that would sometimes cause bowtie-build to crash when
     building a large index with a low --offrate.
   * Fixed build issue that would cause bowtie-build built on one
     version of Linux to die with a "floating point error" on other
     versions.
   * Fixed a bug whereby alignment cost could sometimes be
     miscalculated.  Stratum was unaffected.
   * bowtie now simply skips reads with 0 characters.  Previously it
     would print an error and exit.

Version 0.12.5 - April 10, 2010
   * Fixed spurious "Error while writing string output; not all
     characters written" errors in -S/--sam mode.

Version 0.12.4 - April 5, 2010
   * Periods in read sequences are now treated as Ns instead of
     ignored.  This should help with some problems where Bowtie
     erroneously reports "Reads file contained a pattern with more than
     1024 quality values..." for data from recent versions of the
     Illumina GA pipeline.
   * Fixed a bug whereby some error and warning messages would be
     printed on top of each other in -p mode.
   * Chunk-exhaustion warnings messages are now suppressed when --quiet
     is specified.
   * Fixed small issue in quality decoding whereby no-confidence colors
     would incorrectly influence decoded quality of adjacent bases.

Version 0.12.3 - February 17, 2010
   * Fixed a significant bug in -C/--color mode whereby quality values
     for SNP nucleotide positions were erroneously penalized.
   * Fixed a bug in -S/--sam mode whereby if whitespace occurred in the
     original read name, it would be printed in the QNAME field in
     violation of the SAM spec. Bowtie now truncates read names at the
     first whitespace character before printing.
   * When input is FASTQ and -C/--color is enabled, Bowtie is now
     tolerant of output from FASTQ converters that include the primer
     base and where the quality string is one character shorter than
     the sequence string (due to the primer). This fixes issues users
     have reported using output from BFAST's solid2fastq.
   * Added support for SOLiD-style _QV files, via the -Q/--quals, --Q1
     and --Q2 options. These options are used in combination with
     -C/--color and -f to align parallel colorspace read/quality files
     without having to convert to FASTQ.

Version 0.12.2 - February 2, 2010
   * When -C/--color is enabled, the default paired-end orientation is
     now --ff, not --fr.  The new default fits typical SOLiD output.
   * Fixed a bug whereby very large reads could cause bowtie to crash
     in --best mode
   * After 0.12.1, some issues remained whereby bowtie would fail to
     trim the primer in colorspace (e.g. in -c mode).  All input modes
     should now have fully-functioning
   * Fixed a bug whereby decoded colorspace qualities could overflow
     and erroneously become low qualities.
   * Fixed a bug that could produce incorrect paired-end alignment
     results in -n mode when using paired-end orientation modes other
     than --fr.  Even with the bug, reported results are reasonable;
     but the seed edit constraint (-n) may have been applied to the
     wrong end of one of the mates.
   * Changed --chunkmbs default up to 64 from 32.
   * Better error checking and reporting for some bowtie options.
   * Some basic testing scripts are now bundled with Bowtie (in
     scripts/test), which should make it easier to regression-test.

Version 0.12.1 - January 8, 2010
   * IMPORTANT: Fixed bug whereby bowtie would fail to remove both the
     primer base and the first color when parsing .csfasta files with
     primer bases in -C -f mode.  A workaround for users of version
     0.12.0 is to use "-5 1" in that situation.
   * Fixed bug whereby, when -M limit was exceeded for an unpaired
     read, the number printed in the 7th column for the random
     alignment was too low by 1.
   * Added documentation discussing a pitfall regarding SOLiD paired-
     end input, i.e., not all entries necessarily have corresponding
     mates in the other file.

Version 0.12.0 - December 23, 2009
   * Added missing README.markdown file
   * Minor documentation additions

Version 0.12.0-beta1 - December 12, 2009
   * Added SOLiD colorspace support
     * Colorspace indexes are distinct from standard letterspace
       indexes and must be built with a separate invocation of
       bowtie-build (with -C option)
     * Running bowtie with -C causes Bowtie to align in colorspace;
       both index and reads must be in colorspace
     * Colorspace memory requirement is the same as paired-end
       alignment in nucleotide-space (normal) mode.  Paired-end
       alignment does not increase the memory requirement further in
       colorspace.
     * csfasta, csfastq, and "raw" read formats are all supported with
       -C; '0' means "blue" and is intechangeable with 'A', likewise
       '1' ('C') means "green", '2' ('G') means "orange" and '3' ('T')
       means "red"
     * Colorspace versions of pre-built indexes added (see Bowtie web
       site)
     * New manual section discussing colorspace features
   * Fixed a few SAM output issues
     * @PG line now properly uses colons instead of equals signs
     * Removed /1, /2 suffixes for paired-end reads in SAM mode
     * Added --sam-RG option that permits the user to insert set values
       for flags that appear on the @RG line
   * Fixed lingering pthreads bugs that would cause Bowtie to hang or
     crash toward the end of execution with -p > 1.
   * Fixed performance-related bug that would cause paired-end
     alignment to be artificially slow in many situations.
   * Fixed issue with random number generation that would result in
     non-random selection of alignments in some situations.
   * bowtie -f now supports fasta files with reads split across
     multiple lines
   * New --suppress option suppresses unwanted columns of output
   * The MANUAL file was converted to markdown format, facilitating
     conversion to various other formats using tools like pandoc
   * DEPRECATED: bowtie: --concise, bowtie-build: --big, --little
   * REMOVED: -z/--phased, -b/--binout, bowtie-maptool,
     bowtie-maqconvert

Version 0.11.3 - October 12, 2009
   * Fixed crashing bug in -S/--sam mode when the number of reference
     sequences in the index is very large.
   * Added --sam-nohead option to suppress output of SAM headers in
     -S/--sam mode.
   * Added --sam-nosq option to suppress output of @SQ SAM headers in
     -S/--sam mode.  These can become a nuisance when the reference
     index contains a very large number of sequences.
   * Fixed a bug in bowtie-build's auto-configure mode that would cause
     it to underestimate the amount of memory required by a set of
     parameters.  This in turn would cause the index to be corrupted.

Version 0.11.2 - October 7, 2009
   * Fixed issue whereby --max option was disabled.

Version 0.11.1 - October 5, 2009
   * SAM output: changed XS:i optional field to be named XA:i to avoid
     a conflict with TopHat's XS:i field.

Version 0.11.0 - October 5, 2009
   * Initial SAM output support with -S/--sam option.  Bowtie sets all
     fields according to the SAM spec (Version 0.1.2-draft, 20090820).
     See the new "SAM Output" section of the manual for details.
   * Added --shmem option: --shmem is similar to --mm in that it allows
     concurrent 'bowtie' processes querying the same index to share a
     single memory image of the index.  Unlike --mm, shared memory
     alocated by --shmem is permanent.
   * The alignment summary printed to stderr at the end of an alignment
     run is now more friendly and includes data about the number and
     proportion of reads that aligned, failed to align, or were
     suppressed via the -m option.
   * When too-short reads are encountered, Bowtie now always prints
     warnings, not errors.  --quiet now suppresses those warnings.
   * By default, when bowtie prints a reference sequence name it now
     stops at the first whitespace.  In 0.10.1, the default was to
     print the entire name, which could cause confusion when parsing
     Bowtie output.  To revert to printing the full name, use the new
     --fullref option.
   * Bowtie now prints the command-line before exiting with an error.
   * Fixed mistake in the manual's "Default output" section: the offset
     in field 4 is 0-based, not 1-based.  To obtain a 1-based offset
     instead, use the -B 1 option.
   * Various minor bug fixes.
   * DEPRECATED: -z/--phased, -b/--binout, bowtie-maptool,
     bowtie-maqconvert.  These features will be removed in a future
     version of Bowtie.  Note that -b/--binout, bowtie-maptool, and
     bowtie-maqconvert are largely superseded by the SAM output format
     (-S/--sam), BAM, and SAMtools (http://samtools.sf.net).  Contact
     the authors if this is a problem.
   * REMOVED: --unfq/--unfa/--maxfq/--maxfa/--alfq/--alfa.  Please use
     --un/--max/--al instead.  Contact the authors if this is a
     problem.

Version 0.10.1 - July 19, 2009
   * Now when -3/-5 are used in combination with -I/-X, the -I/-X
     constraints are interpreted as applying to the original insert,
     not the trimmed insert.
   * Fixed issue whereby -I option was ignored; -I option works now.
   * Fixed a bug whereby some large indexes were incorrectly reported
     as corrupt by bowtie-build.
   * Fixed issue whereby negative quality values were wrongly rejected
     when both --integer-quals and --solexa-quals were specified.
   * The -l/--seedlen parameter can now be adjusted down to 5
     (previously had to be >= 20).
   * Fixed several minor memory leaks and out-of-bounds issues.  The
     Linux version of the bowtie aligner now receives a clean bill of
     health from valgrind's memcheck.
   * Other minor bugfixes.

Version 0.10.0.2 - 6/28/09
    * Second bugfix for Windows version. src and bin-win32 packages
      updated. Linux and Mac users are not affected. Thanks for your
      bug reports and patience.

Version 0.10.0.1 - 6/23/09
    * Fix for crashing bug in Windows version. src and bin-win32
      packages updated. Linux and Mac users are not affected.

Version 0.10.0 - June 12, 2009
   * Major change: All alignment modes are now unstratified by default.
     The --nostrata option has been removed, since it is now the
     default.  A --strata option has been added to override the default
     and force stratified reporting.  Reporting is stratified if and
     only if --strata is specified.  --strata now cannot be specified
     without also specifying --best.  Please note that, because of this
     change, specifying the same arguments to this version of Bowtie
     may yield different reported results.
   * Replaced the --unfa/--unfq options with a single --un option,
     which writes unaligned reads to an output file (or pair of output
     files) but keeps reads in their original form.  This is in
     contrast to the old --unfa/--unfq options, which only supported
     FASTA or FASTQ formats, and which would print a post-trimming and
     post-quality-value translation version of the read.  Likewise, the
     --alfa/--alfq and --maxfa/--maxfq options have been replaced with
     --al and --max options.  The old options are still present, but
     are deprecated and will be removed in a future version.
   * Added --nofw and --norc options, allowing alignment to just one
     reference strand or the other.
   * Added --mm option that causes bowtie to use memory-mapped files
     instead of traditional file I/O to access the reference index.
     This allows multiple bowtie processes running on the same computer
     to share a single in-memory image of a given index.  This is a
     useful feature for parallelizing bowtie in situations where memory
     is limited and where -p is inappropriate or insufficient.  This
     feature is not available in the Windows version of Bowtie.
   * Added a section to the manual ("Reporting Modes") clarifying and
     giving examples of how to use Bowtie's reporting options.
   * The --al and -z/--phased options previously interacted in such a
     way that the --al file could contain multiple entries for the same
     aligned read.  --al and -z/--phased are now incompatible.
   * The --oldpmap option, deprecated in version 0.9.8, has been
     removed.

Version 0.9.9.3 - May 12, 2009
   * Fixed an issue where bowtie --best would sometimes use excessive
     amounts of memory to store path descriptors.  There is now a per-
     thread 32-MB ceiling (configurable with new option --chunkmbs
     <int>) on the memory taken by path descriptors.  If the ceiling is
     exceeded Bowtie will skip the offending read, print a warning
     message identifying the read, and continue.
   * More options are available for defining the quality-value format,
     including new --phred64-quals/--solexa1.3-quals options
     appropriate for the 64-based-Phred output of Illumina's GA
     Pipeline 1.3.  Added option --phred33-quals to to handle the more
     typical 33-based-Phred scale (the default).  The --solexa-quals
     option still handles the 64-based-Solexa scale output by GA
     Pipeline versions prior to 1.3.
   * bowtie-build now checks output files for obvious corruption due,
     for example, to disk exhaustion.
   * Specifying "-" (meaning stdin) as an input to bowtie is now
     supported and documented.
   * Fixed a bug whereby bowtie-maqconvert could fail to notice that it
     had exhausted memory and output a corrupt Maq map file.
   * Fixed a bug whereby bowtie would crash when trying to use an index
     built on a machine with different endianness.
   * Fixed several issues that prevented Bowtie from compiling on
     Solaris.  I confirm that Bowtie builds and runs on Solaris.
   * Added _LARGEFILE_SOURCE _FILE_OFFSET_BITS=64 _GNU_SOURCE to the
     default build options in an attempt to resolve some of the large-
     file issues users are having.
   * Clarified column 7 in the manual.  We received many queries from
     users curious about this number.
   * Moderate speed improvements in --best mode.

Version 0.9.9.2 - April 6, 2009
   * Paired-end alignment is now available in all alignment modes,
     including all -n modes.
   * --best now provides better guarantees.  Reported alignments are
     now guaranteed to be "best" both in terms of stratum (i.e. number
     of mismatches, or mismatches in the seed in the case of -n mode),
     and in terms of the quality values at the mismatched position(s).
     Stratum always trumps quality when determining best alignments.
     Also, --best mode resolves the strand bias issue (see manual for a
     discussion of the issue).
   * Speed improvements for --best mode in most alignment modes.
   * Major speed improvement for the -v 3 alignment mode (except when
     -z is also used)
   * The "Reported X alignments..." message is now printed to stderr
     rather than stdout.  Only alignments are written to stdout.
   * In bowtie-maqconvert, read names longer than Maq's limit (36) are
     now truncated to a suffix of the original name, rather than a
     prefix.  This mimics Maq's behavior and prevents "/1" and "/2"
     suffixes for paired-end reads from being destroyed.
   * Added --alfq/--alfa options to dump aligned reads to FASTQ and/or
     FASTA files.
   * Removed many extraneous source files.

Version 0.9.9.1 - March 10, 2009
   * Added paired-end alignment for -v 2 and -v 3 alignment modes (-n
     modes coming soon).
   * Minor bug fixes and speed improvements for all paired-end modes.
   * Added -s/--skip <int> option to skip over the first <int> reads or
     pairs in the input.
   * --unfq/--unfa/--maxfq/--maxfa modes no longer create empty output
     files.
   * All Bowtie tools now compile under GCC 4.3.3.
   * Fixed bug whereby bowtie -b would sometimes write garbage into the
     reference offset field.
   * Paired-end info is now persisted in the -b format, allowing
     bowtie-maptool output to add "/1" and "/2" suffixes as
     appropriate.

Version 0.9.9 - February 19, 2009
   * Added some preliminary support for paired-end alignment in -v 0
     and -v 1 modes.  -1/-2 options to specify the paired-end files,
     -I/-X to specify min and max insert sizes, and --fr/--rf/--ff
     specify relative orientation of upstream and downstream mates.
     bowtie-build now builds two additional files: NAME.3.ebwt and
     NAME.4.ebwt.  Together, these files store a bitpacked version of
     the reference and they are required for paired-end alignment.  If
     your index does not include these files and you would like to
     perform a paired-end alignment, you will have to rebuild the index
     with bowtie-build version 0.9.9 or later.  Paired-end alignment is
     not compatible with -z mode, and it incurs about a 30% greater
     memory overhead than single-end mode.
   * Pre-built indexes available from Bowtie website have been updated
     to include .3/.4.ebwt index files.  These new pre-built indexes
     are no longer compatible with bowtie versions prior to 0.9.8.
   * New -B/--offbase option allows user to specify how bowtie numbers
     reference positions in its output.  E.g. -B 1 causes bowtie to
     number leftmost char as 1.  -B 0 is the default, but -B 1 will
     likely become the default in the 1.0 release.
   * Fixed a bug that caused trimming options -3 and -5 not to work
     properly in -r (raw input) mode.
   * bowtie-build now prints a friendly error message and exits if an
     input file doesn't exist.
   * Fixed a bug that caused the Win32 version of bowtie to hang just
     before it would normally have exited.
   * Fixed bug that could prevent successful read-in of very large
     (>1GB) .2.ebwt index files.
   * Removed --maxns option since it's mostly redundant with what -v
     and -n already do.
   * Removed --ntoa option.
   * bowtie usage message is now divided into sections for clarity.

Version 0.9.8.1 - January 7, 2009
   * Fixed all known problems with the --unfa/--unfq options:
     * They now work properly with multiple threads.
     * Fixed issue where sequence and quals were sometimes reversed.
     * Fixed other issues causing spurious omission of unaligned reads.
   * Added --maxfa/--maxfq options so that reads that don't align due
     to the -m limit can be dumped separately from reads that don't
     align at all.
   * Alignment output is now guaranteed to be "deterministic" even when
     multiple threads are used.  I.e., given the same input reads (in
     any order) and the same --seed, bowtie will produce the same
     alignments every time it is run, though not necessarily in the
     same order.  This does not hold across different versions of
     Bowtie.
   * Multiple other bug fixes.

Version 0.9.8 - November 25, 2008
   * --unfa/--unfq <filename> options cause bowtie to dump unaligned
     reads to FASTA and/or FASTQ files.
   * bowtie-build now selects its memory-efficiency parameters (--bmax,
     --dcv, --packed) automatically by default; this makes it far
     easier to build an index under memory constraints by eliminating
     tedious trail-and-error.  New -a option disables this, yielding
     old behavior.
   * bowtie-build-packed is no longer a separate binary.  Supplying the
     new -p/--packed argument to bowtie-build is the new equivalent.
   * New tool bowtie-maptool converts between Bowtie's output formats.
   * New tool bowtie-inspect recreates reference strings from Bowtie
     index.
   * Renamed bowtie-convert to bowtie-maqconvert for clarity.
   * New universal Mac binary combines i386 & x86_64 binaries.  PowerPC
     still not supported.
   * Added --nomaqround option to bowtie.
   * Fixed memory leaks in bowtie.
   * Switched to a new scheme for mapping positions in "joined"
     reference string to positions in original strings.  This changes
     the index format.  bowtie-build's --oldpmap parameter reverts to
     the old format.  Versions of bowtie prior to 0.9.8 cannot search
     indexes produced by bowtie-build 0.9.8 unless bowtie-build is run
     with --oldpmap.  bowtie 0.9.8 can search either index format.
     Pre-built indexes are still in the old format, but will switch to
     new format when Bowtie 1.0 is released.

Version 0.9.7.1 - November 11, 2008
   * Fixed an issue that caused a spurious loss of sensitivity between
     Bowtie versions 0.9.6 and 0.9.7 in certain modes.  Many thanks to
     Ali Mortazavi for bringing this to our attention.

Version 0.9.7 - November 8, 2008
   * Added new reporting option -m <int> which suppresses all
     alignments for a particular read if more than <int> reportable
     alignments exist for it.
   * Threads now buffer all alignments for a particular read/phase then
     output all alignments in one critical section.  This guarantees
     that all alignments for a given read/phase appear in one
     consecutive block of the output, even when multiple threads are
     operating in parallel.
   * Separated the quality-conversion and parsing aspects of the old
     --solexa-quals argument into separate arguments: --solexa-quals
     (quality conversion) and --integer-quals (parsing).
   * bowtie-convert now handles the new (post-0.7.0) Maq alignment
     format.  The new format allows Maq tools to handle reads up to
     127 bases, whereas the old format was limited to 63 bases.  Added
     a -o option to opt for the old Maq format.
   * New --refout argument sends alignments to a set of files named
     refXXXXX.map, where XXXXX is the 0-padded index of the reference
     sequence aligned to.  Useful for dealing with large datasets
     aligned to, e.g., the assembled human genome.
   * Improved tutorial to use a simple simulated read set (included)
     to do SNP calls with Maq.
   * Added --nota option to bowtie-build
   * Fixed make_h_sapiens_asm.sh script to include mitochondrial DNA.

Version 0.9.6 - October 10, 2008
   * 'bowtie' now supports a host of options that allow the user to
     specify which and how many valid alignments to report per read.
     The default is still to report 1 "good" alignment, which is by far
     the fastest mode.  See -k/-a/--best/--nostrata options described
     in the manual for details.
   * 'bowtie' now supports reads up to 1024 bases long.  Note that for
     reads much longer than, say, 35 bases, the user must be careful to
     set alignment policy parameters (especially -e) appropriately.
   * --fast flag eliminated, double-index mode is now the default.
     Added the -z/--phased flag to revert to phased, half-index mode.
   * --concise output mode now officially supported.  Now outputs one
     alignment per line.
   * Changed 'bowtie-build' default back to --bmaxdivn 4.
   * -h/--help now prints much more verbose help for 'bowtie' and
     'bowtie-build' (verbatim from MANUAL file)
   * BWT-searching code streamlined; much old code eliminated

Version 0.9.5 - September 27, 2008
   * Last column of output now additionally reports the reference and
     query bases (in that order) for mismatches.  E.g., old: "30,32",
     new: "30:C>A,32:C>T".
   * Eliminated spurious trailing space in first column of output.
   * Minor performance and sensitivity improvements.
   * New option '-p' spawns a user-specified number of pthreads for
     parallel processing of reads.  For example, use '-p 4' to run
     'bowtie' on 4 processor cores simultaneously.
   * Due to the new '-p' option, 'bowtie' needs pthreads to compile and
     run.  To compile 'bowtie' without pthreads support (which disables
     the '-p' option), use 'make BOWTIE_PTHREADS=0'.
   * Also due to '-p' option, the Windows version of Bowtie now comes
     with the pthreadGC2.dll file from the pthreads for Win32 project
     (http://sourceware.org/pthreads-win32).  This library is released
     under the LGPL license.
   * New option '--fast' causes Bowtie to load both the "forward" and
     "mirror" halves of the index at once, which eliminates the need
     for multiple phases and speeds up matching at the cost of using
     about twice as much memory.  '--fast' also causes 'bowtie' to
     scale better when used in combination with '-p'.
   * Fixed crashing bug with -o/--offrate in 'bowtie'.
   * Improved error reporting.

Version 0.9.4 - September 16, 2008
   * New method for handling gaps and ambiguity codes in the reference.
     New 'bowtie-build' method handles long stretches of gaps
     gracefully.  New 'bowtie' rejects alignments that overlap a gap or
     ambiguous character in the reference.
   * Due to above change, index file format has been changed.  All
     pre-built indexes available on this site have been updated to the
     new format.  To obtain indexes with the old format, contact us.
   * In 'bowtie' unnamed reads are now given ordinal names (rather than
     "default") in the alignment output.  Works for all input modes.
   * New 'bowtie' input mode: Raw, activated with -r.  Expects one read
     sequence per line; no quality values or names.
   * Fixed 'bowtie' bug whereby trimming did not work in -c mode.
   * Changed 'bowtie-build' default to not use blockwise mode.
   * Changed 'bowtie-build' to avoid certain infinite-loop and very-
     long-runtime scenarios.
   * Packaging improvements: archives now explode into subdirectories
     and scripts are executable.

Version 0.9.3 - September 6, 2008
   * Major reference-name bug fixes to bowtie-convert

Version 0.9.2 - September 4, 2008
   * Now allows 3-mismatches: -n and -v options accept 3
   * Output format prints reference name instead of id in third column
   * Pre-built indexes updated to encode reference names
   * Ns in reads now match nothing (previously, they matched As/Ts)
   * Dropped -l/--linerate and -i/--linesperside arguments to bowtie-
     build
   * Fixed bug in Maq-like mode that allowed some poor alignments
   * Minor speed improvements

Version 0.9.1 - August 25, 2008
   * Integrated relevant SeqAn-1.1 sources into Bowtie source release
   * Now builds on Windows under MinGW (needs pthreads and zlib)
   * Binary releases for Linux (i386, x86_64), Windows (i386) and MacOS
     X (i386)

Version 0.9.0 - August 18, 2008
   * First stable release of Bowtie.
   * Includes the three core Bowtie tools: the indexer 'bowtie-build',
     the read aligner 'bowtie' and the converter from Bowtie's to Maq's
     mapping output format, 'bowtie-convert'.
   * Compatible pre-built indexes for many model organisms are
     available from http://bowtie-bio.sf.net.
   * FASTA, FASTQ inputs supported; tested with Solexa FASTQ
   * Supports Maq alignment policy (-n and -e behave as in Maq)
   * Supports X-mismatch policy (-v option behaves as in SOAP)
   * -n and -v options accept 0, 1, or 2