Version 1.0.0 (31 July 2009)

  * packaging changes:
    - autoconf bugfix: the PKG_CONFIG variable wasn't initialized properly,
      causing e.g. the libgig test to fail when
      "./configure --disable-jack-driver" was used
      (patch by Alexis Ballier)
    - fixed compilation with gcc 4.3
    - fixes for building on OS X (thanks to Ebrahim Mayat for testing)
    - fixed configure so it detects x86_64 (#107)
    - fixes for building with newer MinGW versions
    - fix for building with bison 2.4 (#111)
    - fixed building with libgig installed in a non-standard directory
    - minor fix in configure for mmsystem.h detection on MinGW
    - Windows: look for editor plugins and Fantasia using base
      directory of liblinuxsampler dll (look in the same directory and one
      directory above)
    - configure script fix: removed unconditional use of SSE
    - fixed building with sqlite installed in a non-standard directory
    - when cross-compiling, don't try to create instruments.db
    - fix for new mingw-w64 version, which has usleep

  * general changes:
    - bugfix: on some POSIX systems instrument editor plugins refused to
      load as we used a non-portable Linux specific struct field
      (fixes bug #70, patch by Ronald Baljeu)
    - fixed endless loop which occured when loading an instrument editor
      plugin DLL on Windows systems failed
    - fixed memory leaks that occurred when liblinuxsampler was unloaded
    - optimized the SynchronizedConfig class so it doesn't wait
      unnecessarily long after an update
    - added support for notifying instrument editors on note-on / note-off
      events (e.g. to highlight the pressed keys on the virtual keyboard
      of gigedit)
    - added support for triggering notes by instrument editors (see above)
    - be verbose on DLL load errors (on Linux)
    - fixed di-harmonic triangle LFO implementation (this LFO implementation
      is only used by very few systems, most pick the int math
      implementation, which is usually faster)
    - fixes for audio drivers with varying buffer sizes
    - experimental support for running LinuxSampler as a DSSI, LV2 and
      VST plugin
    - notification events for stream/voice count statistics are now sent
      only when there are actual changes
    - added memory ordering constraints to improve stability on
      multi-core and multi-cpu systems
    - maximum voices and disk streams can now be altered at runtime
    - fixed CPU feature detection on x86_64 (maybe fixes #108)
    - automatic stacktrace mechanism is now turned off by default and can
      be switched on by command line option "--stacktrace" (the automatic
      stacktrace mechanism seems to be broken on most systems at the moment)
    - C++ API method InstrumentManager::LaunchInstrumentEditor() now returns
      a pointer to the launched InstrumentEditor object
    - added optional 3rd party user data parameter for following
      liblinuxsampler C++ API methods: InstrumentEditor::Main(),
      InstrumentEditor::Launch(),
      InstrumentManager::LaunchInstrumentEditor()
    - theoretical fix: made SynchronizedConfig follow C++0x memory
      model more strictly
    - fixes for using large audio device buffers
    - Windows: add the installation directory to the DLL search path
      when loading an editor plugin (solves problems with VST and
      gigedit on systems with other GTK versions installed)

  * audio driver:
    - removed the nonsense audio channel constraint (which was hard coded to
      max. 100 audio channels) for most audio drivers
    - JACK audio driver did not offer a device parameter "SAMPLERATE" as
      opposed to the LSCP specs
    - bugfix: the SAMPLERATE parameter of some drivers (e.g. JACK)
      reflected the wrong value
    - fixed a memory management bug in ASIO driver
    - Makefile fix: JACK_CFLAGS wasn't used
    - JACK: use jack_client_open instead of the deprecated
      jack_client_new
    - added (experimental) CoreAudio driver
    - applied old fixes to the ASIO driver that were included in the
      previous binary release but accidentally never committed to CVS
      (fixes #117)
    - fixes for ASIO on mingw-w64 (iasio wrapper is not needed on
      win64)
    - VST: added support for sample rate and buffer size changes
    - VST: close editor (Fantasia) when the VST is removed
    - VST: avoid opening Fantasia more than once for each VST instance
    - VST: export main function as "main" on Linux too (fix for
      energyXT)
    - VST: changed number of output channels from one stereo to 16
      stereo
    - added channel routing, fxsends and midi maps to the settings
      stored in the plugin state
    - performance optimization of AudioChannel::MixTo() and
      AudioChannel::CopyTo() methods using GCC vector exensions
      (if available)
    - ASIO fixes: avoid initializing the device twice, avoid throwing
      exception when getting parameters from a disconnected device

  * MIDI driver:
    - added JACK MIDI driver
    - dispatch bank select as ordinary CC as well, the user might seriously
      want to (mis)use it for some purpose ("fixed" in all current MIDI
      input drivers: ALSA, CoreMIDI, JACK, MidiShare, MME)
    - bugfix: pitch bend wasn't working with JackMidi, VST, LV2, MME,
      CoreMidi or AU
    - fixed mingw-w64 compilation error in MME driver
    - made program change handling in MIDI thread real-time safe by
      moving the logic to a non-RT thread
    - fixed minor memory leak in ALSA driver

  * instruments database:
    - avoid time consuming samples scanning when adding instruments
      to the instruments database
    - added support for handling lost files in the instruments database
    - Implemented option for adding instruments in separate directories
      in the instruments database
      (patch by Chris Cherrett & Andrew Williams, a bit adjusted)
    - work-around for missing fnmatch function on Windows to make
      instrument database compilable
    - added instrument database support on Windows
      works with both standalone mode and VST plugin.
      the instruments DB file is located in
      %USERPROFILE%\.linuxsampler\instruments.db which allows different
      databases for each windows user
      if no DB is present it automatically creates the .linuxsampler subdir
      and then creates an empty DB
    - fixed recursive import, which was broken on Windows

  * Gigasampler format engine:
    - fixed a memory leak that could happen when a channel was deleted
      while notes were playing
    - made it possible to load an instrument even if the audio thread
      isn't running
    - added partial support for the "Controller Triggered" MIDI rule,
      enough for piano gigs with pedal noise samples
    - minor fix: only mark FX sends as being modified if really the
      respective FX send MIDI controller was used
    - added support for GM global device volume SysEx message
      (can be selected at compile time to either apply globally to the
      whole sampler [default] or only to the sampler channels that are
      connected to the respective MIDI input port the SysEx message arrived
      on)
    - bugfix: notes triggered at position 0 in the audio buffer were
      sometimes wrongly killed in the same buffer, causing no sound to
      be played
    - added support for chromatic / drumkit mode Roland GS Sysex message
      which is usally used to switch a part between chromatic sounds and
      drumkit sounds (as opposed to their common default setting of MIDI
      part 10 being a drumkit part and all other ones chromatic parts), in
      LS however one can switch between the first 16 MIDI instrument maps
      defined for the sampler
    - made it possible to create multiple sample channels even if the
      audio thread isn't running
    - fixed crash when removing channel with active voices (#116)
    - bugfix: on sample reference changes (instrument editor), only
      un-cache the respective sample if it's really not used by any
      sampler engine anymore
    - re-cache samples in case they were changed by an instrument editor,
      e.g. when a sample was added while playing (#82)
    - fixed hanging notes which occured when note-off event had the exact
      same time stamp as the voice's note-on event and occured both in the
      same audio fragment cycle (fixes bug #112)
    - added support for the "fine tune" and "pitch bend range"
      instrument-level gig parameters
    - fixed minor artifacts in pitch bend handling
    - added support for GS Reset SysEx message
    - allow gig files to use unlimited downward pitch shifting
    - added a limit check for upward pitch shifting
    - bugfix: sometimes, when playing a note twice fast, the second
      note was silent
    - fixed crash happening when a pitch bend event arrived at the
      same time a new instrument was loading

  * LSCP server:
    - added new LSCP event "CHANNEL_MIDI" which can be used by frontends to
      react on MIDI data arriving on certain sampler channels (so far only
      Note-On and Note-Off events are sent via this LSCP event)
    - added new LSCP event "DEVICE_MIDI" which can be used by frontends to
      react on MIDI data arriving on certain MIDI input devices (so far only
      Note-On and Note-Off events are sent via this LSCP event)
    - added new LSCP commands: FIND LOST DB_INSTRUMENT_FILES and
      SET DB_INSTRUMENT FILE_PATH
    - added new LSCP command: SEND CHANNEL MIDI_DATA which can be used by
      frontends to send MIDI messages to specific sampler channel
    - added two additional fields to GET FILE INSTRUMENT INFO command -
      KEY_BINDINGS and KEYSWITCH_BINDINGS
    - bugfix: the bank number provided by MIDI_INSTRUMENT_INFO notifications
      was incorrect
    - Optimized the retrieval of the MIDI instrument mappings
    - added new LSCP command "SET VOICES" to globally alter the maximum
      amount of voices, added new LSCP event "GLOBAL_INFO:VOICES" which will
      be triggered respectively
    - added new LSCP command "SET STREAMS" to globally alter the maximum
      amount of disk streams, added new LSCP event "GLOBAL_INFO:STREAMS"
      which will be triggered respectively
    - bugfix: retry if "select" returns EINTR (this fixes a crash when
      a gigedit file dialog is opened)
    - close all connections when LSCPServer is deleted
    - hard close of all sockets on exit
    - bugfix: SET CHANNEL MIDI_INPUT_TYPE didn't work with the MME
      driver

  * Bug fixes:
    - fixed a crash which occurs when removing a sampler channel waiting
      to start instrument loading after another channel
    - fixed a crash which occurs when removing a sampler channel with
      instrument loading in progress (bug #113)
    - fixed termination caused by uncaught exception when adding MIDI
      instrument with PERSISTENT load mode
    - fixed possible iterator invalidations when resetting the sampler
    - fixed memory leaks when issuing the following LSCP commands:
      GET AUDIO_OUTPUT_DRIVER INFO
      GET AUDIO_OUTPUT_DRIVER_PARAMETER INFO
      GET MIDI_INPUT_DRIVER INFO
      GET MIDI_INPUT_DRIVER_PARAMETER INFO
    - fixed possible compilation error when sqlite is not present
    - fixed orphaned pointers when setting maximum voices limit (bug #118)
    - fixed crash when changing the audio output device of a sampler
      channel with loaded instrument and start playing notes
    - fixed endless loop in Engine::SuspendAll() (bug #120)
    - fixed a low-level atomic load function that was broken on 64-bit
      PowerPC, which probably could cause crashes on that platform
    - fixed a memory management error which could cause a crash when a
      plugin was unloaded
    - bugfix: two private structs had the same name, which could cause
      problems if the linker chose the wrong constructor
    - fixed low-level ConditionServer usage bug that caused lockups on
      Windows


Version 0.5.1 (6 December 2007)

  * packaging changes:
    - added autoconf checks for pthread library
    - added autoconf check for pthread bug found on certain NPTL-enabled
      glibc versions (see Gentoo bug report #194076)
    - added autoconf checks for MS Windows

  * general changes:
    - bugfix: the thread used by an editor plugin didn't die when the
      editor closed
    - bugfix: Ringbuffer.h: fill_write_space_with_null() did not zero
      out all the space. operator--() did not apply size_mask after
      decrementing the read_ptr. DEFAULT_WRAP_ELEMENTS set to 0 in
      order to avoid problems with the _NonVolatileReader functions.
    - bugfix: Stream.h: added a missing divide by BytesPerSample in
      GetWriteSpace(). Since this function is currently only used in
      the stream's qsort() compare function, it didn't trigger any
      bugs.
    - Resampler.h, Synthesizer.h: cubic interpolation now works in
      24bit mode too. Faster method to read 24bit words on little
      endian machines (x86): replaced 3 byte reads + shifts with a 1
      unaligned 32bit read and shift
    - experimental support for MS Windows (MIDI input via MME, AUDIO
      output via ASIO)
    - made handling of SIGINT signal (Ctrl-C) a bit more robust
    - support for monitoring the total number of active disk streams
      (new LSCP commands: GET TOTAL_STREAM_COUNT,
      SUBSCRIBE TOTAL_STREAM_COUNT, UNSUBSCRIBE TOTAL_STREAM_COUNT)

  * AUDIO driver:
    - added Windows ASIO low latency audio driver

  * MIDI driver:
    - added MME Windows MIDI driver

  * LSCP server:
    - added support for Windows style path / filenames, however with
      forward slash path separators instead of backslash
      (i.e. "C:/foo/bar.gig")
    - allow naughty liblscp to send non-string device parameters within
      apostrophes as well
    - added new LSCP commands: "GET FILE INSTRUMENTS <file>",
      "LIST FILE INSTRUMENTS <file>" and
      "GET FILE INSTRUMENT INFO <file> <index>" for retrieving informations
      about an arbitrary instrument file on the system where the sampler is
      running on

Version 0.5.0 (15 October 2007)

  * packaging changes:
    - config.h is not going to be installed along with liblinuxsampler's
      API header files anymore
    - only the API relevant header (and source) files will be exported to
      the Doxygen API documentation (explicitly listed in Doxyfile.in)
    - added completely new XCode project files for Mac OSX which is now
      capable to execute our autoconf environment, thus no need anymore
      to maintain the compile time configuration file (osx/version.h) for
      OSX manually (patch by Toshi Nagata)
    - fixed buggy boolean --enable-foo / --disable-foo configure script
      parameters
    - global.h now only covers global definitions that are needed for the
      C++ API header files, all implementation internal global definitions
      are now in global_private.h
    - atomic.h is not exposed to the C++ API anymore
    - no need to include config.h anymore for using LS's API header files
    - fixed warnings in API doc generation

  * general changes:
    - replaced the old, confusing MIDI program change mechanism by a
      flexible MIDI instrument mapper which allows to map arbitrary
      (MIDI bank, MIDI program) pairs with arbitrary
      (engine type, instrument file, file index) triplets which will be
      loaded on the respective channel when such MIDI program change
      messages arrive, beside that, each entry allows to define a life-time
      strategy for the instrument, so the user can define whether the
      instrument should i.e. be permanently loaded (to be able to switch
      quickly among sounds) or i.e. loaded just on demand when the
      respective program change arrives (to safe RAM space), as well as a
      global volume factor for each entry, so the user can adjust the volume
      dynamic of his mapped instrument collection without having to modify
      the instrument files, also one can manage arbitrary amount of such
      MIDI instrument maps and assign each sampler channel individually
      a certain map, so that i.e. separation of normal instruments and
      drumkits is possible
    - new notification events for tracking changes to audio/MIDI devices,
      MIDI instrument maps, MIDI instruments, FX sends, global volume.
    - sampler was limited to load max. 200 instruments in the background
      due to a constant size RingBuffer FIFO which is now replaced by a
      dynamic (unlimited) size std::list FIFO
    - added FX sends, these allow to route audio signals to arbitrary audio
      output channels for being processed by external effect processors
      (i.e. jack-rack), the send levels are controllable via arbitrary MIDI
      controllers
    - global (sampler wide) volume can now be controlled at runtime
    - Implemented new, improved notification system
    - fixed compilation errors regarding OSX
      (patch by Toshi Nagata)
    - implemented instruments database
    - added support for escape sequences to the instruments database
    - added highly experimental support for on-the-fly instrument editing
      within the sampler's process (by using instrument editor plugins),
      you'll notice the new "Registered instrument editors:" message on
      startup, added a new LSCP command:
      "EDIT CHANNEL INSTRUMENT <sampler-channel>"
      to spawn a matching instrument editor for the instrument on the
      given sampler channel, the plugin path can be overridden at compile
      time with ./configure --enable-plugin-dir=/some/dir
    - added experimental code for synchronizing instrument editors hosted
      in the sampler's process to safely edit instruments while playing
      without a crash (hopefully) by either suspending single regions
      wherever possible or if unavoidable whole engine(s)
    - fixed several issues in fundamental "Thread" class: set scheduling
      policy and priority on thread level, set a minimum stack size for
      thread (TODO: a reasonable value yet to be tested), bugfix: non-RT
      threads simply inherited properties of starting thread instead of
      setting their own policy and priority
    - minor fix in our automatic stack trace mechanism on crashes, the
      main process did not wait for the stack trace process to finish
      its output
    - fixed some minor memory leaks
    - reenabled assembly features support, at the moment only for
      enabling a fast denormal FPU mode (x86 platforms supporting SSE2)
    - minor assembly fix in x86 features detection (don't use the PIC
      register, to avoid relocations in the text segment at runtime)
    - POSIX callback functions of Thread.h are hidden

  * MIDI driver:
    - dispatch bank select (MSB and LSB) messages

  * audio driver:
    - the ALSA audio output driver parameters now reflect the correct
      parameter value ranges for the respective selected sound card
      (patch by Till Wimmer, a bit fixed and extended)

  * Gigasampler format engine:
    - if a filter is used and EG2 finishes before EG1, let the voice
      die when EG2 has finished (this fixes a problem with clicks and
      voice starvation for some gigs)
    - playback is no longer disabled during instrument loading
    - all notes playing on a channel that changes its instrument keep
      playing with the old instrument until they get a note off
      command
    - EG fix: a release value of zero could cause noises or crash
    - handle MIDI coarse tuning messages (MIDI RPN #0 MSB #2 LSB)
    - EG fine tuning: when attack is zero the EG starts at a level
      above max sustain level, which means that there is a decay phase
      even if sustain is 100%
    - more EG fixes: the level could sometimes go below zero and cause
      noises or crashes
    - minor fix of EGDecay (patch by Toshi Nagata)
    - fixed compiler error when --enable-override-filter-type was
      supplied to the configure script (fixes #46)
    - disk thread: queue sizes are now proportional to CONFIG_MAX_STREAMS
      instead of a fix value
    - behavior fix: on MIDI CC# 65 (portamento on / off), 126 (mono mode),
      127 (solo mode) only kill voices if the respective mode really
      changed

  * LSCP server:
    - fixed compile time error for old Bison versions
      (i.e. v1.28 found on MacOS 10.4, patch by Toshi Nagata)
    - parser now supports extended ASCII character set
      (up to ASCII code 255, i.e. includes now umlauts and accents)
    - filename arguments in LSCP commands now allow to use escape
      sequences, that is directly literal as one of: \', \", \\, \n, \r,
      \f, \t, \v, or as octal ASCII code value like \132, or as hex ASCII
      code value like \xf2) (fixes bug #24)
    - the following LSCP commands now also support escape sequences for at
      least one of their text-based parameters (i.e. name, description):
      "ADD MIDI_INSTRUMENT_MAP", "MAP MIDI_INSTRUMENT",
      "SET MIDI_INSTRUMENT_MAP NAME", "SET FX_SEND NAME", "CREATE FX_SEND",
      "SET DB_INSTRUMENT_DIRECTORY NAME",
      "SET DB_INSTRUMENT_DIRECTORY DESCRIPTION", "SET DB_INSTRUMENT NAME",
      "SET DB_INSTRUMENT DESCRIPTION", "FIND DB_INSTRUMENTS",
      "FIND DB_INSTRUMENT_DIRECTORIES"
    - returns verbose syntax errors (line and column where syntax error
      occured, the unexpected character and the actually expected, possible
      character(s), the latter only if less than 5 possibilities)
    - made sure that LSCP syntax is not affected by gigedit locale
      settings
    - bugfix regarding strings parameter lists: all comma separated lists
      of strings were treated as being one string containing commas
      (fixes #57)

  * Bug fixes:
    - fixed crash occurring on certain LSCP scripts (Bug 39)
    - another thread safety fix for lscp "load engine" and "set
      channel audio output device"
    - fixed a crash which occurs when reassigning the same engine
      on a sampler channel with connected MIDI device
    - fixed a crash which occurs when changing the number of ports of a MIDI
      device connected to a sampler channel to number less then or equal
      to the index of the port to which the sampler channel is connected.
    - The previous bindings were not been disconnected when altering
      the ALSA_SEQ_BINDINGS parameter. Introduced a NONE keyword for
      unsubscribing from all bindings (e.g. ALSA_SEQ_BINDINGS=NONE).
    - The active stream/voice count statistic was incorrect.
    - notification events were not been sent for some sampler
      channel changes
    - added default min and max values to restrict the number of allowed
      audio output channels and MIDI input ports
    - the connection to the PCM interface is now closed when destroying
      an audio output device
    - files with slash in their path or filename could not be loaded

  * test cases:
    - updated and fixed (haven't been touched in a while)

Version 0.4.0 (24 November 2006)

  * packaging changes:
    - changed deprecated copyright attribute to license; added ldconfig
      to post-(un)install steps; added devel package for liblinuxsampler;
      to linuxsampler.spec (RPM)
    - install necessary development header files for allowing 3rd party
      applications to link against liblinuxsampler
    - liblinuxsampler's API documentation can be generated with 'make docs'
      (Doxygen required)
    - added benchmark to automatically detect the best triangle LFO
      implementation (currently either an integer math solution or a
      di-harmonic approximation), automatic detection can be overriden
      with --enable-signed-triang-algo=x and --enable-unsigned-triang-algo=x
      configure script argument though (mandatory for cross-compilation)
    - do not automatically pick optimized gcc flags if the user already
      provided some on his own (as CXXFLAGS)
    - added compile time option to disable processing of All-Notes-Off MIDI
      messages
    - added compile time options to allow disabling the various audio and
      MIDI drivers
    - fixed automatic GCC CPU switch detection on PPC
      (patch by Ebrahim Mayat)

  * Gigasampler format engine:
    - extensive synthesis optimization
      (reimplementation of EGs and LFO(s), removed synthesis parameter
      prerendering and the synthesis parameter matrix in general, splitting
      each audio fragment into subfragments now where each subfragment uses
      constant pitch and filter coefficients. The volume coefficient is
      linearly interpolated inside a subfragment, unless
      --disable-interpolate-volume is set.)
    - fine tuning of the EG modulation parameters
    - improved filter cutoff calculation by adding support for the
      following gig parameters: Cutoff freq (used when no cutoff
      controller is defined), Control invert, Minimum cutoff, Velocity
      curve and Velocity range. The keyboard tracking now scales
      cutoff frequency instead of resonance.
    - added support for gig parameter Resonance.
    - fixed bug in sysex handling (patch by Juan Linietsky)
    - added global volume attenuation of -9 dB (0.35f) to prevent clipping
      which can be overridden with --enable-global-attenuation
    - EG fixes: made the length of "attack hold" stage more
      accurate. Release stage can now start before attack stage
      ends. Cancel release didn't work when sustain was zero. Attack
      time now has a minimal value to prevent clicks.
    - fixed pitch changes larger than one octave
    - fixed EG3 (pitch envelope) synthesis which was neutral all the time
    - implemented portamento mode and solo mode (a.k.a 'mono mode'):
      all modes can be altered via standard GM messages, that is CC5 for
      altering portamento time, CC65 for enabling / disabling portamento
      mode, CC126 for enabling solo mode and CC127 for disabling solo mode
    - fine tuning of the curves for volume (CC7), pan (CC10 and gig
      parameter) and crossfade
    - added support for the "attenuation controller threshold" gig
      parameter
    - added smoothing of volume changes caused by control change
      messages
    - sample loop parameters are now taken from the DimensionRegion
      instead of the wave chunk
    - fixed keyswitching for v3 gigs with a number of keyswitch splits
      not equal to a power of two
    - reimplementation of the filter algorithm. The new filter is more
      accurate and supports all gig filter types, including bandreject
      and lowpass turbo.
    - real support for 24 bit samples - samples are not truncated to
      16 bits anymore
    - support for aftertouch (channel pressure, not polyphonic
      aftertouch)
    - LFO1 behaviour fixed (dampening from max volume instead of
      amplifying from 0)

  * LSCP server:
    - fixed application exit on broken pipe error (fixes bug #20)
    - fixed the notification messages delay due to lack of
      network activity (fixes bug #26)
    - fixed parser bug which occured on space(s) within device parameters
      (e.g. "SET AUDIO_OUTPUT_CHANNEL_PARAMETER 0 0 SOMEPARAM='foo bar'")

  * audio driver:
    - added aRts audio output driver (by no means RT safe)

  * MIDI driver:
    - fixed legacy sysex code which caused dispatching of MIDI SysEx
      messages several times instead of once
    - API extension for MIDI drivers which already supply exact time stamps
      for events (i.e. for offline rendering based MIDI drivers)
    - added checks for bad MIDI data

  * linuxsampler application:
    - show available sampler engine types on startup

  * general changes:
    - support for muting sampler channels and solo mode of the same, two new
      LSCP commands ("SET CHANNEL MUTE" and "SET CHANNEL SOLO") and two new
      fields ("MUTE" and "SOLO") for command "GET CHANNEL INFO" were
      introduced for this, the behavior is the same like on a mixer console
      (patch by Grigor Iliev, a bit adjusted). Also added configure option
      --enable-process-muted-channels which can be used to enable the
      processing of muted channels.
    - support for sostenuto pedal
    - support for monitoring the total number of active voices
      (new LSCP commands: GET TOTAL_VOICE_COUNT, GET TOTAL_VOICE_COUNT_MAX,
      SUBSCRIBE TOTAL_VOICE_COUNT, UNSUBSCRIBE TOTAL_VOICE_COUNT)
    - fixed some memory management errors
    - fixed some concurrency problems that could lead to crashes when
      LSCP commands were executed
    - fixed crash when instrument loading failed previously on the same
      sampler channel (fixes bug #36)

Version 0.3.3 (15 July 2005)

  * packaging changes:
    - fixed compilation with gcc 4.0

  * Gigasampler format engine:
    - LFO preprocessing was performed even though the respective LFO was
      disabled by the instrument patch (minor efficiency fix)
    - if period time of chosen audio device is too small (< MIN_RELEASE_TIME)
      for volume ramp downs in the same fragment (needed for current voice
      stealing implementation) then simply inform the user with a warning
      about possible click sounds and reduce the volume ramp down
      appropriately instead of cancelling the audio device connection

Version 0.3.2 (24 June 2005)

  * packaging changes:
    - updated autotools build files to compile on OS X
      (conditional compilation of CoreMIDI and MidiShare drivers)
    - hand-crafted assembly optimization code can be disabled with
      './configure --disable-asm' (definitely not recommended)
    - fixed 'make dist' rule to include all necessary files
    - require automake (>= 1.5) for 'make -f Makefile.cvs'
      (due to 'dist-bzip2' automake option)

  * Gigasampler format engine:
    - support for the gig parameters for "release velocity response" curves
    - fine tuning of the EGADSR envelope
    - volume of release triggered samples now depends on note-on velocity,
      note length and gig parameter "release trigger decay" instead of
      note-off velocity.
    - revised voice stealing
      (fixes crash and endless loop caused by voice stealing)
    - don't reset scale tuning on instrument or audio output device change
    - handle key group conflicts right at the beginning of each fragment
      instead of when the respective voice is actually launched
      (fixes undefined behavior if stolen voices belonged to a key group -
      this case was followed by a "killed voice survived" error message)
    - fixed minor issue with null/silence samples
      (those stole voices even though they don't need a voice at all which
      resulted in "voice stealing didn't work out" messages)
    - don't reset volume, pan, pitch and MIDI controller values on
      instrument or audio output device change

  * LSCP server:
    - fixed some crashes (patch by Grigor Iliev, fixes #19)
    - fixed LSCP event "CHANNEL_INFO" notification
      (e.g. did not notify on volume changes or MIDI program change events)

  * linuxsampler application:
    - added command line parameters --lscp-addr and --lscp-port to override
      default IP address and port of LSCP server

Version 0.3.1 (24 May 2005)

  * initial release
