Blob Blame History Raw

- libout123: Fix error messages beginning from OUT123_ARG_ERROR (bug 261).

- mpg123: Fix --icy-interval handling to work with stream from stdin.
  (curl | mpg123 --icy-interval=n -)
- libmpg123: Fix another invalid read and segfault on damaged (fuzzed)
  files with part2_3_length == 0 (set maxband=1, pulled from upcoming 1.26.0).

- mpg123:
-- Also disable cursor/video games for empty TERM (not just unset and dumb).
- libmpg123:
-- Accept changing mode extension bits when looking for next header for
   detecting free-format streams (bug 257).
-- Fix compute_bpf() for free format streams (needed to estimate track
   length and working fuzzy seeking in absence of an Info tag).

- mpg123:
-- Do not play with cursor and inverse video for progress bar
   when TERM=dumb.
-- Fix parsing of host port for numerical IPv6 addresses (just did
   not work before, only for textual host names).
- libmpg123:
-- Proper fix for the xrpnt overflow problems by correctly
   initialising certain tables for MPEG 2.x layer III. The checks that
   catch the resulting overflow are still in place, but likely superfluous
   now. Note that this means certain valid files would have been misdecoded
   before, if anyone actually produced them. Thanks to Robert Hegemann for
   the fix!
-- Silently handle granules with part2_3_length == 0, but
   scalefac_compress != 0 (ignore the latter).

- Hotfix for bug 255: Overflow reading frame data bits in layer II decoding.
  Now, all-zero data is returned if the frame data is exhausted. This might
  have a slight impact on performance, but not easily measurable so far.

- Avoid another buffer read overflow in the ID3 parser on 32 bit platforms
  (bug 254).

- Better configure checks for i?86-apple-darwin (bug 253).
- libmpg123:
-- Prevent harmless call to memcpy(NULL, NULL, 0).
-- More early checking of ID3v2 encoding values to avoid bogus text being

- libmpg123:
-- Better checks for xrpnt overflow in III_dequantize_sample() before each
   use, avoiding false positives and catching cases that were rendered
   harmless by alignment-enlarged buffers.

- libmpg123:
-- Extend pow tables for layer III to properly handle files with i-stereo and
   5-bit scalefactors. Never observed them for real, just as fuzzed input to
   trigger the read overflow. Note: This one goes on record as CVE-2017-11126,
   calling remote denial of service. While the accesses are out of bounds for
   the pow tables, they still are safely within libmpg123's memory (other
   static tables). Just wrong values are used for computation, no actual crash
   unless you use something like GCC's AddressSanitizer, nor any information
-- Avoid left-shifts of negative integers in layer I decoding.

1.25.1: Hot Fuzz
- libmpg123:
-- Avoid memset(NULL, 0, 0) to calm down the paranoid.
-- Fix bug 252, invalid read of size 1 in ID3v2 parser due to forgotten
   offset from the frame flag bytes (unnoticed in practice for a long
   time). Fuzzers are in the house again. This one got CVE-2017-10683.
-- Avoid a mostly harmless conditional jump depending on uninitialised
   fr->lay in compute_bpf() (mpg123_position()) when track is not ready yet.
-- Fix undefined shifts on signed long mask in layer3.c (worked in practice,
   never right in theory). Code might be a bit faster now, even.
   Thanks to Agostino Sarubbo for reporting.

1.25.0: MP3 now patent-free worldwide!
- Silence test for artsc-config if it is not there.
- Make sure -static-libgcc from LDFLAGS gets through libtool,
  fixing 32 bit Windows builds (depend on libgcc DLL otherwise).
- Fix build with non-GNU make by using plain rm -f instead of silly $(RM)
  in libout123/modules makefile fragment.
- Make build work on iOS, including coreaudio backend.
- libmpg123:
-- Finally provide position-independent code for x86 with assembly
   optimisations.The textrels are gone thanks to Won Kyu Park and Taihei Momma.
-- Clarify some license language in files descending from the original MMX
-- Fix return value overflow check for MPG123_BUFFERFILL.
-- Introduced mpg123_getformat2() to enable the FORMAT command
   for the generic control not stealing MPG123_NEW_FORMAT from the main
   playback loop. The sequence LOADPAUSED-FORMAT-PAUSE (play) is supposed
   to work now.
-- Enable aarch64 optimisations on *BSD by default, too. You can always
   override that stupid OS whitelist using --with-optimization, anyway.
-- Use of the i486 decoder is now discouraged more prominently, in configure
- out123: Fix stupid crash with verbose mode and tone generation (print
  the string if the pointer is non-null, not if it is null).
- libout123: More consistent error messages for dynamic and legacy
  (built-in) modules. Namely, you get a hint how if you choose a different
  module than the built-in ones for a static libout123.

- Avoid repeating genre in metadata printout for specifications like
  (144)Thrash Metal.
- In remote control mode, only enforce --quiet if no verbosity was required.
- Prevent --loop and --shuffle or --random from messing with the remote
  control LOADLIST command (printout of the list would loop without reason).
- Fix the mpg123 command (esp. our provided binaries on Windows) to now find
  modules again relative to the executable directory, not the current working
  directory. This was a regression in 1.23 and might be security-relevant if
  you called mpg123 in working directories with untrusted content.
  Note that mpg123 1.23 looked for modules relative to the current working
  directory only if the installation prefix for modules did not exist.
  So, usage on an intact installation (with /usr/lib/mpg123 or the like) was
  safe. Nevertheless this new version fixes the search to be relative to the
  binary path as it was with 1.22 and before.
- At least consistent behaviour of playlist code in the face of looping.
  Looping is about individual tracks, always. They are looped also in random
  mode. Jumping (prev/next keys) is between tracks and resets the loop counter.
  The display of currently playing track in the playlist is fixed for random
  and looped play now (bug 198).
- Looping is now mentioned for a to-be-repeated track with --verbose.
- Move some compiler nagging from --enable-debug to --enable-nagging, fix up
  some new build failures by adding some pesky feature test macros.
- Try not to pollute the terminal buffer with old progress bars in inverse
  video. Only the currently live one shall be seen. That one is pretty. The
  others are not.
- Using plain dlopen()/LoadLibrary() for opening modules instead of libltdl.
  This also means that --with-module-suffix is gone in configure.
- Windows builds only work when Unicode support is there (older than Windows
  2000/XP will definitely not work anymore).
- The out123 tool now features tone generation, with a mix of differing
  wave patterns. Makes sense to be able to test the audio output by itself,
  and it's fun. See --wave-freq and related parameters.
- libmpg123 version 43:
-- Add flags MPG123_NO_PEEK_END and MPG123_FORCE_SEEKABLE, as suggested
   by Bent Bisballe Nyeng.
-- Build fix for MSVC (consistent definition of ssize_t, spotted by manx,
   bug 243).
-- Build fix for --with-cpu=ppc_nofpu (thanks to Michael Kostylev, bug 244).
-- Add asm optimized MSVC++ Win32|x64 and UWP|x64 builds
-- Remove old, broken MSVC++ builds
- libout123 version 2:
-- Added OUT123_BINDIR.
-- New search order for output plugin directory: MPG123_MODDIR, or (relative
   to executable directory OUT123_BINDIR) ../lib/mpg123, plugins
   libout123/modules/.libs, libout123/modules, ../libout123/modules/.libs,
   ../libout123/modules, and at last the installation prefix $libdir/mpg213/.
   This shall ensure that a build inside a source tree does not try to use old
   modules from the system prefix. The normal libtool wrapper deals with the
   shared libout123 or libmpg123 only, not modules.
   Note that if you set MPG123_MODDIR to a non-existing directory, no modules
   will be found (earlier versions fell back to other choices).
-- The OUT123_NAME parameter is now copied by out123_param_from(), as is
   the newly added OUT123_BINDIR.
-- Coreaudio: Use AudioComponents API on OSX >= 10.6 (thanks to Michael Weiser).
-- Coreaudio: Fix behaviour of out123_drop(), not killing the output anymore
   without re-opening the device (bug 236, thanks to Taihei for the fix).

- Fix long-standing bad memory read (via integer underflow) in ID3 parser
  for crafted ID3v2 tags with tiny size information (bug 240). Thanks to
  Han Lee & Jerold Hoong for the detailed report.

- Fix the bad RPM spec file (broken since out123 addition, thanks
  to Srikanth Rao for noticing).
- More out123_pause() in generic control mode, working around
  bug 236. The coreaudio output needs to be more robust, though.
- Do not undef MPG123_EXPORT in the headers, rather be prepared for it
  being defined already. This fixes MSVC builds, broken since 1.23.0
  (bug 239).

- call out123_pause() instead of out123_stop() in remote control STOP
  command, fixing the regression in 1.23.x of not being able to resume
  playback with another LOAD unless the new track has differing audio
  format (bug 234)

- libout123:
-- fix build of portaudio and SDL on win32 (too many parentheses,
   thanks to Sandro Cumerlato for pointing it out)
-- make out123_errcode() return OUT123_BAD_HANDLE as documented
   (adding that value to the error enum, was documented but missing),
   again thanks to Sandro
- fix (shell quoting in copying stage) and add some
  bits to shipped header to make it directly usable in MSVC (define ssize_t,
  bug 232)
- no overwriting of LIBS in Makefile (build system regression in 1.23,
  thanks to Peter Korsgaard for reporting that one)
- no printing of message in response to SIGPIPE, as that could trigger
  a hang if stderr itself is the piped culprit (regression from 1.22, bug 233)

- mpg123: Fix logic of prebuffering small bits for glitch-free start of
  playback. The logic as introduced in 1.23.0 actually introduced stuttering!
  The fix for this is really minimal and I urge everyone to at least include
  that one. It is the changeset of rev. 4041.
- libout123:
-- Actually make OUT123_KEEP_PLAYING on by default, as documented.
-- Fix buffer logic with regard to draining, use common device writing
   code to handle resuming from pause. This fixes sdl output with buffer,
   possibly other sfifo-based outputs like CoreAudio (those were poorly
   tested with buffer, as the combination is indeed unusual, see bug 230).
-- Buffer pauses devices for prebuffering, to avoid underruns.
-- More reliable playback with SDL, Portaudio and CoreAudio outputs.
   This includes FIFO draining on close to avoid dropping of the end.
   It should fix parts of bug 230.
-- Fixed memory leaks with some outputs where a call to the deinit()
   function is necessary during available/working module checking.
- mpg123, out123: Use the default output flags to get OUT123_KEEP_PLAYING.
  Without this, there is a good chance to abort on SIGSTOP/CONT while playing.
  This is a regression from 1.22 .
- Build fixes (bugs 227 and 228) and some build system hackery.
  Gone are libc feature test macros in compat.h . This means always
  using our own strdup() implementation. Simpler than being bugged by
  feature test macros in strict compiler mode (when _DEFAULT_SOURCE is not
- ./configure --enable-nagging does something also without --enable-debug,
  explictly triggers -std=c89, too (hence the above).
- Some fixup to make things work in pedantic C89 mode again.
  It really might be time to swich on some serious C99 for the next release,
  since there's a followup standard to _that_ which we can ignore now.
- Update list of renamed internal symbols to reduce conflicts during static
  linking of libmpg123 or libout123.
- Internal compatibility functions now handled via libtool convenience
  archives, avoiding the hackish code copies.

- mpg123: Man page fixes by ESR.
- Actually check for sys/select.h (again?) in configure so that it is used.
  Fixes build on a Cygwin install, at least.
- libmpg123: Always default to enabled IEEE 754 hack for rounding on all
  platforms, use ./configure --disable-ieeefloat if you do not want it.
  As a protection for the exotic machine that does not use IEEE floating
  point storage format, mpg123_init() will error out early if the rounding
  does not work out.

- mpg123: Fix compact ID3 tag printing to take UTF-8 character count
  into account again (thoughtless regression in 1.23.0).

- libout123:
-- Fix logic to prevent cases of output devices being closed more than once
   (stop after pausing triggered multiple close calls, which broke sndio, for
-- Fix JACK output handling of port wishlist, was simply broken for more than
   two ports (-o jack -a port1,port2,port3). Now, it is ensured that each
   output ports gets some connection to an out123 channel, possibly splitting
   one to multiple ports or mixing multiple channels into one port (cyclic
   connection scheme). Note that this does only apply for an explicitly given
   list of ports. You can still use --stereo to ensure that two output ports
   (if found) are used also for mono tracks.
-- Also use jack_free() in JACK output for ports list.
-- And finally, ensure that pthread library is linked to JACK module (would need
   hacking to work on non-POSIX, replacing sem_t with equivalent).

- mpg123:
-- Hide networking-related command-line parameters when disabled.

- libmpg123 version 42
-- This adds mpg123_framelength() and makes mpg123_position() truly obsolete.
-- Equalizer optional now (./configure --disable-equalizer) to save precious
-- Clip decode tables for large amplification with fixed-point decoders.
   Without that, high-pitched distortion enters really quickly when
   trying to increase volume even if output samples would not be clipped,
   weren't it for fixed-point math.
- libout123 version 1: First incarnation of libout123, a basic library to
  get audio data written to audio devices (or files) in simple blocking
  manner as opposed to callback APIs (which it wraps over). This collects
  the output modules of mpg123 and makes them available to the wider
  masses. Also, the separate shared memory buffer process (contained
  in libout123) got a lot of reworking which should be noticable in the
  interaction with terminal control.
-- Some tuning of several output drivers (modules) regarding flexible
   buffer size, client names, supported formats.
-- Notably the JACK output got a major overhaul and now does not swallow
   audio at the end anymore. Huge latency also reduced and tunable. Encoding
   fixed to floating point, so that libmpg123 actually chooses what is
- Added mpg123 --no-infoframe.
- Detect terminal on input side and enable control keys automatically.
  There is --no-control now to disable terminal control anyway.
- Display stuff:
-- Avoid spillover of status line that caused unwanted terminal scrolling.
   If the terminal width is smaller than the full status string, a cut is
   made to make it fit.
-- Draw a progress bar on the terminal by inversing colors behind status
-- Rework the format of the status line to be more compact and consistent.
-- Rework smart tag printout for adapting to terminal width.
-- Warning messages also start with a line break now to better fit in with
   verbose playback.
-- Doubly-verbose mode now needed for long MPEG header print out.
   It is a higher level of verbosity than printing the current position.
   The compact header is a better fit.
-- Compact header printout also compacted, example:
   MPEG 1.0 L III cbr128 44100 j-s
-- Reporting of clipped samples also includes a line break now.
-- Keep ID3v2 data around for terminal control to be able to print on
   demand. That was broken for some time.
- Default for --preload now is 0.2 instead of 1 (fill whole buffer before
  playback). The maximum is 0.5 . This is mandated by corrected buffer
  interaction in libout123.
- Improved interaction with buffer process in terminal control mode.
  Seeking is more transparent now, taking dropped samples from buffer into
  account to avoid unintended jumps. Direct seeks with number row do not
  pause playback anymore.
- Silently skip APE tags (thanks to Hans de Goede).
- Some reduction in bitrot on AIX (typos in output module, build with
  --disable-largefile --with-audio=aix, real test welcome).
- Added '[' and ']' keys to skip to previous/next directory in playlist.
- Added support for Haiku to configure script (thanks to Jerome Duval).
- Remove spurious Frankenstein warning when scanning files that have no
  gapless info (the reported value of -1152 gapless sample length should be
  a giveaway anyway).
- Fix a bug in the command line parser that was present from the beginning,
  apparently, causing certain parameters (--au, --cdr) to kill off parsing
  of following ones. Nobody complained in all these years. Astounding.
- Forget content type from HTTP redirects (relate to error pages, not the
  stream redirected to), fixes web radio streams that do not send any
  content type for the actual stream.
- Accept application/pls+xml for PLS playlists. It happens. Don't ask
  mpg123 to actually parse any XML!
- Ignore whitespace after HTTP MIME types (esp. before a ";").
- Added --name to mpg123 and out123 for providing client name p.ex. for JACK.
  There is no default of appending the PID to "mpg123" as client name anymore,
  just plain "mpg123" or "out123".
- Added --devbuffer to mpg123 and out123 for influencing device buffer and
  possible latency a bit (perhaps more so in future when drivers are adapted
  to pay more attention to that).
- Added out123 --channels <n>, --list-encodings, --test-format,
  --test-encodings, --query-format.
- Calling ./configure --enable-nagging does not include --enable-debug anymore,
  you specify that separately.
- Fix for newer glibc that wants higher POSIX feature test for getaddrinfo()

- Fix stupid regression introduced with a bugfix in 1.14.1 to make
  free format streams work again. It took 3 years for someone to notice.
  I added a regression test for that now!

- Fix UTF-16 surrogate detection because that used to falsely detect
  characters above 0xDFFF as surrogates (bug 216).

- Fix buffering for funny sample formats (namely, 24 bit), that do not
  fit nicely into 32768 bytes. Effect was a nasty endless loop where
  mpg123 needs to be externally killed.

- Fix mpg123-id3dump when writing images with funny (manipulated) MIME type.
  Stupid mistake in length computation of the fallback file extension caused
  junk from memory being appended to the filename if the pointer size
  is less than 64 bit. For 64 bit pointers (or longer) it was correct by
- Fix pedantic build by cleaning up out123 source, also now really showing
  the encoding list in --longhelp instead of possibly, again, writing junk
  from memory in there.
- Not linking libmpg123 against libltdl anymore (bug 215).
- Update MSVC++ ports a bit to make them work again.

- libmpg123 version 41
- Build fix: Avoid name clash on OSX SDK with "normal" being defined in
  an enum already.
- More resilience against folks who do not check errors and rather pass
  NULL pointers around (see NEWS.libmpg123).
  This changes some return values to MPG123_BAD_HANDLE instead of MPG123_ERR
  (always check for != MPG123_OK).
- Fix WAV writing to stdout on Windows (bug 214).
- Rework Xing/Info/Lame tag parsing, also detecting short ones without
  the TOC now.
  In addition, the warning about file size being off is a _lot_ more
  accurate now.

- Use LL as shortcut for LOADLIST remote command, L is taken already (bug 210).
- Less namespace pollution (MPG123_EXPORT in header instead of EXPORT, bug 212).

- Make feed reader more efficient when skipping junk by early discarding
  of data from buffers to avoid re-parsing.

- Added NEON optimized decoder for AArch64 (ARM 64bit environment)
-- enabled for --with-cpu=neon64 (NEON only) or --with-cpu=aarch64 (runtime
   switch between neon and generic_fpu, like arm_fpu) configuration
-- compatible with aarch64-linux-gnu toolchains (from Ubuntu, debian)
   and Xcode 5
- Added new binary out123 only with the audio output part of mpg123.
  This is a precursor to separating out said code into a simple audio
  output library. It also allows plugging in some processing in a pipe
  before feeding to output.
- Prevent opendir(NULL) for unknown module directory when listing modules.
- some build fixes, among those fixing mpg123-id3dump on Windows

- libmpg123 version 40
- Support float and 32/24 bit integer output also with fixed-point decoders
  (conversion from 16 bit precision).
- A-Law output encoding is actually implemented now (Michael introduced
  the encoding back in the old days without ever implementing theconversion
  table, apparently).
- Added --with-cpu=arm_fpu as decoder choice for runtime switch between
  plain fpu code and NEON.
- Fix ARM builds to work properly on debian systems without special CFLAGS
  (NEON decoder, specifically).
- Make scripts/ check the return value instead of delivering
  crazy timings for crashed mpg123.
- Trigger track initialization in mpg123_info() to avoid use of uninitialized
  variables (watch out for MPG123_NEED_MORE return value!).
- Fix integer code to 32 bit numbers where it would waste 64 bit on
  modern systems.
- Add non-executable stack marking also to ARM assembly.

- Do not close buffer twice with mpg123 -R (delete outdated code in
  control_generic(), fixing bug 207.
- Do not mess around with first command line argument to extract program
  name, use a copy instead, fixing bug 205.
- Prevent premature exit on timing of failure to open a track in the
  playlist and the user wanting to skip over it (bug 206).
- Fixup of HTTP socket mess, also avoiding unnecessary FILE objects (bug
  204, patch by Rajeev V. Pillai).
- Fix playlist behaviour on slow HTTP streams, jumping back ('d' key)
  from first track stays at first track (thanks to Rajeev again, sub-bug of
  bug 206).
- Fix Frankenstein check not to stumble over ID3v1 tags. My apologies. As
  a bonus, the position of the concerned code in the parser is now far
  more logical.

- libmpg123 version 39
- Big version increase for clear distinction regarding the 1.14.x regression
  fixes in libmpg123.
- Regression fix: Ensure decoder reinitialization on combination of seek
  and resync (buffer overflow bug 201, introduced in 1.14.1).
- Regression fix: Make parsing really end on reaching of junk limit
  (introduced in 1.14.4).
- Add check for bad bit allocation value in layer I decoder (removes
  audible glitch in bug 201 test file).
- Avoid wrong error message about Frankenstein streams when gapless
  decoding disabled.
- Be verbose about big header changes (Frankenstein streams). Note that
  changes in samples per frame will confuse the seeking code, but decoding
  can get through.
- Actually use the full routine to close tracks. Avoids socket leak when
  playing multiple HTTP streams from a playlist (bug 203).
- Send correct host header when connecting through HTTP proxy (bug 202,
  figured out by Rajeev V. Pillai).
- The console application now informs about changed major stream properties
  (prints out another header line).

- mpg123-id3dump --no-scan; for speedup
- hide -C parameter from mpg123 --help when it is not supported
- added tinyalsa Output (by Jarno Lehtinen)
- build system tweaking
- some care for fixing Cygwin builds

- Taihei strikes again:
-- New Huffman decoding scheme (configurable, on by default) brings nice
   boost to mpg123 to make it the fastest decoder on modern CPUs again.
-- new AVX decoder
-- assembly-optimized DCT36 for SSE and AVX
- added configs for vintage 3DNow(ext) CPUs (AMD K6) utilizing assembly DCT36
- added configs for vintage SSE CPUs (Pentium M), utilizing C DCT36
- added FORMAT command to generic control
- added support for ID3v2 APIC frames
- added mpg123-id3dump as official companion to extract meta data
  (including writing of album art from APIC to files)
- added mpg123-strip as official companion to strip dirt out of MPEG streams
  (including meta data)
- not deleting ID3 data right after printing in generic control
- module loader a bit more flexible with relative MPG123_MODDIR
- Fix chopped-off playback with pulseaudio output: draining the output
  now on closing.
- SDL output fixup, a) not chopping off end and b) non-fatal underrun
- Fix obvious errors in win32 output ... does it now play the end of
  tracks? Need someone to test this.
- added support for screen and iris-ansi terminals with --title (as prompted
  by patch from user canavan)
- fixes/enhancements for sgi audio output
- libmpg123: proper largefile aliases for proper systems (FreeBSD with
  one one-and-only off_t
  This enables client software that insists on defining pointless
- wrapper script src/mpg123-with-modules to ease the pain of running mpg123
  with modules without installation (setting MPG123_MODDIR)
- removed ALSA 0.5 code (unused for longtime, last GPL-only bit)

- Writing CDDA data via --cdr works again; had also been broken by attempts
  to catch failing writes because of full disk.
- Also, the return values of WAV/CDR writing routines now properly indicate
  error to avoid stupid endless loop of "smart" audio writing code that wants
  to handle interruptions. Really, --cdr could just go and nobody would miss
  it ...
- Fix stupid bug in mpg123_chomp_string() (introduced in 1.15.1), empty
  lines were badly treated. Only usage in mpg123 is for parsing HTTP redirect
  headers; malicious HTTP servers might exploit that (possibility to zero
  bytes in memory).
- some cleanup for nagging mode
- more resync strictness to really catch (all) cases with changing decoder
  structure; in practice badly needed when disabling seekbuffer (so it's
  your fault, somehow ...) for non-seekable streams, otherwise not really
  an issue because of readahead checking
  I repeat: This fixes possibly fatal behaviour in the parser when readahead
  checking is explicitly disabled (p.ex. via mpg123 --no-seekbuffer on
  http streams). By default, readahead checking is _always_ active!
- build system fixed for automake 1.13 (thanks to Nix)

- Fix WAV writing. AGAIN. People love to decode WAV to stdout in various
  ways that started to get broken with me starting to care for people who
  like to write to full disks. I frikkin' HATE this rat race! I'm even
  starting to SHOUT. Now, the code prefers to write no header at all (when
  there is no actual output) instead of possibly writing too many misleading
  ones. Getting the blame for breaking dir2ogg repeatedly while trying to
  cope with one fatal situation that you cannot really cope with sucks. Big
  time. I hope that's it now!

- build fix with older shells (== in configure)

- libmpg123 API version 37
- Fix corner case in HTTP communication with missing path in Location header
  (bug 187).
- Prevent nasty alignment issues by not guessing about assembler alignment
  when .balign is available (bug 188).

- Added --continue (also see scripts/conplay).
- Added number row to terminal control keys for 10% jump points (for
  navigating long tracks).
- Set xterm title always if asked, disregarding the -q switch (if you want
  fully quiet, don't specify --title).
- Enable terminal mode in OS/2 by explicitly checking for and using
  os2term library.
- Getting WAV/AU/CDR writing back to a sane state while still catching
  out-of-disk when writing headers (none in case of CDR) but removing
  that initial test that prevented writing on any non-seekable file (like,
  /dev/stdout to a pipe).
- Added native audio output for QNX (thanks to Mike Gorchak), also improving
  portability of configure script while at that.
- more robust recovery from ALSA issues (device suspend)
- Fix crash in module loader on *BSD.
- Extend time display to switch to hours above 60 minutes (hh:mm:ss,
  ditching the sub-second precision)
- really include dithering in default x86-64 build.
- Fix mpg123_scan() not properly seeking back for all types of files (bug
  3582628 ... it got a different number on now ... brilliant, folks!).
- Prevent false positives for "Frankenstein!" when using mpg123_scan().
- Fix printing of negative positions with buffer (bug 186).
- Accept application/x-scpls as additional MIME type.
- Try to handle certain Windows toolchains that miss EOVERFLOW.

Who reworks a parsing and data mangling without really making sure that
things still work like before? I hope I won't do so in future. Not again.

- Fix resync logic to properly work again (used to prematurely end stream
  on bad headers). A regression again, this sucks, but at least we got a
  regression test for that now.
- Fix aeon-old regression (since 1.11.0) about short seeks: Bit reservoir
  was reset also for short seeks that did not break continuity.
- Another damned regression got fixed: Sample-accurate seek for files
  without full gapless info.

- Fix regression from 1.14.1 in parsing of bad free format streams,
  avoiding apparently endless loops (just very slow reading, in fact).

- Fix writing of WAV to stdout.
  It is still not the best of ideas to do so, though. At least make sure
  to fix rate and channels.

- Fix subtle bug that could trigger failure to detect a format change on
  resync when using libmpg123's feeder API.
  That bug has been there since ages but was neutralized by other (buggy)
  behaviour of the parser before the little overhaul. The Law of an Even
  Number of Bugs.

- libmpg123 API version 36 (see NEWS.libmpg123)
- Add --ignore-streamlength.
- But also deal with concatenated (Frankenstein) streams: Not applying
  gapless cutting once we're over the announced number of frames.
- Also, gapless cutting is only applied if relevant information is
  present. Previously, at least the decoder delay has been cut.
- Add --lyrics for printout of lyrics (from USLT in ID3v2).
- Added handling of SIGUSR1 and SIGUSER2 to terminal control code,
  configurable via --ctrlusr1 and --ctrlusr2.
- Some low-level API added (mpg123_framedata(), mpg123_framepos(), buffering
  and resampling control).
- Added info about input buffer fill to mpg123_getstate().
- ReplayGain: Try to detect LAME below 3.95 for the different reference
  level of 83 dB (adding preamp of 6 dB). Version 3.95 itself (not 3.95.1)
  I cannot detect. User-set gain values are not touched.
- Keep silent about missing bit reservoir when ignoring frames. That is
  the whole point (mostly;-).
- Add the new/old WRITE_SAMPLE variant from MPlayer that makes the generic
  decoder faster on older CPUs (x86 at least). Not on x86-64, though (it's
  not enabled per default there).
- disable 3DNow and 3DNowExt DCT36, which is slow on modern CPUs
- Handle out-of-disk in the file writers (a bit, see
- Added Xcode project files to build libmpg123 on Mac OS and iOS
- 32 bit output for sndio output (patch by Brad Smith).
- Remote control interface knows LOADLIST command now.
- Now freeing meta data memory after printout.
- Some documentation cleanup, including manpage fixes by quadrispro.
- Re-enabled wildcard expansion on Windows (been broken by unicode file
  name support).
- Fix MMX tabinit syntax (thanks to Marcel Müller for pointing that out).
- Fix SGI audio output (been broken since change to modules).
- Fix some fringe behaviour (mpg123_getformat() triggering needless reading
  of next frame, possibly hitting bogus MPG123_NEED_MORE).
- Resync limit now also serves to increase amount of skipped junk on
- Better separation of stream end and read error (in case there is no
  stream opened, even).
- Don't always complain about failed free format header search (only with
  verbosity level 3).

- Fix build for ARM Thumb (incompatible assembly in some macros).

- Fix endless loop in module loader when starting mpg123 in a directory
  that does not exist.

- Fix output of one random character in -vvv mode for ID3v2.2 frames
  (no buffer overflow, move along, folks;-).

- Fix reading of ReplayGain values from Lame tag (nobody seen this before?!)
  (both actual values and the distinction between audiophile / radio).
- Accept application/octet-stream as input from HTTP.
- Update man page with encodings.
- Build fixes.

- Fix bogus error on resync with big offsets (hits systems that have bigger
  off_t than int), bug 3393801.
- Fix windows binary package build script again with plugdir.

- Fix an endless loop on bad headers, bug 3267863.
- Prevent compiler warnings about unused variables with gcc 4.6 .
- Fix return value handling of getaddrinfo(), bug 3288333.
- Make decwin size computation work for OPT_ARM and sizeof(long) != 4
  (bug 3288360).

- Updated .NET wrapper (MPG123_SKIP_ID3V2, encsize and mpg123_strlen
  functions) by Malcolm Boczek.
- Fix crash when calling mpg123_decoder() before opening a track -- a rare
  use case since one can choose a decoder with mpg123_new() already.
- Fix format change detection for certain (artificial) concatenated streams
  (p.ex. MPEG 1 frames following MPEG 2 frames).
- Prevent the nasty of bug 3154889: No change of pitch without a current
  audio format. No proper solution, but more like earlier behaviour.
- Fix some compiler warnings.
- Fix a memory leak in the module loader, also some in the examples,
  for being pedantic.
- More use of AC_LANG_SOURCE to silence autoconf.

- Fix --float and --8bit (MPG123_FORCE_FLOAT, MPG123_FORCE_8BIT) regression,
  caused by introduction of 24 bit formats. (bug 3151597).
  Btw: The use of specific format choice via `mpg123 -e f32` and friends
  is encouraged (meaning mpg123_format_none() / mpg123_format() calls in
  libmpg123 API).
- Loosen the parser restrictions: Allow the private header bit to vary. It
  is not mpg123's business anyway (finally suggested by Marcel).

- libmpg123 API version 29 (see NEWS.libmpg123)
- New optimization: ARM NEON (Cortex A series) -- kudos to Taihei, again!
- Added 24 bit integer output (truncation of 32 bit integer).
- Cleanup of symbols for the static library: Only have global symbols that
  are actually needed, prefix them with INT123_ to avoid conflicts.
- Added --streamdump to get a copy of the data libmpg123 used.
- Added --icy-interval and tweaked libmpg123 to make playback of ICY stream
  dumps possible.
- Added possibility to set module file suffix at build -- to be able to
  avoid .la files for plugins.
- Removed compiler directives for alignment and disable associated tests.
  The tests don't work on a wide range of compilers anyway, and newer gccs
  do have force_align_arg_pointer to fix bad stacks.
- There is no ChangeLog file anymore... it was autogenerated from subversion
  log anyway, so look there if interested.

- Fix bug with the new meta data printout of the mpg123 app; embedded
  null bytes in UTF-8 strings in non-UTF-8 environment would trigger buffer
  Thanks to Jakub Bogusz of PLD Linux for the tip. And yes, shame on Thomas
  for changing working string handling code. At all.

- Fix formatting of meta data printout in UTF-8 (string length computing).
  As a collorary, ID3v1 data is reduced to ASCII in UTF-8 environment,
  as we do never know what specific encoding that has.

- Fix a long-standing bug that subtly broke http streaming with ICY meta
  data when reading a bit more quickly, like when using the buffer.
  I was a very simple and embarrassing bug, actually ... and it made
  me wonder how normal streaming still worked rather well, generally:-/
  Thanks to Lint for staying with us and reporting bug 3022850, to finally
  get a grip on streaming reliability.

- Fix for nasty memory corruption on configurations that hand out addresses
  in the upper half of the memory space (the poiner alignment function broke
  in that case, bug 2996045).
- Various build and portability fixes (including making some exotic
  configure switches work (again)).
- Add dump_seekindex example
- Sync mpg123_clr to 1.12 feature set, patch provided by Malcolm Boczek
  And actually include it in release tarball!
- Rework mpg123.h logic for large file stuff a bit, clients can control
  it by defining MPG123_NO_LARGENAME or MPG123_LARGESUFFIX.
- Include dumb wrappers to provide names suffixed with _64 on 64 bit
  machines and _32 on 32 bit machines, respectively, to help clients that
  insist on defining _FILE_OFFSET_BITS where it is not needed (or with a
  non-large value).
- Tuning of the internal buffer code for feeder mode to minimize its
  performance impact (works in 4K blocks now).
- Workaround for compiler bugs in Open64/PathScale/SunStudio (bug 3004396,
  suggestion by Doug Gilmore).
  Note that Sun Studio is still a tricky fellow, at least when it comes
  to our preprocessed assembly.
  Also, x86-Open64 recently produced another segfault in layer3.c on my
  box... GNU and intel compilers are what one can use as something stable.
- Make mpg123_getformat() return more error codes (like MPG123_NEED_MORE).
- Fix handle I/O for clients with small file offset (32 bit when libmpg123
  has 64 bit).
- Fix 3DNow(Ext) standalone builds.

- Update INSTALL for the dual-mode library. No code changes.

Attention: This reverts the big bad libray name change. People did not
like at all ... so Thomas went along with proving that
being responsible for a open source library package sucks for your life --
he coded large file dual-mode into libmpg123. One library with both APIs,
like glibc.

- libmpg123 25.0.25:
-- The separate libraries are gone again, this one is dual-mode (using
   ca. 750 lines of not very funny code).
-- The library even translates callbacks between large file support modes.
-- Introduce handle-based custom I/O in libmpg123.
-- The header now defines MPG123_API_VERSION to help compile-time detection
   of available functions (lib is version 25.0.25 now... MPG123_API_VERSION
   is 25).
- Make the usage of AI_ADDRCONFIG dependend on a configure check. This
  flag is not available everywhere.

Attention: There was a re-release of the source archive, because the
initial one missed the actual changes for the library name.
Sorry for the inconvenience:-/

- Big bad libray name change for builds with enabled large file support.
  That will break libmpg123 binary compatibility on Linux/x86 when upgrading
  from mpg123>=1.5.0 with enabled large file support. You need to relink
  with the new library called libmpg123_64.
  If you do not enable large file support, or it is implicit on your
  platform (x86-64), the library keeps the old name of libmpg123 and
  everything is fine.
- Assembler cleanup:
-- no use of compiler-based stack alignment (for SSE), helping portability
-- mostly performance neutral, hint at faster decoding for the dynamic
  build and slower for static, but small change in any case
- Windows networking (access http streams in mingw builds).
- Do not ask for addresses that do not have matching interfaces configured
  (IPv6 on a IPv4-only host, for example). That can help with broken DNS
  servers that choke on IPv6 addresses (T-Online Speedport W502V, for example). 
- Fix a potential ICY reading problem were ICY metadata could be read as
  mp3 data (basically for MPEG frames bigger than the ICY interval).

- Help multilib setups that boldly assume that headers are invariant
  by removing the consistency check in mpg123.h and instead just adding a
  suffix to relevant functions when _FILE_OFFSET_BITS is defined. Invalid
  setups will now fail at the linking stage instead of the preprocessing stage.
  That should cover bug 2934488.
- Fixes for gapless decoding:
-- Correctly skip padding larger than one MPEG frame (strange, but occurs).
-- Bug 2950218 (proper gapless cuts for seeking near the end).
- Make sure the search for the first header aborts after 64K of one-byte
  reads, as always intended (bug 2951334).
  Certain files (or even random data) were able to fool mpg123 sufficiently
  enough that it basically reads through the whole file in 1-byte steps
  in search for MPEG data.
  The limit of 64K will very likely be configurable in a future mpg123
- Fix subtle bug with seeking in feeder mode: A seek close to the current
  position (meaning: the needed data offset is already in mpg123's buffer
  chain) resulted in skipped bytes.
- Improve the feeder mode for handling large amounts of junk at the
  beginning, including proper passing of the request for more data when trying
  to determine free format frame size. This also fixes endless looping as
  reported in bug 2951334.
- The parser code still needs overhaul. It is not pretty.
- Compile fix for OpenAL.

- Confirmed/fixed on OS/2 platform.
  At least a static build works with native audio output and generic decoder.
- Make modules work with libtool-2.2.6b .
- Added pitch command to control interface.
- HTTP header field and MIME type parsing now case-insensitive (fixes
  bug 2904736).
  Also, the MIME-parsing for non-playlist URLs is done the same way as
  from playlists now.
  And, we interpret MIME types only to before a ";" (see bug 2836904).
- There is a build-time option to enable unicode file names on windows
  now. UNIX (Linux) does not need that, UTF-8 works natively.
  An unicode-enabled build of libmpg123 expects file names as UTF-8 strings,
  opening of non-utf8 strings (by old applications) still works via fallback
  when the conversion of the string or opening of the file failed.
- Also, on Windows, --realtime now works, along with added granularity of
  the new --priority parameter. Note that you get the latter functionality
  through nice/renice on UNIX, no need to add that to mpg123.
- Added experimental framebyframe API.
- Added a new Microsoft common language runtime wrapper (aka .NET wrapper)
  written by Malcolm Boczek. See ports\MSVC++\2008clr\mpg123clr\ReadMe.txt
  for further information
- Added Visual Studio 2010 Beta 2 project files for libmpg123
- Updated autotools machinery, is gone, use `autoreconf -iv`.

- Fix logic failure in gapless code: The trailing silence was only cut on
  the first decode. So, seek-back and decode again never was gapless:-(
- Fix linker flags on MacOS 10.6 .
- Improve portability (mainly for Solaris/x86, possibly others).
- Fix reading from standard input in Windows (use binary mode).

- Fix OpenAL output issue (bug 2840085)
- Fix theoretically possible discarding of some audio samples when getting
  a new format.
- Fix unwanted verboseness (some error messages from library even when
  MPG123_QUIET was specified).
- Fix back-to-beginning for feed_skip (fixes bug 2859531).
- Fix the CPU creep bug with NtoM decoding.
  That one has been nasty on you with large files since bad frame fillup
  has been introduced.
- Fix regression (since 1.9.0) in ID3 parsing: Now Foobar2000 RVA fields
  are interpreted again (among others).
- Fix some compiler warnings, minor cleanup.

- Free format support.
- ARM assembly optimizations (yay Taihei again!).
- Dither noise now runtime-generated, thus less bloated binary (thanks
  This means that dithered decoding is now included per default.
  Please note that work is ongoing to improve the dithering quality,
  more to come.
  For example: Accurate rounding to 16bits is now always used for generic
  dithered decoder, there are plans to adapt the noise for different
  sampling frequencies.
- Buffer plays any audio type now.
- New output module: OpenAL
- Bug fixes:
-- Broken frame fillup for non-signed 8 bit formats with correct zero
   (bug 2805665)
-- Crash on invalid UTF-16 in ID3 tags fixed (thanks to Andy Hefner)
-- Bad detection of separated UTF-16 strings in ID3 tags (bug 2834636).
- Topic ID3 and UTF: Now we are more tolerant with UTF16, even accepting
  those broken tags that id3v2 produces (multiple conflicing BOMs).
- Build system allows AltiVec on non-Macs now (don't enforce unsupported
  gcc flags).
- Terminal control mode is more verbose (gives hint about pressing 'h').
- The library now offers a switch and small API to defer translation of
  ID3 text data until client app wants to do it (or not).

- Include improvements to integer decoder from SVN trunk.
  They have been accidentally promised on 1.8.0 release announcements,
  though they were not there. Here it is: ISO compliant fixed point decoding
  (speed optimizations are still reserved for next release).
- Fixed build with ./configure --enable-modules (bug in configure script).
- Fix name of binary in help screen when called with path.
- Build fix for old gcc with AltiVec code.
- Update of Sony PSP build files.

1.8.0: (The Taihei Monmament)
- New and improved SSE optimizations! For x86-64, too! Also AltiVec! Fast
  float output! Faster stereo!
  Finally, this should put mpg123 into the efficiency-leading position on
  current hardware! Thanks go out to Taihei Monma for pushing lots of new
  assembler code.
- Added proper rounding modes to 16bit integer.
-- Use ./configure --enable-int-quality to prefer accuracy over speed.
-- Accurate code (generic, SSE, 3dnow -- i586 has always been good)
   delivers ISO compliant quality with 16bit output.
-- Uses fast rounding for IEEE754 floats.
-- Mpg123 can do fully ISO compliant MPEG audio decoding to 16bits (of
   course to 32bit float/int, too).
- subtle performance enhancements via broader data alignment
- Prefer SSE code over 3DNowExt.
-- It can be tight, but generally our new SSE code is better on CPUs that
   can do both.
- Remove HF cutoff (SFB21), has been introduced in 0.59r .
- Fix layer 1 regression (been there since version 1.x).
-- Layer 1 ISO compliance test is positive now.
- Fix layer 2 regression (false positives header change).
-- Layer 2 ISO compliance test is positive now.
- more sample-accurateness
-- Clear frame buffers before seek.
-- Decode 4 frames in advance for layer 3 (2 for layer 1 and 2) per default.
-- Number of layer 3 frames to decoded in advance (and fill bit reservoir) is
   runtime tunable (--preframes).
- Do not stop decoding after stray ID3v1 tags in streams (bug 2793677).
- Fix handling of negative RVA2 values in ID3v2 (bug 2792417).
- Discard bitreservoir of previous frames if a resync is needed.
- improve perceived randomness of --random (do not repeat too early)
- Fix special crash case in UTF16 conversion for ID3v2.
- MPG123_MODDIR environment variable for changing module directory at runtime
- arts output fixes (bug 2715128)
- win32 output swallowing less buffers at end (but still some audio is lost,
  bug 2722991)
- tested mpg123 on OS/2; it works, but not the native audio output
  (SDL works)
- WAV output announces supported output formats correctly
- fix buffer startup to make sure the first bits of audio are not discarded
  (bug 2796802)
- re-enable the --gain parameter (the move to modules disabled that),
- Libltdl is _not_ included anymore.
-- The host system is expected to provide it, otherwise plugins are not used.
-- Our copy does not get in the way anymore.
- enhancements of the MSVC++ builds, including optimizations using yasm
- general build system fixes
- Legacy build system is gone.
- code cleanup
-- signed/unsigned integer usage, make compilers more happy)
-- simplified synth selection code (it is _less_ code now, and does more)
- Fixed behaviour of MPG123_NEW_FORMAT: Now it does not stop you from
  just continuing decoding without querying the (new) output format, this
  behaviour regression had been there since somewhen after 1.4.3 .

- Fix a regression introduced with the --doublespeed fix. Now seeking should
  be a lot smoother again.

- Fix a possible vulnerability in the ID3 code (negative encoding values).

- Just one fix for making the RVA switch in terminal and generic control mode
  work again (instead of plain silence).

- Lots of output format fun!
-- now supporting 8, 16, 32 bit signed/unsigned integers,
   32bit float (64bit float can be hacked in quickly)
-- optimized decoders use fallback routines for non-optimized
   formats transparently
-- flexible resampling is automatically chosen when output does
   not support standard rates (like a JACK server)
-- That meant quite some restructuring in the decoder code...
- Lots of core functionality now optional, enabling small-footprint
  builds of libmpg123.
-- choosing among the MPEG layer codes
-- NtoM resampling
-- downsampling
-- feeder API (mpg123_open_feed())
-- ID3v2
-- String API
-- ICY
-- output formats
-- error/warning message printouts
-- ... a layer3-only shared libmpg123 can be just 107K on x86-64
   or 94K on x86
- --preload (tune prefill before playback for buffered mode)
- win64 support (one report at least;-), modules working on Windows
- Windows/MSVC++: fixed handling of unicode file names,
  mpg123_topen() now actually working there
-- You are obliged to use mpg123_tclose() now!
- using "plugins" subdir relative to mpg123 binary
  (in addition to ../lib/mpg123)
- switch to libtool 2.2 and automake 1.10 for the build scripts
-- though no update to new libtool 2 API, yet
- some portability fixes
- prevent user from creating non-working I/O setups
  (replacing timeout reader, ICY with feeding)
- JACK output accepts port names for -a, stereo gets mixed down
  when only one port is chosen
- added sndio output, contributed by Christian Weisgerber
- Make --doublespeed paramter work, porting of which has been
  forgotten in the 1.x series!
  Along with that, a regression in decoding has been fixed
  (concerning skipping of frames in general). 

- Fix for consistent NtoM resampled decoding and broken frames fillup.

- Fix bug #2413608: Make sure a new format does not slip through
  after early close.
- Include equalizer example file and a piece of wrapper code for MSVC++
  in the distribution.
- Do not claim to support unsigned 16 bit encoding (never been the case).
- Fix default fifo detection, now it actally activates fifo support!

- Fix default IPv6 detection in (a typo has been introduced

- Fix printing of largefile offsets in remote control interface.
- Some portability fixes (reduced compiler warnings on Solaris, p.ex.).

- small things inside that I forgot to mention
- some new commands for remote control interface:
- added --keep-open command line flag (for remote control)
- fixed equalizer file parsing to really handle comment lines as example file
- more elaborate detection of UTF-8 locale, also --utf8 switch to force it
- added --fuzzy to enable inexact seeking (see below)
- added --index-size parameter for tuning accurate frame seek index
- enabling largefile support via configure
- libmpg123 11.0.11:
-- added mpg123_geteq(), mpg123_tell_stream() and mpg123_getstate()
-- API cleanup (take const char* for mpg123_open())
-- calm down complainling (C++) compilers by completing the mpg123_channels
-- fixed several possible crashes when user provides NULLs where it isn't
-- workaround for frames that fail decoding (bad frame body, missing bit
   reservoir): fill up with silence
   That also fixes seeking issues related to bad frames since now the
   fixed relation between frame and sample offsets is restored.
-- Finally realized what that the resampling code for integer math
   (generic_nofpu) was just incomplete!
   Now it works, using the defined integer multiplication.
-- Detect UTF-8 in ICY conversion (instead of just assuming CP-1252,
   provided by Torsten Glaser).
-- Prevent ugly noise when the bit reservoir is not fully there
   (p.ex. ignorantly cut mp3 files).
   Concerned frames are silenced, but reservoir for following frames
   is kept.
   Thanks to Patrick Dehne for the investigation and patch!
-- New option for fuzzy seeking; trading sample accuracy for speed.
   With that option, seeks positions are guessed using mean framesize or
   the Xing VBR TOC is used.
-- Feeder streams work with seek index now.
-- Seek frame index is runtime tunable.
-- more caution with alignment issues for varying library users
- new project files to build/use libmpg123 with MSVC++ 2005 and 2008,
  provided by Patrick Dehne
- fixed a regression that made the pause/loop mode in terminal control
  slowly go backwards

Big note for libmpg123 users: Make sure your application is compiled
with the same largefile setting as libmpg123!  If that does not match,
you will get unresolved symbols (which is better than weird errors when
using a nonmatching off_t in arguments;-).

- fix wav/au/cdr writing to stdout (bug #2044423)
- add more sensible --no-resync to replace --resync (bug #2044404)
- libmpg123 7.0.7
-- fix thread safety of SSE decoder (bug #2053181)
-- fix robustness of length handling, added mpg123_set_filesize() (bug #2064322)

- libmpg123 6.0.6
-- Floating point decoding back again, (build time choice for now;
-- More robust string handling; extended string API.
-- fixed an (so far) unnoticed ID3 bug that may have obscured comment fields
-- added mpg123_feed() as shortcut; using const input buffer now
-- resync is enabled again for ICY streams (the trouble in the past was
   due to a reader bug, not the streams)
-- decode ICY text data to UTF-8 (thanks, Thorsten)
-- preserve file flags for timeout mode
- More elaborate (helpful?) error message for output format problems.
- Reworked HTTP code... and added IPv6 support!
- Fix the --proxy parameter (has been accidentally ignored before).
- Added --ignore-mime to work with web streams with funny playlist MIME
  types like "audio/megurl".
- Know MIME audio/mpegurl.
- --timeout is now also used for the connection itself (not just reading).
- Now trying to resume ALSA device after suspend.
- subtle build changes (your CFLAGS rule now)

- Fix mpg123_tell() and mpg123_seek() return values with respect to the
  samples in decoding buffer (bug 1947373).
- Fix (bogus) 3DNow detection on IDT C6 (bug 1947982).
- Fix mono playback of win32 output (bug 1955157).
- Fix ABI compatibility with compilers that don't align the stack like
  gcc does (fixing SSE crashes on win32).

- Fix build problems (for Cygwin and Win32, perhaps others, too).

- Fix a serious regression of terminal control mode (forgot to remove a
  term_restore(), bug 1936548).

- Enable runtime check for available and working output modules.
  Now mpg123 tries all built modules (or a list specified in
  mpg123 -o moda,modb,modc) in turn to find a working output. This minimizes the
  situation of having to specify a device when default output doesn't work for
  some reason.  See bug 1910500.
- libmpg123 2.2.2
-- Fixing an endless loop for ICY streams that suddenly end without error
   (just EOF).
-- Further fixes around EOF handling in (buffered/ICY) streams.
- Enable user to interrupt mpg123 when stuck in a loop of failing to open a
  track (with --loop).
- This time with correct RPM spec (hopefully... sorry, Michael!).

- Enable snd_pcm_sw_params_set_xfer_align() for alsa-lib < 1.0.16 (bug 1908603).
- libmpg123 2.1.2 (no-op change)

This is a _major_ fix release for http streaming!
Much more usable than earlier mpg123 1.x versions.

- libmpg123 libtool-version 2.0.2
-- Fix nasty bug in ICY reader that failed to account for partial reads
   from slower network.
-- Now non-seekable streams can get peek-ahead for MPEG sync, too, triggered
   by the MPG123_SEEKBUFFER flag.
-- Do not try to resync on streams with ICY metadata, that won't work.
   Consequently, if there's unreliable network, use --loop -1 to make
   mpg123 reconnect to the server on errors.
-- Applying a few layer 3 decoder fixes from old mhipp CVS (hail to the
   creator;-) That includes the fix for bug 1895025.
-- Fix ID3v2 parser with very invalid claimed ID3v2 frame sizes (could
   have triggered DOS before).
-- Oh, and fixed a steady memory leak in ID3 handling. Should mention that;-)
- Print ICY-NAME and ICY-URL again, as mpg123-0.6x did.
- Added --smooth option for not flusing buffer between tracks but at program
  end only.
- Included some contributed ports/bindings of libmpg123 under ports/ .

- libmpg123 libtool-version 1.2.1 (coincidence!)
-- Earlier check for free format headers to prevent them from screwing up
   resync (free format still not supported)
-- bug 1889051: more flexible resync, not insisting on the format of the
   first frame (which could have been bogus)
- Add an output write loop to handle SIGSTOP/CONT operation (bug 1890563).
- Fixed a bug where playback status for some http streams was not shown.

- libmpg123 libtool-version 1.0.1
-- See NEWS.libmpg123 for details.
-- This adds --resync-limit parameter to mpg123.
- Some build system work, including:
-- It is now possible to build a mpg123 rpm from the source tarball via
   rpmbuild -tb .
-- configure: Accept --with-audio=foo,bar,bla (comma-separated list) in addition
   to --with-audio="foo bar bla".
-- included in distribution.
-- Include the forgotten i486 files in the distribution (now that's embarrasing...).
   Also a little fix for wrong --list-cpu on i486 build.
-- Fix build with MinGW32.
- Fix -v stat output (time display) for MinGW32 build.

- Added arts output.
- More verbosity on lame/xing/info tags (on demand) instead of compile-time
  debugging switch.
- XMMS2 plugin cleanup (submitted to XMMS2 for inclusion)
- Improved ID3v2 robustness (let bad tags not affect playback of the important
- Terminal control:
-- Key for going to previous track ([d], because it's left of [f] usually...
   no config file yet for that).
-- Key for printing current playlist, [l].
-- Now possible to switch on verbosity also for mpg123 -q -C
-- Key for (re)printing ID3 tag info ([t]).
-- Key for (re)printing MPEG frame header info ([m]).
- Soften the -q option: Now you are able to increase verbosity after -q
  again, like mpg123 -q -v (this fixes cplay's progress display).

- Fix module loader to prevent crash in working paths with more than 49
  characters length (bug 1859413).

- Fixed Pulseaudio.
- Cleaned up configure.
- Pushed remote control version to v3.
- Updated documentation.
- Fix VPATH build for generated mpg123.h .

- Fixed JACK output.
- Disable buffer when statically building CoreAudio in (MacOSX); with
  CoreAudio module the buffer works, with static output it doesn't.
  This is the situation because of fork() discrimination by Apple Inc. .
- Added verbosity to tell about ID3v2 text encodings (level 4).
- minor cleanup

- Fixed some bugs, the ID3v2 interface of libmpg123 is safer now (there
  were possibly stale pointers in there).
- Buffer mode got vital improvements (yeah, that is sort of feature creep
  during release process, but worth it)
  The improved communication makes it possible to skip tracks in terminal
  control mode without the hack to immediately restart playback.
  To be able to do seeking in terminal mode with buffer properly, buffered
  playback stops on seek keys. You restart playback after you are done
  with seeking.
- Support either portaudio v18 or v19 API (v18 works for me, v19 works
  according to user report).
- Audio output handling got safer and more clean, a byproduct of implementing
  the above. I stumbled over some things and had to fix it (yeah, feature
  creep in parts again, but still: Worth it!;-)
  Any audio access is now restricted to either that standalone mpg123
  process or the buffer process, the output being opened _after_ the fork().
  There is hope that this fixes buffered output on MacOSX.
- More feature creep, this time because of something like that was already
  included in the Debian mpg123:
  --delay parameter to issue a pause before each track (does not make
  sense for file writing outputs, though).
- Even more feature creep: remote control can now give ID3v2 data, also
  in more raw form via new "tag" command.
  This sort of qualifies as bugfix, actually, since it's item 1786430 on
  the tracker;-)
- Make flexible rate computations more safe (especially on 32bit platforms)
  by switching to the looping code instead of direct multiplication/division
  (which blows up with integers). A possible effect was premature track
  ending with a forced rate.
- Flexible rate mode works _at_all_ again with fixed output support
  detection code (comparison of -1 with a size_t ended a loop all too early).
- check for error in flushing output (this ends mpg123 for a broken pipe,
  for example)

- In one word: LOTS!
- Many small things that happened while migrating to the new setup.
- The decoder is now a library (libmpg123) which can be installed also
  for other programs to use. This finally provides a heavily improved
  replacement for the aged mpglib. We have doxygen API doc, example codes.
  You can still choose to compile the decoder statically into the mpg123
  binary, though.
- We have a plugin system for audio output! This should be relieving to
  package maintainers for binary distributions, drastically reducing the
  number of builds they have to provide.
  Here, too, you can choose to build one audio output into the binary
  like before; you can have your old-school one-piece mpg123 or the hip
  I-am-legion shared-lib-and-plugin flexible mpg123;-)
  One exception: For now, you have to compile without modules in
- ID3v2 support has been enhanced and, in the case of specific comments
  (those with a description), fixed.
  The ID3 parsing is part of libmpg123, offering support for virtually
  all textual information in ID3v2.2, ID3v2.3 and ID3v2.4 tags (along with
  the usual 128 bytes of ID3v1 or ID3v1.1 in a block).
  All tag texts are converted to UTF-8, allowing mpg123 to properly display
  special chars in an Unicode terminal without fuzz (for non-UTF locales,
  there is still the ASCII reduction).
- Gapless decoding (of layer 3 streams) is now considered to be a standard
  feature and is enabled per default both at build- and runtime.
- ...

- Fix bug 1807964: distorted playback of MPEG 2.5 files

Thanks go out to the ROM 106.5FM team ( for supporting
this release with a donation on and a bottle of finest Gewuerztraminer
wine from Luxembourg!

- verified that it builds on SGI IRIX 6.5 for MIPS32 and MIPS64
- enable flexible rate for stdout decoding, too
- UTF8 filtering to ASCII finally works thanks to Mika Tiainen, see bug 1745847
- corrected frame offset; -k was off by one
  This means now that -k disagrees with mpg123 0.59r on many files because
  the old version does play a frame of silence when encountering LAME tag
  instead of dropping it for audio.
- added --timeout <n>: Possibility to end playback on a stream that doesn't give
  data for <n> seconds.
- added --loop <n>: for ... well... looping playback of a track; motivation was
  persistent internet streaming with unreliable connection (together with --timeout)
- build fixes
- use movap instruction in SSE assembly when compiler supports the needed
  alignment (idea from MPlayer)
- show decoder options in their priority order in --list-cpu/--test-cpu
- added Perl script to benchmark each of mpg123's CPU optimizations
- portability enhancements and cleanups along with testing on SunOS 4.1.2
  with gcc-2.7.0
- workaround for ugly flexible rate for MMX/SSE/3DNowExt code:
  --with-cpu=mmx also includes i586 code as fallback; similar with the others

- support for a combined build of x86 optimizations and runtime selection
  among these The (supposedly) "highest" CPU-supported optimization is chosen
  by default, --cpu, --list-cpu and --test-cpu parameters are there for you to
  take control.
- default build for x86 systems is combined "x86" with generic, i386,
  i586, MMX, 3DNow, SSE, 3DNowExt; "x86_dither" also includes the dithering
  i586 decoder
- no automatic -march compiler flags anymore; the --with-cpu configure option
  selects specific assembler code but special CFLAGS are up to you
- extended 3DNow and SSE optmizations from MPlayer's mp3lib
- x86 assembler optimizations adapted for MacOS X/Intel
- straight support for Windows using MinGW32, reactivated the native
  WinMM output. Core functionality including assembler optmimizations
  and WAV writing is there.
- assembler optimizations also for cygwin
- set remote-err automatically when putting audio to stdout
- a bit more general portability in the new build system
- fix file creation mode for -O
- experimental decoding to 32bit float (raw and WAV file): ./configure
- generic control interface (-R) can create and read from a named pipe,
  parameter --fifo <path>
- audio bugfix: playback of mono file on stereo-only hardware, better
  error reporting there
- ALSA output handling underruns
- MMX, SSE and 3DNowExt fallback to i586 and below or 3Dnow for forced
  sample rate
- more terminal control: volume, RVA and verbosity controls

- now allow crc bit to vary over stream (fixes files with bad lame tag
  and some internet streams)
- do not try to parse id3v2.2 (as opposed to 2.3 and 2.4)
- correct id3v2.3 parsing for not expecting syncsafe integers for framesize,
  2.4 introduced that
- omit gcc specific compiler flags when gcc is not there
- initital attempt to support sgi/irix with configure (please test)

- fixing a potential denial of service problem; broken connection to server
  could trigger an endless loop

- build fixes, only relevant to developers and people who had problems building

- JUMP command in generic frontend accepts offsets/positions in seconds
  (JUMP 4s vs. JUMP 4)
- initial support for ICY meta data in SHOUTcast streams
- a bit of fixing with rva code, added rva and general volume control to
  generic frontend
- accepting more MIME types for http streaming (application/pls and
- ID3v2 info output taking less space per default - more resembling the
  old ID3v1 view for short tags, full view with always one line per item
  via extra --long-tag switch
- generally less verbosity (i.e. noise on console) per default
- ID3v2 now having some kind of support for unicode; trying to print what's
  latin1-compatible and prevent messing up of terminal with unknown bytes
- several bug fixes and small improvements

- improved AltiVec code from Taihei Monma
- an i486 optimization by Petr Salinger
- improving interactive experience (terminal, frontends)
-- prevent/lessen bad effect after seeking while paused
-- show targeted position while seeking in terminal mode while paused
-- not messing up terminal on unexpected errors
- decoder robustness - better file sync (weird junk at beginning, some vbr
- build system improvements:
-- fixing configure isses on some platforms
-- adding cygwin, kfreebsd
- extended fix for CVE-2006-1655
- generic code should compiling again...

0.60 final: (The Team!)
- no functional changes

0.60-beta6: (Thomas Orgis)
- more use of ID3v2 tags, mpg123 now displays both (merged) info from
  ID3v2 and ID3v1; trying to support both ID3v2.3 and ID3v2.4, but ignoring
  encoding (UTF8 should still work if your terminal supports it, straight
  unicode should work except for parsing genre numbers and multiline strings)

0.60-beta5: (Thomas Orgis)
- more robust playlist support:
-- path/url length not limited to 1023 characters anymore
-- reading in of whole playlist at program start (old operation was not
   really safe with playlist from http)
-- fixed bug where reading of playlist form stdin caused segfault (introduced
   with the pls format support)
- smaller cleanups, screen output cosmetics

0.60-beta4: (Nicholas Humfrey & Thomas Orgis)
- proper seeking, also for VBR streams
- real two-way seeking in remote control mode at all
- better estimates of remaining frames (on vbr files), exact if Xing header
  is found
- Removed libao support (as it is GPL and we aren't - and we don't need
  it with native Alsa)
- Added ALSA 0.9/1.0 support by Clemens Ladisch (old alsa is in
  audio_alsa05.c but will most likely vanish soon)
- Added PortAudio audio-output support
- Added SDL audio-output support
- Fixed NAS audio-output support
- re-enabled MPEG 2.5, code just seems to work
- improved stream checking; being more willing in general (favor resync
  instead of bail out) but directly give up if free format stream is obvious
- Whee! Initial Relative Volume Adjustment (buzz) support with reading of
  ReplayGain values in lame/info tag, the RVA2 frames in ID3v2.4, ID3v2 TXXX
  tags from foobar2000 (please test, someone!) and custom ID3v2.3 comments
- more persistence in trying to get http streams

0.60-beta3: (Thomas Orgis)
- fixed bug with segfault in http code on stream from -
  bug id 1529266
- made http request HTTP/1.0 compilant again (tested only without proxy) -
  bug id 1529319
- now using http content-type header to check for correct types mpg123 can
-- you can now always use -@ for a single http resource (direct stream or
   playlist) since a stream will be reopened based on MIME type,
   though omitting it will save you one http request to the server in the
   direct stream case...
- now simple parsing of shoutcast pls playlists (how did you people use
  shoutcast before?)
- playlist mode enhanced with selective playback of one entry in list
  (--listentry parameter)

0.60-beta2: (Thomas Orgis)
- fixed bug in terminal control mode that caused the f key to quit program
  isntead of advance to next file (introduced by the first attempt to restore
  "normal" Ctrl+C behaviour in terminal control mode)

0.60-beta1: (Thomas Orgis & Nicholas Humfrey)
On the way to the first new offical realease since years!

- new license: all LGPL except the old audio_alsa.c
- merge with Nicholas' sourceforge package
-- autooled configure/make system
-- MacOSX, Jack, libao (alsa 0.9/1.x support through that)
- experimental gapless playback for layer 3
- merged in MMX and AltiVec optimizations
- security bug fix of GLSA 200607-01 and catching possible crashes found by fuzzing 
- fixed many issues here and there
- polished help messages, updated man page...

UNDERGROUND 0.59r-thor7: (Thomas Orgis <>) (well, I'm going to
------------------------ release a _really_ official mpg123 in future)
- applied relevant stuff from Debian's 22:
-- fix for CVE-2006-1655 (buffer overflow in III_i_stereo())
-- strict aliasing hack in WRITE_SAMPLE of decode_i386.c
-- fixed http streaming through proxy by actually using the request url...
-- buffer resync on Ctrl-C (cleaning, reducing pauses with multiple files)
-- removed unused esdserver variable from mpg123.h
-- did NOT apply the Debian Makefile/manpage changes (going to rewrite that
   stuff anyway)
- made the netbsd-i386 targets visible as netbsd-x86

UNOFFICIAL 0.59r-thor6 (Thomas Orgis <>)
- split nedbsd targets for realtime priority selection code; since one
  needs -lpthread with it. standard netbsd target doesn't use realtime code --
  I won't recommend running mpg123 as root anyway!

UNOFFICIAL 0.59r-thor5 (Thomas Orgis <>)
- Separated the the use of char* equalfile as filename from the use as
  TRUE/FALSE flag by introducing int have_eq_settings. Compilers kept nagging
  about assigning and integer (TRUE) to char* ... and it feels more right
  this way.
- Made control_generic aware of real being possibly different than float (seq
  command used to raise an FPE on OSF1 when this assumption was not the case.
- Fixed the command line arg parsing to work properly on Alpha (32 and 64
  bit mixup)

UNOFFICIAL 0.59r-thor4 (Thomas Orgis <>)
- merged with 0.59r-19 from Debian project (Daniel Kobras
  <>), hopefully fixing all those security vulnerabilities
  -> my dirty http_open and open_stream variants now replaced by more
  proper code
  -> still my (rather workloaded) parsing/processing in control_generic,
  but now including the idea to watch out for \n, \r\n and \r (or \n\r,
  for that matter;-)

UNOFFICIAL 0.59r-thor3: (Thomas Orgis <>)
- hacked additional variants of http_open and open_stream that return
  errors instead of exit()ing; makes more sense when used by a frontend
- further cleanup (and return to using continue instead of nested
  if(){}else{if()else{}}) in command processing, better error messages there

UNOFFICIAL 0.59r-ThOr(remote-err,eq,flushing): (Thomas Orgis <>)
- added audio_flush() calls to various places in control_generic.c to
  make sure there is nothing left in there that hurts later (not sure if
  this has effects)
- (re?)enabled (?) code for stream info
- corrected answer of JUMP (now gives actual reached position, +/- 1 frame
  since I'll still have to figure out what frame 0 really is...)

UNOFFICIAL 0.59r-realeq-remoteerr: (Thomas Orgis <>)
- applied patch from Cool Feet Audio project (, see to use real-time equalizer in remote mode
- applied patch from AMF ( by Steve Grundell for remote
  communication not polluting the STDOUT where the audio data is read from
  (instead using STDERR) ... and faster seeking, maybe
  The patches go not cleanly together, but rejects are trivial. Thus,
  a combined patch with both would makes sense.
- modified (IMHO improved) the control_generic command reading: now
  always making sure that the full command is read up to a mandatory \n;
  also reworked the application of the commands afterwards
- added SEQ command: simple eq with only 3 bands
- added SILENCE command: I don't want the playback messages when I don't
  read them (since my frontend is reading the STDOUT it knows what is going on)
- added SLJ command: SILENCE-LOAD-JUMP... a hack for my hacked
  mixplayd... maybe one should remove that again.

0.59r: (MH)
- applied audio_alib changes from Ralf Hildebrandt <R.Hildebrandt@TU-BS.DE>
- some 'paetsches' from DK
- minor optimizations in layer3 (dequantize/huffman) 
- BSDOS 4.0 with gcc added to Makefile (<>)
- Solaris-ESD patch from Ryan Prosser <>
- Added Frontend changes from Bertrand Petit <>
- more DK 'paetsches' (final dk7-clubhaus edition), includes extended term 
- SGI audio fix for non RAD machines from Erik Mouw <>
- fixed ugly bug in layer3 dequant for bt==2 (may cause high frequency clitchs)
- AMD-3DNow optimizations in decode and dct64
- OS/2 reworked from Samuel Audet (see audio_os2.c for more)
- added freebsd-esd make-entry from Daniel O'Connor (
- added enhanced head_check from D. Skarda <>
- applied 'xterm-title-patch' from "Wilson, Jeff D" <>
- more esd changes from Robert Bihlmeyer <>
- applied Hannu Napari's <> SGI audio patches
- minor changes in httpget.c and mpg123.c 
- added native AIX support from
- applied playlist patch from <>
- more SGI audio patches from Gilles Zunino <>
- applied esdserver patch from Matthew Parslow <>
- added some fixes for control_generic

0.59q: (MH)
- ESD fixes
- minor bugfixes (DK patch ;)
- au and cdr 'support' (it's a hack)
- Fixed 'VBR' Bug in get_scale_fac1(). Thanx to <> for
  this. (Wasn't really a VBR Bug. The new encoders just triggered it.)
- Added equalizer patch from <>

0.59p: (MH)
- Added mpglib (see mpglib directory)
- 'make generic' is more generic now
- Added HPUX/ALib support from Ducroquet Erwan ( 
- Added EsoundD support from <fill it in>
- Added exp. relatime support (compile with -DSET_RT) 
- Added linux-sparc patch from <hmm lost email)
- Added genrepatch from Shane Wegner <shane@CM.NU>
- Added wav-File patch Samuel Audet <>
- Added SGI-RAD support from "J. Dean Brederson" <>
- Added sajberplay/FreeBSD patch from Chou Ye-chi <>
- Added 486 optimizations from Fabrice Bellard <>
- Added combined output devices on SUN (ie Speaker and headphones together)
- Added audio_alsa.c (under GPL license) for the ALSA sound system from
   A. Hermansen <> and J. Kysela <>
- Added K6-3DNow patches from  KIMURA Takuhiro <>
- Added MINT patches from Petr Stehlik <>
- Fixed minor bug for linux-i486 and mono 

0.59o: (MH)
- fixed Bug with stereo streams on mono audio hardware for OSS
- fixed a bug with mono output and 1:1 (no) rate conversion
- again changes in the probe code
- fix in the buffer code for the case, that audio parameters are changing
- more changes to the flexibel rate converter, should now work when
  writing to STDOUT.
- Added Andy's (<>) float2int speed up proposal 
  (check WRITE_SAMPLE in decode_i386.c) .. Maybe someone wanna take the 
  idea and put it into the decode_i586.s part?
- Added TK3Play support from Brian Foutz <>
  (compiles but untested)
- RIFF Header fix (according to Thomas Niederreiter's
  <> patch)
- m68k patch from Stefan Gybas <> applied
- minor additional PPC changes applied
- resync and EOF hang fixed
- resync and bitrate correction relaxed 
- Slightly change verbose output; added countdown

0.59n: (MH)
- fixed shuffle bug
- added memory mapped IO for file reading
- added MPEG 2.5 12kHz and 8kHz table entry
- started flexibel rate conversion .. 
- new '-Z' / '--random' option 
- minor clean-up
- splitted audio file in audio system dependend smaller parts
- started system stream support (external program: system.c)
- unsigned 8 bit (instead of ulaw) is now default for 8bit operation
- Linux PPC patch from Grant Erickson <> applied
- added BSDi patch from Peter Berger <>, who reported
  a failure when catching signals
- minor bug fix in the control interface for sajber jukebox
- fixed the ugly 'Ctrl-Z in buffer mode' bug .. YES ;)
- added buffer preload when underrun occurs (preload = 1/8 buffer size)
- minor SUN audio clean up

0.59m: (MH)
- Bug with RIFF header streams fixed.
- HTTP auth option patch from Henrik P Johnson <>
- NetBSD patch(s) added from <mycroft@NetBSD.ORG> and <>
- Added advanced shuffle support from Steven Tiger Lang <>
- fixed bug with too large big_value field 
- fixed (hopefully) the bug with the '-a' option 
- work-around for some pedantic compilers in fr->synth init
- some changes for the 'generic' version
- applied NAS bug-fix patch
- new (good) MPEG 2.5 tables (from Niklas Beisert)
- added WIN32 support from Tony Million
- added some bug fixes from Niklas Beisert
- fixed another ugly bug in layer3 dequantize (wrote over the border
  of the 'xr' field)
- a minor HPUX clean up
- added BSD patch from Kevin Brintnall <>
- added Makefile.win32
- added NAS big endian fix

0.59l: (MH)
- simple equalizer support (to be continued)
- added NAS patch from Martin Denn <>
- real mono support
- added pentium optimized decode_i586 from Stefan Bieschewski <>
- fixed a bug in MS dequantize function 

0.59k: (OF)
- fixed "Accept" header in httpget.c
- fixed buffer overflow (causing core dumps) in httpget.c
- fixed display of song length
- fixed -b (buffer) problem on SGI / IRIX -- it definitely works now
- fixed problem when files with different sample rates where played,
  it now works with -b (buffer), too
- fixed order of local variables in decode.c to work around a
  compiler bug in ultrix
- fixed Makefile for ultrix
- reverted to old getbits.s (the new one could cause coredumps)
- more compact MPEG header display in non-verbose mode
- small fixes in buffer code and elsewhere

0.59j: (MH)
- added sajber jukebox control interface (it's currently a big hack)
  (see 'JUKEBOX' file for more)
- added simple rewind facility .. (only used by sajber interface at the moment)

0.59i: (MH)
- added front-end (remote) patch from "Brian J. Swetland" <>
- added OS2 patches from Niclas Lindstrom <>
- fixed 'rewind bug'
- added patches from Leo Broukhis <>
- added shuffle patch from Tillmann Steinbrecher <>
- added i386-getbits from M.Stekelenburg <>
- added fallback to 8bit if no 16bit samples support (currently linux only)
- added (slightly changed) outburst patch
  from Antti Andreimann <>
- added 'http accept' patch from Hur TaeSung <>
- fixed buffer mode bug 

0.59h: (MH)
- added '--8bit' option to force 'ULAW 8 BIT' playing
- pcm buffer type changed to 'unsigned char *' 
- plays 2:1 and 4:1 with the real frequency (ie 11025/22050 instead of 44100)
- program (re)sets (changed) frequency for every frame
- buffer mode is broken

0.59g: (MH)
- added support for MPEG 2.5 (it's a HACK!) .. mpg123 understands and plays
  2.5 but it does it wrong. (you will hear this!) 
- Fixed wrong values in BandInfo struct for 32kHz mode
- Some performance improvements and minor changes and bug-fixes.
- Tried the new imdct from the maplay1.2+ (version 1.81) package. (but 
  it was slightly slower on my system, currently disabled)

0.59f: (OF)
- HP-UX audio fix (provided by Damien Clermonte)
- Buffering (-b option) hopefully fixed.  Please test!

- added Audio MPEG 2.0 support (but near untested) (MH)

0.59d: (MH)
- minor modifications to dct64_i386.c
- removed duplicated dependency 'dct64.o' in Makefile

0.59c: (OF)
- Another complete rework of the buffer system (-b option) (normally uses
  mmap(), but it can also use SYSV-ShMem because mmap() is broken on linux).

0.59: (MH)
- extracted dct64() from decode*.c files.
- some changes to dct64() and decode*.c to allow auto-increment/decrement
- removed bit-counter in getbit-functions. Layer3 part2 now counts it itself.

0.58: (OF)
- Bugfix: not closing the audio device caused a few samples to be lost.
- Bugfix: audio_set_rate() was not called (!?!).
- Additional command line options "--doublespeed" and "--halfspeed" (couldn't
  find better names).  See the manpage.
- Complete redesign of command line options.  See the manpage.
- Changed the buffer algorithm (-b).  I'm not sure whether it's better or worse
- Supports multiple files on the command line.  Pressing Ctrl-C will skip to the
  next MPEG file, pressing it twice within one second will abort mpg123
  completely.  See the "INTERRUPT" section in the manpage.
- Support for http URLs.  See also the -p option in the manpage.
- Support for list files.  See also the -@ option in the manpage.

0.57: (MH)
- Put the downsampling code together with the full quality decode into one
  binary. (new options -2to1,-4to1)
- Optimizations in the dct12.
- Removed the 'int' version.

0.56: (OF)
- Additional command line option "-rs" to try to resync on "broken" MPEG streams
  (now called "--resync" or "-y").
- Additional command line option "-b" to use audio output buffering. See the
  manpage for additional information.
- Changed a few variable names and some other minor fixes, so the
  source compiles more cleanly (without warnings) on most platforms (for
  example, the FreeBSD port compiles without warnings using "-Wall -ansi
- Makefile changes: moved "solaris" to "solaris-gcc", new target
  "solaris" (using Sun's SparcWorks cc, faster tha gcc!), new targets
  "aix" and "generic".

- removed some obsolete code from layer3.c (MH)
- changed scalefac stoing in layer3 (MH)
- decode_4to1.c added again. Renamed decode_lp.c to decode_2to1.c (MH)
  (both not optimized .. still expterimental)
- Minor cleanups and optimizations in layer3. (MH)
- Removed some (probable) minor bugs. (MH)
- Changed 2**(...) scaling (and omitted a few muls) (MH)
- Changed Makefile (OF)
- Changed dequantize to reorder with a mapping table 
  (you can disable this by undefining MAP in layer3.c) (MH)
- removed the 'dummy' scale step for mixed mode (MH)
- Changed copy policy (MH)
- Modified dequant-reorder-mapping (MH)

0.54: (MH)
- Minor changes in the documents .. 
- huffman.c: x->len,y->len removed .. always '15' for the important cases
- audio.c: some additions to the SOLARIS audio code

0.53b: (OF)
- Makefile:
-- Renamed SPARC10 to SUNOS (which is more appropriate).
   Tested with SunOS 4.1.4 on a SPARCstation SLC, 20 MHz (322% CPU,
   so no realtime decoding).
-- Added SOLARIS.  Tested with Solaris 2.5.1 on an Ultra 1, 167 MHz (14%
   CPU), a SPARCstation 5, 110 MHz (47% CPU), and a SPARCstation 10, 33 MHz
   (68% CPU).
-- Added FreeBSD.  Tested with 2.1.7 on a Pentium-90 (38% CPU) and 2.2 on a
   Pentium-166 (20% CPU).
-- Added Digital Unix.  Tested with DU 3.2 on a DEC 3000/500 (Alpha AXP
   150 MHz, 33% CPU).  Does not support DEC's audio interface ("AudioFile"),
   decodes to stdout only.
-- Improved "clean" target to remove core files and editor backup files.
-- Added "install" target.
-- Added "dist" target to create a .tar.gz file of the source directory.
-- In the distributed Makefile, all system sections are commented out.  When
   you type "make" without uncommenting one of the sections, an appropriate
   error message will be displayed.
- audio.c
  Fixed problem when opening /dev/audio on Solaris: I had to remove O_NDELAY,
  because writing to /dev/audio didn't block until the requested number of
  samples could be written, but instead it wrote only as much samples as
  it was able too.  Sounded funny.  Removing O_NDELAY fixed the problem.
  I don't think that it hurts SunOS either, so I didn't make another #ifdef.
- common.c
  Fixed bug by adding function finish_output(): The decoder writes the audio
  samples into a buffer, and each time the buffer is full, it is written
  to the audio device (or to stdout).  But when the decoding is done,
  there are some samples left in the buffer which aren't written yet --
  they were just discarded.  This additional function fixes the problem by
  writing the remaining samples from the buffer.  It is called immediately
  after the main frame loop in mpg123.c.
- decode*.c
  Fixed compiler warning on symbol "div": Some compilers issue warnings
  because the symbol div is already used for something else (see "man div"),
  so I changed it to "divv".
- layer*.c Functions do_layer*:
-- Moved the declaration for audio_play_samples into mpg123.h.  Makes the
   code cleaner.
-- I moved the variables pcm_samples and pcm_point to common.c, so they are
   shared between all modules.  This was necessary because the finish_output()
   function (see common.c above) needs to access them.
-- Renamed loop variables "i" and "j" to "k" in order to fix compiler warnings
   for duplicate/hidden symbols (similar to "div" in decode*.c, see above).
   Changed the preprocessor symbol AUDIOBUFSIZE into a global variable
   "audiobufsize".  This makes it possible to change it by a command
   line option.
- layer1.c
  Fixed error in function I_step_one, second parameter: unsigned int
  scale_index[2][]  -->  ...[2][SBLIMIT] This caused an error with the DU
  compiler (which is ok, because [2][] is incorrect according to ANSI).
  I don't know whether [2][SBLIMIT] actually works correctly (I have no
  layer-1 streams to test it), but at least it compiles without error, and
  it appears to be correct (looking at the rest of the source).  Heaven knows
  why gcc doesn't issue an error, and what it actually makes from it.
- mpg123.1
  A simple manual page.  Type "nroff -man mpg123.1 | more" to view it, or
  install it in /usr/local/man/man1 (this is done automatically by "make
- mpg123.c
  New boolean variable "quiet": If this variable is set to TRUE (command line
  option -q), no messages are printed.