/* CHANGELOG FOR PORTMIDI
 *
 * 17Jan07 Roger Dannenberg
 * - Lots more help for Common Lisp user in pm_cl
 * - Minor fix to eliminate a compiler warning
 * - Went back to single library in OS X for both portmidi and porttime
 *
 * 16Jan07 Roger Dannenberg
 * - OOPS! fixed bug where short messages all had zero data
 * - Makefile.osx static library build now makes universal (i386 + ppc)
 *    binaries
 *
 * 15Jan07 Roger Dannenberg
 *  - multiple rewrites of sysex handling code to take care of 
 *     error-handling, embedded messages, message filtering,
 *     driver bugs, and host limitations.
 *  - fixed windows to use dwBufferLength rather than 
 *     dwBytesRecorded for long buffer output (fix by Nigel Brown)
 *  - Win32 MME code always appends an extra zero to long buffer
 *     output to work around a problem with earlier versions of Midi Yoke
 *  - Added mm, a command line Midi Monitor to pm_test suite
 *  - Revised copyright notice to match PortAudio/MIT license (requests 
 *     are moved out of the license proper and into a separate paragraph)
 *
 * 18Oct06 Roger Dannenberg
 *  - replace FIFO in pmutil with Light Pipe-based multiprocessor-safe alg.
 *  - replace FIFO in portmidi.c with PmQueue from pmutil
 *
 * 07Oct06 cpr & Roger Dannenberg
 *  - overhaul of CoreMIDI input to handle running status and multiple
 *  -    messages per packet, with additional error detection
 *  - added Leigh Smith and Rick Taube support for Common Lisp and
 *  -    dynamic link libraries in OSX
 *  - initialize static global seq = NULL in pmlinuxalsa.c
 *
 * 05Sep06 Sebastien Frippiat
 *  - check if (ALSA) seq exists before closing it in pm_linuxalsa_term()
 *
 * 05Sep06 Andreas Micheler and Cecilio
 *  - fixed memory leak by freeing someo objects in pm_winmm_term()
 *  - and another leak by freeing descriptors in Pm_Terminate()
 *
 * 23Aug06 RBD
 *  - various minor fixes
 *
 * 04Nov05 Olivier Tristan
 *  - changes to OS X to properly retrieve real device name on CoreMidi
 *
 * 19Jul05 Roger Dannenberg
 *  - included pmBufferMaxSize in Pm_GetErrorText()
 *
 * 23Mar05 Torgier Strand Henriksen
 *  - cleaner termination of porttime thread under Linux
 *
 * 15Nov04 Ben Allison
 *  - sysex output now uses one buffer/message and reallocates buffer
 *  -    if needed
 *  - filters expanded for many message types and channels
 *  - detailed changes are as follows:
 *  ------------- in pmwinmm.c --------------
 *  - new #define symbol: OUTPUT_BYTES_PER_BUFFER
 *  - change SYSEX_BYTES_PER_BUFFER to 1024
 *  - added MIDIHDR_BUFFER_LENGTH(x) to correctly count midihdr buffer length
 *  - change MIDIHDR_SIZE(x) to (MIDIHDR_BUFFER_LENGTH(x) + sizeof(MIDIHDR))
 *  - change allocate_buffer to use new MIDIHDR_BUFFER_LENGTH macro
 *  - new macros for MIDIHDR_SYSEX_SIZE and MIDIHDR_SYSEX_BUFFER_LENGTH 
 *  -    similar to above, but counts appropriately for sysex messages
 *  - added the following members to midiwinmm_struct for sysex data:
 *  -    LPMIDIHDR *sysex_buffers;   ** pool of buffers for sysex data **
 *  -    int num_sysex_buffers;      ** how many sysex buffers **
 *  -    int next_sysex_buffer;      ** index of next sysexbuffer to send **
 *  -    HANDLE sysex_buffer_signal; ** to wait for free sysex buffer **
 *  - duplicated allocate_buffer, alocate_buffers and get_free_output_buffer 
 *  -    into equivalent sysex_buffer form
 *  - changed winmm_in_open to initialize new midiwinmm_struct members and 
 *  -    to use the new allocate_sysex_buffer() function instead of 
 *  -    allocate_buffer()
 *  - changed winmm_out_open to initialize new members, create sysex buffer 
 *  -    signal, and allocate 2 sysex buffers
 *  - changed winmm_out_delete to free sysex buffers and shut down the sysex
 *  -    buffer signal
 *  - create new function resize_sysex_buffer which resizes m->hdr to the 
 *  -    passed size, and corrects the midiwinmm_struct accordingly.
 *  - changed winmm_write_byte to use new resize_sysex_buffer function,
 *  -    if resize fails, write current buffer to output and continue
 *  - changed winmm_out_callback to use buffer_signal or sysex_buffer_signal
 *  -    depending on which buffer was finished
 *  ------------- in portmidi.h --------------
 *  - added pmBufferMaxSize to PmError to indicate that the buffer would be
 *  -    too large for the underlying API
 *  - added additional filters
 *  - added prototype, documentation, and helper macro for Pm_SetChannelMask
 *  ------------- in portmidi.c --------------
 *  - added pm_status_filtered() and pm_realtime_filtered() functions to
 *       separate filtering logic from buffer logic in pm_read_short
 *  - added Pm_SetChannelMask function
 *  - added pm_channel_filtered() function
 *  ------------- in pminternal.h --------------
 *  - added member to PortMidiStream for channel mask
 *
 * 25May04 RBD
 *  - removed support for MIDI THRU
 *  - moved filtering from Pm_Read to pm_enqueue to avoid buffer ovfl
 *  - extensive work on Mac OS X port, especially sysex and error handling
 *
 * 18May04 RBD
 *  - removed side-effects from assert() calls. Now you can disable assert().
 *  - no longer check pm_hosterror everywhere, fixing a bug where an open
 *    failure could cause a write not to work on a previously opened port
 *    until you call Pm_GetHostErrorText().
 * 16May04 RBD and Chris Roberts
 *  - Some documentation wordsmithing in portmidi.h
 *  - Dynamically allocate port descriptor structures
 *  - Fixed parameter error in midiInPrepareBuffer and midiInAddBuffer.
 *
 * 09Oct03 RBD
 *  - Changed Thru handling. Now the client does all the work and the client
 *    must poll or read to keep thru messages flowing.
 *
 * 31May03 RBD
 *  - Fixed various bugs.
 *  - Added linux ALSA support with help from Clemens Ladisch
 *  - Added Mac OS X support, implemented by Jon Parise, updated and 
 *       integrated by Andrew Zeldis and Zico Kolter
 *  - Added latency program to build histogram of system latency using PortTime.
 *
 * 30Jun02 RBD Extensive rewrite of sysex handling. It works now.
 *             Extensive reworking of error reporting and error text -- no
 *             longer use dictionary call to delete data; instead, Pm_Open
 *             and Pm_Close clean up before returning an error code, and 
 *             error text is saved in a system-independent location.
 *             Wrote sysex.c to test sysex message handling.
 *
 * 15Jun02 BCT changes:
 *  - Added pmHostError text handling.
 *  - For robustness, check PortMidi stream args not NULL.
 *  - Re-C-ANSI-fied code (changed many C++ comments to C style)
 *  - Reorganized code in pmwinmm according to input/output functionality (made
 *    cleanup handling easier to reason about)
 *  - Fixed Pm_Write calls (portmidi.h says these should not return length but Pm_Error)
 *  - Cleaned up memory handling (now system specific data deleted via dictionary
 *    call in PortMidi, allows client to query host errors).
 *  - Added explicit asserts to verify various aspects of pmwinmm implementation behaves as
 *    logic implies it should. Specifically: verified callback routines not reentrant and
 *    all verified status for all unchecked Win32 MMedia API calls perform successfully
 *  - Moved portmidi initialization and clean-up routines into DLL to fix Win32 MMedia API 
 *    bug (i.e. if devices not explicitly closed, must reboot to debug application further).
 *    With this change, clients no longer need explicitly call Pm_Initialize, Pm_Terminate, or
 *    explicitly Pm_Close open devices when using WinMM version of PortMidi.
 *
 * 23Jan02 RBD Fixed bug in pmwinmm.c thru handling
 *
 * 21Jan02 RBD Added tests in Pm_OpenInput() and Pm_OpenOutput() to prevent 
 *               opening an input as output and vice versa.
 *             Added comments and documentation.
 *             Implemented Pm_Terminate().
 *
 */