===============================================================================
 TSE3 History File
===============================================================================

This history file documents the entire development of the TSE3 library. It
includes every painstaking detail and so is ridiculously boring to read. It's
only really here for my own benefit, and I haven't actually decided what that
is yet.

-------------------------------------------------------------------------------
 Contents
-------------------------------------------------------------------------------

  - Daily development history
  - TODO list

-------------------------------------------------------------------------------
 Daily development history
-------------------------------------------------------------------------------

-------------------------------------------------------------------------------
May 1999

14 Decided this was a good way to learn decent C++ and the STL, so embarked
   on a rewrite of the proven TSE v2.00. I'll build this under Unix, and
   document it well. ATM I'm going to use KDOC because it looks enough like
   javadoc, and I can get it to work.
17 Made a ridiculous build structure somewhat like the the Java one, with
   separate build and src trees to try to keep things neat.
20 Put in the Notifier/Listener framework taking ideas from Java. The funny
   thing is that it flipping well works!
24 Over the weekend extended how I think playback will work. Quite
   different from TSE. I now have a Playable interface that everyhing
   implements: from MidiParams right up to Song. The Transport class just
   reads MIDI events from Song and schedules them. The implementation of
   this means that each object can cache it's next event which saves
   reevaluating them as much as in TSE2 (I think). Made this even neater
   by adding meta events to the MidiEvent class so that tempo and timesig
   chages can filter through this system too. The only thing that won't
   fit into this framework now is audio.
   Added the Tempo and TimeSigTracks.
   Added NoteNoConversion class and wished that I didn't have to use stdio
   features.
25 Added the Metronome class, and made it return values through the
   Playable interface. I toyed with making the TimeSigTrack responsible
   for producing the metronome, after all it knows what the time sig is,
   but decided against it since it doesn't know what the playing status
   is. Also even if there were no more MIDI events in the Song it would
   still return stuff, so Song could never return false to consume which
   would be a shame: as it stands an auto stop facility should be fairly
   easy...
   First stab at some of the Transport functionality including drafting
   a version of the playback/scheduling loop.
   Started to include the Part. In my quest to make everything as general
   as possible to beat TSE2 for sheer coolness, added the MidiEventFilter
   class, and the SimpleMidiEventFilter.
26 Got the Part class doing the Playable interface properly. It returns
   any MidiParameters first, and then steps through the Phrase. Marvellous.
   Work finishing off the PhraseList. Added the PhraseListener.
   Fixed some rubbishness in my Notifier design.
   Then came accross a big trip up in the notifier design: what if you have
   one class inheriting from a notifing class that want to add it's
   own notification. The use of Notifier get disgusting.
27 Got Tracks and Songs implementing the Playable interface. Sorted!
   Changed the Notifier design *again*. This time removed the templates
   altogether. Now we only have a basic type of Listener and Notifier.
   There is an integer code for the different types of event which means
   that you have to work out what sort of Notifier just send you an
   updated() and then switch on the event code. Not nice, but it scales
   a little better.
28 Added a simple version of the FileOperations class in order to load
   a demo song so I have data to play with.

-------------------------------------------------------------------------------
June

1  Got the Playable interfaces working better for Track and Part.
   Put it into the EventTrack and then Tempo and TimeSig Tracks.
2  Made the PlayableIterators more robust by handling the deleting of
   their Playable objects. I'm not entirely sure how to test that code yet.
3  Element-14 is sold to Pace. Marvellous. Maybe work will slow down on
   TSE3 now ;-)
7  Got the saving mechanism working using a Saveable interface. Created
   a new 'TSE3MDL' which is a lot more verbose than TSEMDL, but who cares?
8  Implemented loading of TSE3MDL files.
9  Added TSE3MDL file decsripton document. It doesn't really tell you all
   that much, but it kept me occupied.
   Added Song::replacePhrase. Everything that was in SongUtilities in TSE2
   belongs in appropriate classes in TSE3.
   If MidiParams is going to implement forceNone, then it has to be a
   MidiEventFilter.
10 Oo look, there's an io manipulator called ws. Nice. There's also a
   strstream class. I can get rid of <stdio.h> stuff. Better.
   Had a first stab at the PhraseEdit::tidy method. It will probably be a
   while until I can test it.
   Fixed a core dumping bug where the MetronomeIterator couldn't cope
   properly with the Metronome being deleted.
   Started thinking about a TSE3 command design pattern implementation.
11 Got the Commands working with the CommandHistory. Added the
   PhraseUtilities class and implemented add/subtract.
21 Put a hook into the MidiScheduler to handle MIDI remote control.
   Added the Panic class drawing the information from TSE2. Started
   implementing the PanicIterator class.
22 Autostop flag in Transport class (it always did autostop, it's just now
   configurable!)
   FileOperations.cpp wasn't using the istrstream class. Does now.
   Added saving and loadng to the Panic class. Implemented everything that
   doesn't require sysex messages. Not sure how to do those yet.
   Added a RepeatTrack to replace the left/right markers. Made the
   SongIterator generate these, and the Transport act on them. Scheduler
   had to implement moveTo, and Transport react to move updates from the
   Scheduler. Fixed various bugs on the way.
25 Spent some time looking for a segmentation fault. Didn't find the cause.
   D'oh! However, a few minor things fixed, TrackIterators detach
   themselves from Tracks etc.
28 Added a status flag to Repeat events. *** UNTESTED ***
30 Added the MidiFileImport class with code taken from TSE2 MidiImport.

-------------------------------------------------------------------------------
July

1  Still investigating that segmentation fault. When a Track is deleted
   a deleted callnack is being sent that is for a deleted object and
   so the program crashes.
2  Found out why the segmentation fault occurs. The MidiParamsIterator
   didn't have a dtr that detached it from the object. So when the
   Track was deleted, and it's MidiParams object a callback to the non
   existent iterator caused a crash. Fixed now.
5  Made the TempOTrack and the TimeSigTrack attach themselves to their
   appropriate EventTrack. Put some notify()s in the EventTrack class.
   Added copy constructors where relevant: some are just inline empty and
   private. The Part has a decent copy constructor that attaches the new
   Part to any used Phrase.

-------------------------------------------------------------------------------
July

19 Added the MidiSchedulerFactory class and converted the meat of
   MidiScheduler to a StreamMidiScheduler.
20 Added a riscos (hypothetical) and linux MidiSchedulers and factories.
   Added a NullMidiScheduler for when things go badly.
   Added a Makefile.common that should do much of the mundane makefile
   stuff.

-------------------------------------------------------------------------------
August

3  Started trying to write a Cakewalk .ins file parser in a moment of
   boredom. I've done it already for Anthem but wanted to do it properly -
   I even found a document that sort of describes them rather than try
   to work it out myself. I'm using a state machine style parser.
...23
   Got the parser working. It loads the whole file and allows you to
   select an Instrument from the hierarchy. Realised this was far too heavy
   weight - we only need to create a few of the many objects created for
   the whole of a file like Yamaha.ins so in my clean up phase I decided
   to work the parser into the Anthem form - only get from the file what
   we need to.
24 Finished off this work. The Instrument parser is now complete - it can
   save .ins files as well as create them.
25 Fixed STL list bug in new Instrument code.
26 Started work on the MidiFileExport class.
   Couldn't play back things trhough Playable interfaces. It's down to an
   error in some load methods that don't test string::compare(..) == 0.
   Twit.
   next discovered bug: the priority_queue in Transport that I've copied
   into MidiFileExport::writeMTrk_outputLoop is the wrong way up.
31 Finished off MidiExport for type 0 files. It works nicely. Implemented
   the type 1 format. Works. Good. Just need to slot in the code that
   actually saves tempo/timesig events and it should be done...

-------------------------------------------------------------------------------
September

1  Finally got the MidiExpoty working well. Not bad. The output seems to
   be correct which is good - it proves that MidiExport works, but also
   that my output iterator system works too. Marvellous!
   Added a compact facility to shorten the file size.
   Removed old <stdio.h> code from MidiImport constructor.
7  Finally I have managed to get the soundcard spitting out MIDI commands!
   This is a big step forward (this is using the AWE32 that Rik has donated
   to the cause.)
14 First time that I've got Linux to spit some MIDI output! I've got
   song.tse3mdl playing (somewhat stutteringly, but it some form of output).
   Hurrah! My big problem is returning the current clock time - I can't see
   a way of polling /dev/sequencer to get it - and I can't see any other
   MIDI programs that do it nicely (kooBase, for example, just seems to
   return the time of the last scheduled note. Yuck).
23 PhraseEdit.insert() is ridiculously slow. Hurray for the STL (and probably
   my dodgy use of it). I've tried to add an insert position cache to the
   class.
24 Tidied up the code so that it compiles quietly under GCC with -Wall.
27 Added the MidiMapper object.
29 Gave TSE3Play the ability to produce snappy VU bar textual output. Cool!
    
-------------------------------------------------------------------------------
October

6  Moved public data members out of Song.
7  Extended FileOperations_TSE2 (which was FileOperations_2) to be able to load
   all of the TSE2MDL filetype. Made the FileRecogniser know about old TSEMDL
   files. Made FileOperations_TSE2 into TSE2FileImport and gave it a
   MidiFileImport-like interface.
   Moved public data members out of Track and Part.

-------------------------------------------------------------------------------
November

9  Moved the build tree into an autoconf/automake system. This will make it
   easier for other people to build and install the libraries. I still haven't
   figured out how to do shared libraries in this system; I don't appear to
   libtool (which appears to be favoured) but do have ranlib. I'll work it out.
   This work involved moving the directories around a bit. Now the commands
   are in TSE3/commands and the different platforms' schedulers are in
   TSE/platform. I think that overall this is more sensible.
10 Fixed the MidiMapper - the defaults weren't being set in the constructor so
   everything was just getting mapped to (0,0).
   Put tempo handling into the Transport meta event handler.
11 Moved the inline stuff from CommandHistory.h into .cpp and forced setLimit
   to shorten the lists if necessary. Also added an 'infinity' choice.
17 Added the application subdirextory of TSE3 with application support classes.
   First ones in are the Application class and the ChoicesManager,
   ChoiceHandler and implementations for Metronome and Application. This is a
   neat way of trying to pull application logic out of the GUI front end.
   Interestingly you can't create a ostream for "~/.anthem", you can for
   "/home/pete/.anthem". I don't like that...
18 ChoiceHandler for Transport class.
19 Moved some of the stuff in the Saveable base class out into helper functions
   so that I could more easily implement the choices saving - the ChoiceHandler
   classes don't beong inherting from Saveable since they don't care about the
   current Song. Put in the framework that allows the choices file chunks to
   be regognised. Just need to actually prod the settings now.
23 When implementing load for the ChoiceHandlers is became aparent that I'm
   continually writing block parsing code. This checks for the closing },
   sometimes checks for sub-blocks (when it always should) and parses data
   lines. Created the FileBlockParser utility class to do the spade work for
   me. This involved a lot of moving things around. Redesigned the layout of
   te FileOperations classes to suit it better. The interface is now a lot
   more logical. There is one entry point for file operations: the TSE3MDL
   class. Nice. I created the Serializable base class to replace the
   Saveable. A little more than just a name change, I'm trying to work out
   how to remove the extra data (e.g. song pointer) that's passed around the
   load calls - when loading choices there isn't a Song reference so it's
   redundant. I'm trying to generalise the Serializeable mechanism as much
   as possible.
   Created lots of nice little helper FileItemParsers that are generic so that
   you don't have to do too much tedious typing.
24 Finished off the new FileOperations system. It works pretty nicely all
   things considered. Or so it would seem. It's only really strange data
   lines that you have to write explicit parsing code for now, and even
   then you get given all the standard stuff (sub-blocks, closing brace).
   In order to make the system better I added a catch-all FileItemParser to
   the FileBlockParser. Next step is to completely remove the Saveable
   interface and make all classes use Serializable. Big job.
26 Finally I beilieve that I have finished adding the new Serializable
   interface to the whole TSE3 Song hierarchy in place of the Saveable.
   Various twiddles to bits such as the ChoicesManager::ChoicesChanceHandler
   and so on, and some code reformatting, plus removing some methods which
   were inline when they really shouldn't be (must have been when I started
   and we too into Java ;-)
27 Added a midi, gm, gs, and xg reset facility to tse3play.
28 Fiddled the MidiEvent structure. I was never really happy using the two
   Event<MidiCommands> because e.event.data.data1 is messier than
   e.data.data1 which is what you'd write now. It was always a 'special'
   kind of Event so now it's an easier to use special type of Event.
   Added --oss and --stream to tse3play.
30 Added a Win32 platform. It's not finished, but I was reading the docs.

-------------------------------------------------------------------------------
December

1  Took out that nasty global stuff from OSS.h and made it a little neater.
   The OSS API really does suck.
4  At SYMS - added a lot of new stuff for soundcard support, AWE/GUS/FM stuff.
   Added a VoicesManager - this code is based on the playmidi and kooBase
   sources since the OSS documentation is sadly lacking. It's a real shame,
   really - the API itself sucks, so working it out from (oddly formatted)
   code and not documentation is a pain.
   Added a start facility to tse3play so you can play from a given position.
13 (On a plane over the US ;-) I've been working on FM support in the
   OSSMidiScheduler (I've mainly been seeing how kooBase does it). The code is
   looking a bit complex now - I'll have to sort that out. The patches aren't
   perfect yet, but I get noise out. Time to sleep...
17 During the Western Carble Show I've rejigged the OSSMidiScheduler class
   heirarchy. I was incorporating all the different types of OSS device into the
   one OSSMidiScheduler class since they have to share the nasty OSS API macros.
   This system used a pointer to member jump table for each OSS device
   available. This design smells to much like virtual function calls for comfort
   the only reason they weren't being the need to share the data OSS buffer.
   My alteration to this is to really make each driver a separate class and to
   share the data using references. This is almost certainly the Right Way to
   do it. It kept me occupied, anyway.

-------------------------------------------------------------------------------
January 2000

17 Whilst waiting for the Pace exec to do anything interesting, played about a
   bit with the command objects. Added a new VariableSetCommand to save all the
   hard work with Song_SetTitle, Song_SetAuthor etc. Added new Song commands for
   all the other things you can do with Songs. Added Song::searchForPhrase to
   make undoing replacePhrase work.
21 Changed the ugly src/TSE3 to tse3.

-------------------------------------------------------------------------------
February

7  Fixes to the FileBlockParser, access to the amount of lookAhead in the
   Transport class and a usleep in tse3play which takes processor usage down
   from ~60% to very little. Harrah.
14 Altered the build process to use libtool and install properly.
.. Working on the website to upload to wherever I find.
22 Added support for the solo track to the SongIterator.
   Added --solo-track to tse3play to test it.
24 Added a MidiEcho object (like TSE2).
25 First stab at reading from /dev/sequencer - success! If only OSS was
   documented... same old story.
28 Tidied up the recording stuff, fixed MidiEcho. Bugfixes here and there.
   Implemented Transport::ff and rew methods.
   Made the OSS scheduler take the OSS timing rate into account in timing
   calculations.
   Added channel and port filters to the SimpleEventFilter, plus a time scaling
   factor: it's anything but 'Simple' now.
29 Implemented the recording mute facility. It's not as nice as in TSE2. In the
   later you said 'record on channel 2'. In TSE3 you have to give the
   SimpleMidiEventFilter for the Track that's being recorded on, so that it can
   be muted. Euchy, but it works.

-------------------------------------------------------------------------------
March

1  Wanted to sort out how to do the OSS immediate playback so that I can do the
   MIDI echo sucessfully. The OSS documentation sucks. As ever. The new PDF
   docs helpfully say (p70-1) there is an ioctl to do it. They don't tell you
   what it is. Poured our the driver source until I worked out what it is and
   how to use it. Say hello to SNDCTL_SEQ_OUTOFBAND.
   There appears to be massive delays between input and output, though. Hmm.
   Ha! Fixed the delay. I hadn't enclosed OSS:readInput in a loop, so it was
   getting called many times for a single event (i.e. read time, read status
   byte, read time, read data 1, read time, read data 2 and return an event).
   It works very nicely now, and OSS tech support has still not bothered to
   reply to me.
   Added --sleep switch to tse3play. Doesn't seem to make the vu bars update
   more responsively, though.
3  Added the -ansi flag to the build. Fixed some minor things. Giving egcs
   the -pedantic flag make it guff up on inline functions in the EventTrack
   template class. I think that it shouldn't. Shame.
8  Started on the KDE2 UI application. Serious problems: qt doesn't seem to
   coexist with the STL at all. Hmm. Also, the KDE libs use the -pedantic
   switch, so the EventTrack really screws up. Grr.
9  Changed the bracket matching to exdented style, since I've been meaning to do
   this for ages. Now my eyes hurt.
   Fixed the nasty -pedantic problem. The introduction of a 'typename' was what
   was needed.
13 The 'Rule of Three' (RO3) needed following properly in TSE3, so added a whole
   tonne of private copy constructors and operator= declarations. Added somei
   more docs to constructors here and there.
15 Fixed a really nasty timing bug in the OSS driver - if you say SEQ_WAIT_TIME
   with the same value as the last one, it actually waits for a bit more, which
   can skew playback royally.
   Added colour to the tse3play application. Cheesy. Also added a flush to the
   end of updateBars() which means that they redraw in real-time again now.
16 Altered some function9f names and made documentation about objects taking
   ownership clear - for example Track 'owns' inserted Parts, as does the
   PhraseList for Phrases, etc.
17 Created the tse3/adaptor directory and started working on the ListenerAdaptor
   framework. Created a SongListenerAdaptor.
18 A little work on the ListenerAdaptor idea.
20 Integrated the ListenerAdaptor class into TSE3 and made a SongListener.
21 Added a --mute-track flag to tse3play to help Joerg Anders diagnose dodgy
   AWE64 playback.
   Finished off the ListenerAdaptor work, and made the Song class use it.
22 Added all of the ListenerAdaptor classes. Laborious but necessary.
23 Added const to all the save methods.
25 (At the JaCC conference) incorporated an 'adaptive look ahead' calculation
   into the Transport class which works out the best lookAhead value according
   to the frequency that poll is called. Also added the ability to record the
   number of Transport break ups, and added a minimumLookAhead value to stop
   it being set to zero (frightening).
   Altered tse3play to show the current lookAhead value, and the number of
   break ups. Also fixed the bug which meant that some vu bars got stuck on
   the screen (the callback sets next[channe] to now[channel].
   Got anal about const correctness and size_t.
26 Finished off the size_t fixes - I had to add a couple of static_casts to
   TrackIterator which is distasteful. Ho hum.
   Added lastClock to Playable and implemented it in each Playable class.
27 Fixed another const foible, and removed the \es from tse3play.cpp, replacing
   them with a escChar static const - that removes all those nasty warnings.
   (I wanted to remove them earlier, but couldn't find out what \e was!)
   Created a simple progress bar in tse3play so you can see how far through the
   Song you are. Altered the adaptive look ahead formula slightly so that it
   doesn't break up on my desktop machine (which is more heavily loaded than
   the laptop).
   Fixed AWE drums bug: hurrah for undocumented OSS! AWE_DRUM_CHANNELS can only
   set *one* channel, and you pass it (1<<channel) as the parameter. Ho hum.
   Added a wee credits splash to tse3play, and a status message.
29 Added inject to the Transport class so that we can write useful things like
   on-screen keyboards.
30 Add the PowerQuantise class. For such a cool class, it didn't take much
   thinking since it just had to be converted from TSE2 to TSE3. It's much more
   logical in TSE3 style, which must be a good thing.
31 Added Mixer.cpp and Mixer.h with the Mixer, MixerPort and MixerChannel.
   Altered the Transport callback handling so that I can attach more than
   one callback (thinking specifically of the Mixer).
   Altered Part::lastClock. It did cunningly return the time of the last
   event, but I've had to alter that: there could be MidiCommand_NoteOffs
   that ar any time after that. That's annoying, since the song could stop
   before the lastClock because of that.

-------------------------------------------------------------------------------
April

1  Wrote test harness for the Mixer - it appears to work. Converted Song::track
   to oeprator[].
2  Moved Utilities.h classes into a utilities directory. Changed EventTrack::
   indexAtTime to index. Implemented Snap operator * that does the snap (I
   can't believe I ommitted to implement that, but then I've not yet used this
   kind of stuff). In doing this altered EventTrack::index to take a bool,
   for round up/down.
3  Began to integrate the TSE2 DeMIDIfy (putting it in the new utilities
   directory).
4  More ints to size_ts. Demidify.
5  Finished off the Demidify port. Untested.
6  Fixing and testing Demidify. It works, but I'm not convinced the output is
   the same as for TSE2.
7  Fixing and testing Demidify. I think that it produces the same output as the
   TSE2 version now, but I can't play the output since I've stiffed my sound
   card (sniff).
19 Uploaded version 0.0.1 to sourceforge, put up the website, announced on
   Freshmeat. Today is effectively the first public release.
25 Started on the Phrase utility explode.

-------------------------------------------------------------------------------
May

15 Added more documentation - specifically parameter and return values plus
   other gems that confused users.
16 More documentation and various minor fixes.
17 Joerg had some bugs in his use of the library which cause me to alter
   PhraseList::insert - you can now not insert a Phrase with no title.
27 Updated the documentation drastically.
30 Added the KeySigTrack. I'd never really thought that it was worth it, but
   Joerg Anders needs to be able to import KeySig changes in noteedit, and
   that's the final push I needed to make me add it.
   Blimey, I've been doing this for over a year now!
31 More work on the KeySigTrack, putting it in MidiFile import/export. Should
   be complete now - just for Jeorg to OK it.
   More kdoc following feedback.
   Moved the KDOC output dir to doc/api to help integration with HTML
   documentation and transfer to web site, also added the tse3-docs-clean
   target.
   Altered the Track class - it didn't send notification when the title is
   altered.

-------------------------------------------------------------------------------
June

2  Added the .cpp bits of the TrackListener which were missing.
3  Added unknownChunk and unknownData to SerializableLoadInfo.
   Walked around the API and tidied some of it up - renamed a few methods
   and changed some capitalisation to make it consistent. Perhaps a bit
   gratuitous, but it was worth it. Also changed the formatting of some
   documentation and other comments.
   Moved all of the MidiCommand_ definitions (et al) into enum definitions
   to tidy the Midi.h file up a bit.
4  Fixed the install path for the platform directory.
5  Swapped 'MidiClock' for 'Clock' - it shouldn't really be called Midi since
   it isn't exclusively for Midi bobbins.
   Also moved the Scheduler.* files to MidiScheduler.* since they are the
   only file names that don't match the class names.
6  Linked the utilties subdir into the KDOC output. Put in more redundant ';'s
   so that KDOC produces correct output. Grr. Improved the KDOCs by figuring
   out how to use @libdoc.
7  Implemented the GUS OSS device; largely based on the libkmid sources, but
   then it looks like these are largely based on some other sources :-)
   Added a --patches-dir switch to tse3play to be really useful.
   Shipped a pre-release version to Joerg to see if the changes he need work.
   Spent ages trying to put these changes into CVS over a trans-atlantic
   dialup link...
   Fixed bug in tse3play which was down to TSE3PlayVisual detaching a callback
   from a deleted Transport. Every day I get closer to implementing the 'safe'
   version of the observer pattern...
   Added a --force-port switch to tse3play.
   Fixed a 'MidiImport' bug that manifested itself randomly. It was actually
   the hint cached iterator in the PhraseEdit class - the act of inserting a
   new event invalidation the iterator, which made some inserts crash - my
   guess is when the data is relocated when the vector grows (this only
   showed up on reeeeeeeally big files). Fixed the problem by converting the
   hint to an index.
   Made Playable inherit from Notifier. This meant a little hassle and a few
   changes to the inheritance heirarchy (specifically, I had to add a
   FlagTrackIterator which does nothing). This prevents the warning of the
   cross cast when Transport compares Playable* with Notifier* in deleted().
   The only warning left in is in OSS.cpp - I can't remove that it's from the
   awe_voice header file.
   Started to create a MidiFileImportIterator that allows you to play a MIDI
   file without importing it. Seemed like a clever way to avoid the massive
   delay of import for big files. Got stuck when I realised that I can't match
   note offs to the note ons reliably.
8  Finished off the MidiFileImportIterator. Decided to ignore the note off
   problem - just accept that it can't be done with any efficiency or
   correctness. Either I pair up with the next note off that matches the note
   on, which may not be the correct one, or I preprocess the MIDI file
   prior to playing, which defeats the whole point of the iterator.
   Linked this iterator into tse3play and added a facility to skip empty space
   at the beginning of a MIDI file.
   Added --fast-midi to tse3play.
   Implemented MidiFileImport::lastClock which isn't as slow as I has expected.
   Added a filter to Transport, which means that it's easier to force to port.
   Fixed tse3play's broken --force-port - d'oh!
9  0.0.4 public release
12 Rew/ff in Resting mode. Fixed bug where shiftBy didn't alter the
   lastScheduledClock.
15 Fixes to recording after it failed on Joerg. Fixes to the OSS scheduler too.
   It's hard to test this problem since I can't get MIDI input ehre in the
   states. It will have to wait until I get back.
16 Record timing was out, fixed by turning a += into a =. I wish all problems
   were that simple.
   The TrackIterator moveTo didn't interact properly with getNextEvent. The
   set part pos was ignored, and reset to zero each time. The effect was that
   playback from time PPQN*40, say, included all the events from 0 to that
   point as well.
18 When doing ff/rew in Anthem it became apparent that the Part and Track
   iterators could only cope with playing from the beginning of a Part. Had
   to do quite a bit of surgery to get to it work properly.
19 Added the Record and PartSelection class declaration. Actually finished
   off the TransportListener interface. Implemented the Recording class.
   Changed MidiScheduler port references from ints to size_ts.
20 0.0.5 release to sourceforge.
   (At SFO) implemented the PartSelection class.
   Added the _track member to the Part class and made Track it's friend so
   that you can track which Track owns which Part. I'd always shied
   away from this in the past, but it seems a reasonable addition for the
   needs of Anthem.
   Added index to the Song class so that you can recalculate the index of a
   Track.
22 (Back in England) Added song() to the Track class, to make a path up similar
   to that I added in Part. I don't think I need it, but it's there anyway -
   at least it ensures that you can only add a Track to one Song (was this
   /really/ a problem anyway?).
   Added the InsertAction to the Part class - although not fully implemented
   yet.
23 Moved all exception classes to Error.h. Rationalised their names, and
   added each error code.
   Wrote Error.html documenation page.
24 Moved InsertAction to Part and put an extra parameter in setStart and setEnd.
   Because the Part now knows it's Track I can write code that allows you to
   set a Part's start/end whilst inserted - this is what the extra parameter
   is for.
   Added the DisplayParameters class and linked it into the Phrase and Part
   classes.
27 The MidiFileExport class didn't quite get around to writing key sig changes
   (although it worked out when to send them and had the logic to do so). Fixed.
   Added colour and setColour to DisplayParams.
   Added Track::remove(int,int) and noPartsBetween(...) to help with the new
   insert modes.
   Altered default Metronome note to one that is within GM specs. I thought that
   I had chosen such a drum sound, I guess TSE2 also had this problem!
   Thanks to Joerg cards with synth and MIDI combined now work properly (my test
   cards have one or the other but no both!) the problem was that synth and MIDI
   don't share port address space, but each have their own starting at zero.
   Implemented insert under for the Track::insert(Clock, Clock, InsertAction)
   method. In doing so I added Track::prvInsertPart.
28 In a shocking fit of good practise, tested the new Track insert methods and
   applied appropriate bugfixes. Even 'Under' works properly now.
   Implemented the thoughtful Part::setStart and setEnd that works correctly
   if the Part is inserted in a Track. Also added Part::setStartEnd so that you
   can move a Part without throwing exceptions, since I have also added
   exceptions that are thrown if you set start > end (and v.v.) if the Part is
   in a Track. Also added a ctor that takes time parameters.

-------------------------------------------------------------------------------
July

3  Added the PartDisplay application class to help handle the dual source of
   DisplayParams (in the Part and the Phrase).
   Added DisplayParams loading to the Phrase.
6  Added PhraseList::index.
7  Implemented Application::addSong and Application::history. I'm now actually
   using some of the command stuff which is nice. Made the Song info methods
   send a notification. Added notifications to the CommandHistory class so I
   can grey out buttons when there are no undo/redos left.
8  Added the Track_SetInfo command.
9  Finished the Track_SetInfo command.
   Added the Part_SetInfo command.
12 Added the Track_Snip command. Added Track::remove(size_t index).
   I hadn;t notice this up to now: the SimpleMidiEventFilter offset was
   being applied incorrectly - it should be subtracted, not added to times.
   This only showed up when redrawing the contents of Parts!
   Fixed some badly behaved uses of SEQ_BENDER in OSS.cpp, a bug found by
   Joerg Anders.
13 Fixed bug in PartSelection - didn't attach to Parts. Added some docs
   on the TSE3_Commands.
   Found a pretty difficult to spot bug: if a Listener gets an updated
   message and decided to detach from the Notifier (e.g. PartSelection
   when Part is unparented) then the Notifier's iterator is invalidated.
   This can cause nasty crashes. I don't like the solution, but it's
   to copy the Listener list and then iterator over this 'safe' copy. Euch.
   Finally succame to the tempation to create Notifier.cpp.
14 Added min and max Track index to PartSelection. Added the reparented stuff
   to the Track class so that it has the same good stuff as the Part class.
17 Fixed the PartSelection stuff which wasn't perfect.
18 0.0.6 release
19 Added a tse3-lib-local target to the top level Makefile, so it's easier
   to link to a non-installed library. I wish I could get the libs to build
   here straight out, but I can't see how to bully automake into it. I also
   discover that TSE3 only builds on my box - some autoconf weirdness.
26 Took out the Part::InsertAction and placed it into the utlities/Track.h
   file - it was making the API to clumsy, and the commands too hard to
   implement. The simple and sufficient API approach is better. Renamed
   Track::indexAtTime to just Track::index - it's more logical (the 'at time'
   bit is indicated by the Clock parameter, after all).
31 Moved the TSE3_* namespaces into the TSE3 namespace and renamed them. Maybe a
   gratutitous changes, but it's more logical for them to be contained, and it's
   easier to use if they have shorter names.

-------------------------------------------------------------------------------
August

2  Move the App/Plt/Cmd/Util subdirs to app/plt/cmd/util, to be consistent with
   my file naming conventions. The StreamMidiScheduler TSE3 class had a
   dependancy on the util lib which sucks. The neat solution is to move the
   Stream and Null MidiSchedulers into the Util namespace/lib.
   Moved the concrete MidiSchedulers to a Plt namespace.
   Removed redundant "Commands" from the cmd/*Commands.* filenames. Did some
   source formatting and added some more docs.
16 Altered FileBlockParser::parse by simplifiying the code ("find" is your
   friend ;-). Finally moved the Instrument classes to somewhere more logical,
   a new Ins sub-namespace.
   0.0.8 release
17 Added operator= and copy ctr to PartSelection and docs on how to use to
   avoid iterator invalidation.
21 Added the Destination utility class.
22 Added erase(int) to the EventTrack template for convenience. Also altered
   the insert method so you can't insert two events at the same time - when
   loading I was getting two timesigs at the same time, which mucked up my
   redraw code.
   (The code currently stands at 30692 lines and is 914k of source.)
   Added barBeatPulse to the TimeSigTrack.
29 Moved all that PartInsertAction stuff from util/Track.h to cmd/Part.h.
   0.0.9 release
30 Added Phrase_Merge for Joerg and took all the other Phrase bobbins
   out of the silly class (that was a TSE2 hang-over).
   Some code tidying and alteration of foo++ in ++foo all over the place.
   In an attempt to decrease header dependancies and generally tidy up
   split FileOperations.* into Serializable.* and TSE3MDL.*. Turned
   TSE2Import to TSE2MDL and changed filename. I want to get any nasty
   lingering code-changing bobbins out of the way quickly, ensure that the
   entire library is consistent and correct and then move out of alpha ASAP.
   Moved search and replace methods from Song to Util namespace. Put Pimpls
   into Song, Track and Part.
   Split EventFilter.* into MidiEventFilter.* and SimpleMidiEventFilter.*.
   Loads of repercussions from these changes and other tidies.
31 More of the same.
   Goodness - the MidiCommandFilter was never used or compiled. Builds now.
   SimpleMidiEventFilter was always a really unweildy name. So converted it
   to MidiFilter. The base MidiEventFilter class is now just Filter. Nicer.

-------------------------------------------------------------------------------
September

1  Release 0.0.10 (private)
5  Added the CommandGroup class.
   Added CommandHistory::undoCommand(size_t) and redoCommand(size_t) so I can
   have a menu of recent commands.
6  It seems as if the RepeatTrack was probably too grand an idea. You'd only
   really ever use a left/right marker like TSE2, so took out the RepeatTrack
   and put from/to/repeat into Song. Had to munge the SongIterator quite a
   bit.
7  Removed some cout diagnostics from App stuff.
   Alterered the way SongIterator implements the soloTrack - now sets commands
   to MidiCommand_Invalid rather than removing them completely. This means that
   auto stop still works as you'd expect (otherwise if you soloed to an empty
   Track the playback would stop).
11 Removed all the *Iterator classes from the public API, since they're all
   used via the PlayableIterator base class anyway. This makes the headers
   smaller, and the KDOC nicer ;-)
12 @bf in kdoc -> <b> As it turns out, that doesn't work either. Sigh.
   0.0.11 Release
   First bits of work at putting the new sexier Notifier framework in. It's
   really much nicer, and safer. libtse3.so.0.0.0 size before hand is
   5044711 for the record. The stuff gets a little messy around EventTrack
   but that has alwasy been a notifier nightmare. Touched 20 files.
13 More conversion to the new Notifier framework - it's a lot of hard work.   
14 Finally finished the conversion to the new Notifier framework. Seems to
   work really well. libtse3.so.0.0.0 is now 16522254. Erk.
   Stupid change because gcc < 2.93.0 is broken. The XxxImpl structs in
   Song.cpp, Part.cpp and Track.cpp should have members that default to
   public, as C++ specifies. gcc defaults them to private so the compile
   screws up. Duh. It wouldn't be too bad if the standard compiler versions for
   RedHat 6.2 and SuSE 7.0 (at least) are included in this.
15 Having been frightened by the huge template code bloat, implemented a much
   simpler Notifier framework. That's a real shame, but can't be helped. As it
   happens it's still much nicer than the original system (and more type safe)
   so it's still a win. A lot of conversion work (but not as much as before).
   KDOC bobbins: added @reimplemented in every conceivably unseful place.
   After these major alterations the size of libtse3.so.0.0.0 is 809919. Phew.
   Oh! No they're not - its still 16303602 bytes. Grr.
18 Over the weekend did some investigation into this dumb sizing. It turns
   out that it's all about link names. The 16m is due to use of the std::set
   data type in a template base class. If I turn it into list, it does
   down to about 8m and with vector lower than that. Sigh.
   The solution was to create a hacky void_list Impl class and cast all
   Notifier and Listener pointers to void* so store then in it, casting them
   back when they're needed (need to be careful when passing c_notifier_types
   that the recast type is correct ;-)
   So put back in all the good Event stuff that means listener functions
   can have any number of parameters. Things are nice again; good functionality
   and it only costs arounf 5749141 bytes.
   Put muldiv in the Impl namespace.
   0.0.12 Pre-release
22 Added a path delimiter to the OSSMidiScheduler patches directories so you
   can have multiple search paths. This means I should be able to accomodate
   SuSE and Redhat users better...
   0.0.12 Release
28 The Part copy ctor did evil things to the pimpl. Fixed.

-------------------------------------------------------------------------------
October

2  Added the Mutex class and the related classes to provide TSE3 with
   thread-safety. Added a bit of extra water-tightness to the Notifier class
   by making void_list::attach return a boolean.
11 Put a Destination class into the Application class.
12 Massaged the Song_EraseTrack and Song_InsertTrack commands. Renamed
   Song_EraseTrack to Song_RemoveTrack for consistency. Added Track_RemovePart.
   Added Track::index(Part*p) for consistency (and because I needed it).
13 Added threading support to the main TSE3 library; using the Impl::CritSec
   class.
14 Added TSE3_WITHOUT_MUTEX #define to the Mutex class so I can disable the
   whole shown if needed. Added a check in configure.in for whether Mutex
   support is needed.
15 Started the UnixMidiSchedulerFactory, which can be used to create both
   OSS and Alsa schedulers.
   Work on autoconf malarky to determine whether OSS/ALSA support is either
   wanted, or possible. Largely guesswork to see how to do it!
16 Made the configure.in stuff /much/ nicer. Mmm.
   Also, used the TSE3_WITH_OSS macro in the plt/OSS.* files.
   Implemented the UnixMidiSchedulerFactory. Made tse3play use this, and
   updated the documentation.
   Added new Mutex and Build docs, plus some examples (ripped from my
   article).
   At home much more of the Alsa implementation.
 * Release 0.0.13
17 Jigged around with the docs somewhat, and relayed out this file for laughs.
   Several minor doc changes. Fixed the make "tse3-lib-local" target.
   src/tse3/Makefile.am still used an old libtse3oss for the test prog, which
   has now gone. D'oh.
18 Many people have complained that Alsa support won't work (when the
   configure file clearly says at the end to build --without-alsa. I guess
   that was being a bit hopeful. Hacked it out completely ATM.
 * Release 0.0.14
20 Added Instrument class management to the Destination class.
23 Gratuituous change of Track::song() and Part::track() names, both to
   parent() since it is much more descriptive.
   Fixed bugs in Instrument loading routines.
25 More error handling for save/load operations.
26 Changes to the MidiScheduler interface: added portReadable and portWriteable.
   Changed std::strings to const char *. Removed txByte and added txSysEx.
   Added event in MidiSchedulerListener for when the number of ports changes.
 * Release 0.0.15
27 Added the App::Modified object.
30 Hid MidiMapper data behind a pimpl and made it a MidiSchedulerListener.
31 Application class handles Songs being deleted.

-------------------------------------------------------------------------------
November

3  Fixed bug in new MidiMapper - scheduler wasn't being stored.
7  Modified Part_Move so you can set the end and the start.
8  Added a whole load of new selection mechanisms to the PartSelection class
   (i.e. select all, invert etc). Also made a PartSelectionListener class.
9  Fixed Modified class if you initially specify no song.
10 Moved Modified into Util and altered Part_Move to it can also name itself
   "resize part".
13 Added the Progress class. Linked it to the Demidify utility, instead of
   the DemidifyProgress class.
   Linked this info the FileBlockParser and TSE3MDL classes.
   Added Application::numSongs.
   Added Progress to the TSE2MDL and MidiFileImport classes.
14 Added Progress to the Ins::Instrument classes.
15 The Instrument class remember's its filename - useful for the Application
   choices stuff.
   Added a preset colour capability to the DisplayParams object, and
   added the PresetColours class. The Application now also manages one of these
   objects.
16 Updated the PartDisplay to cope with the new PresetColours stuff. Changed
   the API slightly.
17 Added DisplayParams::presetColourString().
20 Modified's setModified goes public.
 * Release 0.0.16
21 D'oh! Progress.h was missing from src/tse/Makefile.am
 * Release 0.0.17
   Gave Phrase a parent() method to shadow Track and Part. You now can't
   insert a previously inserted Phrase into a PhraseList.
   Record now copes with start() being called when reset() isn't.
   Implemented Cmd::Phrase_Insert.
   Added Util::Phrase_Search, and used it to implement Cmd::Phrase_Remove.
23 Removed Phrase_Search again, since there is already Song_SearchForPhrase.
   More implementation of Phrase commands, including Phrase_Replace. Now
   the App::Record object uses commands.
   Improved Cmd namespace docs.
25 Fixed Modified.setSong(0); it detached from s, not _song.
   Phrase didn't save its DisplayParams.
27 Merged the two PhraseEdit::reset methods.
30 Improved documentation and added the Version file.

-------------------------------------------------------------------------------
December

1  Added Destination class Instrument lookup by name. This is used by the
   App choices file save/load mechanism. Began to implement Destination
   choices file stuff.
4  Destination defaults weren't set up.
6  Got the Destination stuff saved in the App choices files. Now the
   Instruments are saved and loaded properly. Also fixed the Choices loading
   mechanism which didn't work properly.
14 Bugfixed Destination choices saving.
15 Transferred the build across computers - from solomon to a new beefy machine,
   the build time is much faster! However this is a Red Hat 7.0 machine and
   there are a number of build problems - autoconf/automake both behave a little
   differently. PACKAGE and VERSION strings don't work, and libtool is set up
   incorrectly by autoconf.
   Finally go the build into a sane state with autoconf. I now know how to
   check out the pristine sources and run the correct commands to make the
   whole thing build OK.
17 Fixed bug in Destination saving for when instrument pointer is zero.
   Added Progress to MidiFileExport.
18 Fized grevious bug in TimeSigTempoKeyTrackIterator ctor which prevented
   MidiFile export working.
22 Updated tse3play verbose dignaostic MidiScheduler output.

-------------------------------------------------------------------------------
January 2001

3  Fixed bug in Destination class - instruments array was not the correct
   size leading to surreal application crashes.
4  Added .cvsignore files to make "cvs -nq update" a lot less noisy.
6  Lots of doc tidying and a few extra bits. Added consts to the Application
   and Record classes where they were missing and ensured that all channel/port
   params are in the same order.
   Converted TSE3_Version/TSE3_Copyright to return a "const char *".
   Moved App::ChoicesManager::ChoicesChoiceHandler from protected to private.
8  A big shakeup in the documentation, it is now "built" which means that I can
   put a standard header and footer onto it. Haven't worked out how to install
   the docs yet.
   Changed the licence to pure GPL (rather than library version).
9  Moved CommandHistory internal data from protected to private.
 * Release 0.0.18
10 Added a version number to the Application class, and made the choices
   file save both the application name and version - may be useful in the
   future.
   Added a ThisFileCreatedOn to the Application choices block too. It's cute.
   It's also probably going to be useful for debug. If it it aint, well, wibble!
16 Added Transport::playable(). Added htmlize.pl to doc/Makefile.am EXTRA_DIST.
 * Release 0.0.19
   Altered the RecordListener API so listeners can work out whether or not they
   were recording (in case there are two both listening to the one Recording
   object).
17 Fixed PartDisplay for empty Parts (i.e. with no Phrases).
   Added the TrackSelection class, an analogue of the PartSelection class.
   Tidied the PartSelection and TrackSelection together, this required some
   horrible nastiness in the the respective header files that I am not proud
   of.
22 Improved Notifier documentation.
   Rolled the libtse3util and libtse3plt into the main libtse3, since they are
   almost always used together anyway (you'd practically need one of either
   plt and util for a MidiScheduler, anyway, and neither are /that/ big).
   Library structure changed: now TSE3 is one big libtse3.so rather than
   many smaller libraries.
23 Added buffer flushing methods to CommandHistory.
   Fixed bug where TrackSelection could dereference a 0 minTrack pointer after
   recalculateEnds mucked the object state up.
   Added a whole pile of stuff to PhraseEdit to handle selections more
   gracefully and gave it good documentation.
24 Implemented the good PhraseEdit selection caching.
   Added PhraseEdit::eraseSelection.
   Put a whole load of modification status stuff into PhraseEdit and created a
   callback event for it.
 * Release 0.0.20
25 Forgot to remove some references to old libraries in tse3play makefile.
 * Release 0.0.20a
   A lot of investigation to see if I could get the Notifier framework any
   better - not possible! It was interesting, though :-)
26 Moved all Listener interfaces into the "listen" subdirectory. This prevents
   class users from having to include a whole class and everything that it
   depends on when using the Listener interface.
27 Tidied up some more arthermath of the restructuring.
   Now the documentation gets installed properly with the rest of the library.
28 More tidying. Moved Modified into TSE3::App.
   Moved TransportCallback into listen/Transport.h
29 More tidying.
   File loading may throw exceptions for bad format. The TSE3MDL load is
   now safe if an exception is thrown (no memory leak for the Song). It
   also uses a FileBlockParser for the top level.
   Altered TSE3MDL slightly: now everything is inside a chunk. Now it's
   completely orthogonal.
   Modified app/Choices.cpp similarly. Added noChunks to SerializableLoadInfo.
   Made FileBlockParser a little more tollerant - can now have comments
   twixt chunk name and opening brace, and empty lines too.
30 Discovered that autoconf enables debugging by default. The little critter.
   Put a huge warning for this in the configure.in file. This takes the combined
   lib size down from 6109854 to 1257958 bytes. Phew.
   The FileRecogniser now has a load() method to prevent the same code being
   written thousands of times.
   Made the installation of documentation a compile time option. Finished off
   the TSE3 spec file, I can now build an RPM of the TSE3 library. I needed
   to make the docs configurable 'cause otherwise the RPM installation
   generated two sets of documentation.

-------------------------------------------------------------------------------
February

1  Installed ALSA. Ouch. I don't have a card with supported synth capabilities.
2  Imported Matt Gerassimoff's work on the Alsa driver, got it compiling and
   patched it properly into the build system. UnixMidiSchedulerFactory can now
   create Alsa devices.
5  Began to tidy up some of Matt's work, bringing the code into line with the
   rest of the TSE3 style.
6  A lot of rework of the Alsa stuff - reading the Alsa library API and trying
   to implement it in a much cleaner way. Many implementation methods have moved
   into the AlsaImpl class.
   Just discovered that TSE3 0.0.17 and Anthem have made it into SuSE 7.1
   (alongside NoteEdit).
   The code currently stands at 42290 lines and is 1207k of source, not
   including documentation.
7  Alsa: Ports enumerate OK, and the timer stuff now works. No output goes to
   my MIDI device, though.
8  Got Alsa timing working. Got both Alsa input and output working.
   The original Alsa stuff that Matt wrote has now been completely ripped up and
   replaced. It all appears to work nicely now.
10 Fixed Alsa.cpp so it compiles queitly on a non-Alsa machine.
16 Fixed hienous bug in MidiFileExport - all files created were broken
   because of the tse3Message being saved with the incorrect length. This has
   been bust for a while, and is down to the fact I swapped a std::string for
   a const char *, and some accidental sizeof behaviour.
 * Release 0.0.21
   Added some more docs to the Part class to describe more clearly the need
   to set the start and end times - forgetting to do this is a common mistake.
   Put some more cunningness into PhraseEdit::createPhrase - you can now specify
   a PhraseList and get the Phrase automatically named and inserted. Maybe this
   won't get used much, maybe it will. Either way, it emphasised the need to
   name a Phrase before inserting it in a PhraseList.
   Added some extra docs to the Phrase class too.
   Rearchitected some of the Phrase stuff, now Phrases CANNOT exist outside of
   a PhraseList which will prevent a whole class of library user error.
   This required changes to PhraseEdit, PhraseList and Phrase.
   Changed me mind slightly, allowed you to remove and reinsert Phrases
   from the PhraseList - but you now can't use them if you do so - all Parts
   will ignore a reparented Phrase and you can't Part::setPhrase for an
   unparented Phrase.
   At home, added exceptions to Phrase_SetInfo, and did more work on the
   other Phrase commands and the Record object. Not finished yet. Tidied
   up other issues.
17 New implementation of Cmd::Phrase_Replace. Used it in the App::Record
   class again.
18 Bugfixing the updated command. Added an extra parameter to the
   Song_TrackRemoved event - the old Track's index.
19 Song.cpp bugfix - insert(int) set the new Track's parent to zero. Goodness
   knows why.
   Added PhraseEdit::timeShift() so that recorded data can be normalised.
   Made the App::Record class use timeShift().
20 PhraseEdit::tidy bugfix - unmatched note off removal on an empty PhraseEdit
   area would invalidate the data and cause weird segfaults. Doesn't now.
   Part::setPhrase can no longer dereference 0. Fixed Part_Move command wrt
   the newEnd time calculation, and fixed bug where ctor params shadow member
   variables.
   App::Record uses a CommandGroup for the insert Part section.
21 Found an alarming Notifier problem. When a Phrase::setParent call is made
   in one situation, not all Listeners get the event. In fact, if I send
   two events one after the other, one listener gets one event the first time
   it's sent. The other listener gets it the second time. my hair is getting
   pulled out...
22 PhraseList new title base names no longer include a trailing space, so
   the first suggestion is no longer "Phrase ".
   Fixed the messy Notifier problem. The issue was listeners detaching
   themselves in a callback - whilst the "listeners" void_list is being used.
   The for loop that iterates over it couldn't cope, and so had the ability
   to skip some listeners. The fix is a bit nasty - take a copy of the listeners
   list and iterate over that. Of course, some listeners might have removed
   themsleves, so I need to test each listener in the copy_list as it's iterated
   over with the real list, just to check that it still wants to receive the
   event. Sigh. It was a lot to get my head around, but it's working again now.
   Tidied up and gave more docs to the void_list.
   Updated the Notifer HTML docs.
   PhraseEdit::reset always sends a notification and always updates the
   selection information.
   Following a fault report, added #include <cstddef> to every header file
   that uses size_t. I was presuming that stl headers include it, which is
   somewhat bogus.
   Fixed Cmd::Phrase_SetInfo - if the title is the same as the old one, it
   was throwing an exception when it needn't. Fixed this problem.

-------------------------------------------------------------------------------
March

5  Fixed PhraseEdit::tidy() - used size() in a couple of places where it should
   have been size()-1. This was reported by Claus-Justus Heine.
   Tidied up the Alsa pimpl tx method slightly.
6  Joerg has discovered the bug which made Alsa timing dodgy - I did a bad
   calculation to get from nanoseconds to milliseconds in AlsaMidiScheduler::tx.
   Fixing this makes the AlsaMidiScheduler usable for the first time.
   Transport ctor sets _breakUps to 0.
 * Release 0.0.22
   Arse! I missed a zero from plt/Alsa.cpp. Fixed.
 * Release 0.0.22a
13 Updated the OSSMidiScheduler - hid all the nasty implementation classes from
   the public header file, and moved the patch setting APIs into the
   OSSMidiScheduler class.
19 Changed to a much nicer set of default preset colours, and coloured in the
   demo song.
21 Slight efficiency improvement in Notifier.h: in the Event class, the
   parameters are now held by reference, rather than copied.
   Applied a "work around" for an Alsa 0.5.x flaw in Alsa.cpp - the number
   of clock nanoseconds can be a ludicrous number, so its now checked by TSE3.
22 Small mod to StreamMidiScheduler so that clock() only increments the clock
   when it's running. I've been using it to debug Anthem, and the clock
   drifts when the scheduler is stopped which is quite disconcerting. Doesn't
   now.
   Joerg pointed out that my Alsa bug workaround was insufficient. The correct
   fix is now in place.
   Bugfixes for the Panic object - it now plays events properly. Still doesn't
   do the GM/GS/XG malarky, but that's really because I haven't got a decent
   streamable sysex facility in TSE3 yet...
   Gave the Part and Phrase listener interfaces windows into the changed
   events of their subcomponents (DisplayParams et al). This proved to be
   needed by Anthem.
   Added a DisplayParams object to Track. Updated Track_SetInfo. Gave the Track
   class the same forwarding event as I did the Phrase class.
23 Fixed annoying playback bug where the MidiParams weren't being filtered in
   either the Track or Part iterators.
   Added a whole pile more demo songs, and installed them in the Makefile.
   Added Type_Error to the FileRecogniser class for filenames that don't
   exist.
   Added the Melody PresetColour.
   The Alsa bug work around in plt/Alsa.cpp has become much more hideous. It
   was contributed Joerg Anders.
 * Release 0.0.23
27 More thoughts about sysex, as documented in Playable.html. Extended this
   doc file to describe the Midi.h definitons.
   Implemented GM resets in Panic.cpp using this new scheme.
28 Added running status to the OSSMidiScheduler so that sysex events will work.
   Implemented GM/GS/XG resets in the PanicIterator. Correctly handled the
   GS/XG device ID masks. Made the StreamMidiScheduler mention meta events
   (particularly invalid events).
   Got sysex working in the AlsaMidiScheduler. Cripes it was tedious. If only
   the sysex Alsa interface was documented. It turns out that even if you
   call it a sysex stream, you still need to include the start and end status
   bytes.
30 Bugfix (and semantics fix) to Cmd::Phrase_Erase.
   Fixed insideous little error in Song::remove(Track *t). It delegated
   the removal to Song::remove(size_t). However, it sent a Removed notification.
   Mostly worked as expected, but two notifictions got sent. This caused a
   very odd bit of behaviour in Anthem's TrackList and was quite tricky to
   find!
 * Release 0.0.24

-------------------------------------------------------------------------------
April

1  Made link lines in makefiles cleaner.
9  A lot of reworking of the Instrument class (and related classes) to make
   information access a whole bunch easier.
23 Added #define _GNU_SOURCE to the top of plt/Alsa.cpp which fixes errors
   in the linux headers on some platforms.

-------------------------------------------------------------------------------
May

11 Made a concerted effort to incorporate Stefan Westerfeld's aRts scheduler
   code into the TSE3 code base. I'd only put it off because I though it
   involved some hairy MCOP stuff, but thankfully it didn't.
   It's gone in OK, but I can't get it to work yet.
14 Random tse3play errors are in fact a non-handled switch statement for the
   new error case from FileRecogniser. Fixed. Still can't get aRts to work.
   Added ArtsMidiScheduler support to UnixMidiSchedulerFactory.
16 Fixed bug when playing TSE2 files.
   Demo TSE2 file added to the misc directory. Doesn't install, though.
 * Release 0.0.25
31 Got the instrument name lookup stuff working quite nicely. Proved this by
   using it in Anthem.

-------------------------------------------------------------------------------
June

30 Added the ability to get at an Instrument's ControlData and NrpnData.

-------------------------------------------------------------------------------
July

-
11 More documentation.
   Moved to beat release
 * 0.1.0 release
19 Started a marathon stint of work getting TSE3 to compile under gcc 3.0. The
   new compiler is /much/ stricter and there are a number of areas TSE3 was
   loose on. The most common (and easiest to fix) is the tightened use of the
   std namespace.
20 More gcc 3.0 work. Got it to compile, but can't yet run because the rest of
   my system is not built with gcc 3.0. Out of interest the file sizes are,
                             gcc 2.95.3        gcc 3.0
           debug build       6571618           9176054
           non-debug build   1372042           1403374
   That looks quite poor!
24 Started planning for improved serialization and MidiScheduler frameworks.
 * 0.1.1 release

-------------------------------------------------------------------------------
August

6  Work on a new MidiScheduler architecture that is more flexible.
10 Work on a more XML-like file structure - thinking of removing the
   Serializable class completely

-------------------------------------------------------------------------------
September

Somewhat distracted by other events

-------------------------------------------------------------------------------
October

9  Fiddled the MidiFileExport so that you give it a filename, as you do
   for import and TSE3MDL save/load. Added the MidiFileExportError.
   Fixed bug in MidiEvent constructor for NoteOns that meant no note
   off events were ever being created.
   Fixed bug in MidiData::lastClock which deferenced data.end() - that's
   an invalid data member.
   Added a new example - creating a scale, playing it and saving it as a
   MIDI file on request by a user.
...Work on a new MidiScheduler framework, and XML file loading (saving
   is complete)

-------------------------------------------------------------------------------
November

28 Added RPNs to the Instrument class, added the RpnData class and fixed the
   way NRPNs are loaded. Updated the Cakewalk ins format documentation too.
   Added the examples/midifile example.
 * 0.1.2 release (no new MidiScheduler stuff yet)
29 Added docs about MidiFile import/export exceptions.
   Added a reimplemetion of the standard "what" function to the Error class.
   Fixed the configure search for arts midi header files.

-------------------------------------------------------------------------------
December

3  Added work around for OSS MidiScheduler code: it appears that OSS doesn't
   support running status for AWE emulated MIDI ports properly. The
   MidiScheduler now detects them and attempts to work around this limitation.
   Put check for the ALSA verison into configure.in and Alsa.cpp.

-------------------------------------------------------------------------------
January 2002

3  Concerted effort to get the new MidiScheduler framework into the build
   properly, making pragmatic design decisions.
4  Incorporated Jose Maria Sanchez Saez' work on a Win32 MidiScheduler.
   Moved muldiv from MidiScheduler.cpp into Util namespace
5  Working from a pre-new MidiScheduler tree, sorted out the
   MidiSchedulerFactory. There are now no subclasses, just the one class that
   has different implementations on different platforms. That keeps things a
   little more simple. Used it in tse3play.
   Worked out how to detect cygwin on Win32, and added tests for this to
   the configure script, and patched into the build system for the plt
   directory. Now the Win32 scheduler code should be built on those
   platforms. I can't test this here, so I'll send it to Jose to see how it
   works for him.

-------------------------------------------------------------------------------
February

20 Added the contents example.

-------------------------------------------------------------------------------
March 2002

27 [ASIDE]
   Branched 0.1.2 for an Alsa 0.9.x maintenance release. A lot of build
   work to correctly regonise Alsa and the various different Alsa versions.
   Split implementations of Alsa up into different files Alsa-XXX.cpp.
   Integrated 0.9.x work kindly done by Takashi Iwai of SuSE.
   Modified configure.in for aRts header detection too.
 * 0.1.3 release (branched version)
28 Merged 0.1.3 branch onto trunk.

-------------------------------------------------------------------------------
April

4  Added port to Panic
8  Ported OSS MidiScheduler to new implementation.
10 Added concept of "all ports" port number and "no port" port number.
   Began the porting work for Alsa code.
16 Ported Alsa 0.5 to new MidiScheduler framework. Added $KDEDIRS linkage
   to configure.in for aRts library.
17 Implemented AllPorts and NoPort in the MidiScheduler. Added the
   MidiScheduler::portNumber API for simplicity.
   MidiEcho implementation of port/channel acceptance was wrong - it implemented
   forcing to the specified destination, NOT filtering the source. Fixed,
   and also prevented reliance on valid port numbers with the new MidiScheduler.
22 Song::lastClock is now cached.
   Changed Record ctor, since you don't need to specify a MidiScheduler (it is
   available through Transport::scheduler()).
   Fixed MidiScheduler.h bug: clock() didn't account for "_running == false"
   Expanded MidiSchedulerListener to have _PortAdded and _PortRemoved.
   Emitted these in the relevant places.
25 Added time parameter to the EventTrackListener interfaces.
   Added the Track_Sort command.
26 Sped up Track_Sort considerably by moving away from a quicksort algorithm,
   to a more careful and less track-swap-hungry approach.
   The sort command now also reselects selected Tracks once they've all been
   moved about.
   EventTrack::insert returns a size_t index of the new event. Changed
   the listener interface so that the parameter is the index.
28 Named the anonymous enums in the MidiCommand class that describe "magic"
   port and channel numbers - they were causing some compiler warnings.
   Adjusted the demo songs so that they honour the magic channel/ports.
29 Fixed MidiFilter in light of new magic port/channel numbers.
 * Release 0.2.0

-------------------------------------------------------------------------------
May

3  MidiEcho default channel is sensible. The Part_Move command correctly
   names itself when Parts are moved between Tracks.
   Late night hackery to get TSE3 to build under automake 1.6.1. It does
   some weird stuff with .deps which caused failures since I had a rule
   libtse3plt_ls_SOURCES_alsa = Alsa-$(TSE3_ALSA_VERSION).cpp
7  Some fixes for building with Alsa 0.9.
   More documentation for command classes - placing it into the standard
   classes documentation for reference. In doing this realised that
   Song_ReplacePhrase clashed with Phrase_Replace. Removed former.
8  Moved the "misc" directory to "demos".
 * Release 0.2.1
9  Added MidiScheduler::numberToIndex(). MidiFilter has sensible
   default port and channel.
   Alsa 0.9 build fixes and aRts fixes under SuSe 8.0.
 * Release 0.2.2 
13 Fixed some issues: MidiFilter.cpp - removed run time check that was replaced
   by a compile-time check. PartIterator::Part_PhraseAltered doesn't resend
   the MidiParams.
   Sorted out SerializableLoadIterator ctor so that it works with gcc 3.0.x -
   initialising an int with a negative number in an initialiser list doesn't
   work. Sigh.
   PhraseList is more careful if Phrase creation fails during load.
   A little more thoughtful dealing with MidiParams in the PartIterator during
   moves.
14 Fixed crash in Anthem when it is closed whilst playing - the TrackIterator
   didn't check that _track hadn't been deleted when doing a "moveTo".
16 Fixed the PartIterator so that it doesn't throw away the first event
   of the Phrase.
17 Added "-l" option to tse3play, this was Joerg's suggestion.
   Added ports example. Fixed MidiEcho bug.
 * Release 0.2.3 
21 Added more docs to MidiEcho to make it clear what setChannel/setPort do.

-------------------------------------------------------------------------------
July

10 MidiFileExport and TSE3MDL can save to an ostream as well as a filename.
   You can specify "-" as tse3play output parameters.
11 Got MidiFileExport ofstream flags right, so file gets truncated.
16 configure.in fix thanks to Sbastien Prud'homme.
28 Changed email address away from old Pace one since I no longer work there.
 * Release 0.2.4 

-------------------------------------------------------------------------------
August

22 MidiFile.cpp patch so that it compiles under gcc 3.2.
 * Release 0.2.5 

-------------------------------------------------------------------------------
October

22 Bug fix tse3play, default port is NoPort
   Added a MidiScheduler concept of internal and external ports.
   Updated the implementation accordingly.
   Made tse3play a lot more clever in which ports it picks for playback.
23 Added Doxyfile
24 More MidiScheduler implementation docs.
30 Put the doxygen documentation completely in the distribution
 * Release 0.2.6 

-------------------------------------------------------------------------------
February 2003

10 Tidied tse3play so you don't need to open the MIDI devices to convert
   files.

-------------------------------------------------------------------------------
March

6  Added MIDI record example application.
7  Added definition of TSE3::Clock::PPQN to Midi.cpp
 * Release 0.2.7
12 Fixed compilation error ni Alsa-0.9.cpp

-------------------------------------------------------------------------------
 TODO
-------------------------------------------------------------------------------

Core TSE3 library 
  * AllChannels/NoChannels in MidiScheduler - or punt out?
  * XML file format loading
  * Check use of EventTrack::index, now there's a bool parameter, can some
    redundant logic be removed?
  * Lyric track
  * Plugin interface for audio/video tracks
  * Load any type of file utility
  * PowerQuantise with groove taken from another Phrase/Track.
  * Finish off the Demidify utility.
  * Mixer architecture needs sorting out now ports can be arbitary numbers
  * MidiScheduler catches sysex into a bit bucket, and then sends it out
    wholesale

tse3play
  * Reset sent on ctrl-c

Issues
  * disable RepeatTrack, iterator moves to 0, _more == false. Enable it,
    it will move to 0 which will be wrong.
        - I've done some thinking about this - see paper notes (somewhere).

-------------------------------------------------------------------------------