PUAE/CHANGES

20040302
- Fix building of SCSI emulation on OS X. Update config script to link with
  IOKit framework when building against libscg on OS X. Update install_scg
  script to re-run ranlib on libscg/libschily on OS X after copying.
- Tweaks for building scsi emulation on BeOS.
- Tidied up sleep code. Reimplemented my_usleep() to uae_msleep() which
  takes a sleep duration in milliseconds rather then usecs.
- Misc clean-ups in preparation for getting stuff merged in WinUAE tree.
- Hacked support in config script for compiling against libscg/libschily
  external to source tree (this needs tidying up). Yet more options -
  --with-libscg-prefix etc.
- Bug in libscg block driver was causing segfault when CD32 was reset. Fixed.
- Libscg block driver wasn't clearing SCSI error codes before executing SCSI
  inquiry or test unit ready commands (I was passing args in the wrong order
  to memset. Ooops!). This was causing problems when scanning the SCSI bus
  (particularly for real SCSI busses on the host) and was causing CD32
  emulation to fail to detect disk changes. Fixed.

20040222
- Only build enforcer when we have JIT/natmem support.
- Begin merging GTK+ changes back (bearing in mind that this time GTK thread 
  is secondary again). GUI can again be opened after emulation proper has 
  been started and failure to open GTK+ GUI will not cause UAE to fail.
- Fix configure script with regard to libscg. When building in source tree,
  add current directory to library path.
- Fix configure script with regard to CAPS. Use CPPFLAGS rather than CFLAGS
  to specify directory to include CAPS header from.
- Fix install_libscg script to work on *BSD (and OS X). Problems still exist
  linking on OS X due to the whacky linker there.
- Replace mode sense/select 6 to 10 translation in libscg block device driver
  with calls to the new generic code imported from WinUAE 0.8.25.
- Merged WinUAE 0.8.25
- Back out GTK+ GUI changes. It seems that running the GUI as the primary
  thread and SDL as the secondary really hurts performance. Also back out
  SDL timers for clock speed calibration since it's not needed now.
- Added config option --enable-enforcer
- Added AHI emulation glue from WinUAE (this includes support for WinUAE
  enforcer, etc.) This hodge-podge of code is in misc.c (for want of a
  better place to put it). AHI emulation still needs porting and back-ends
  for the various sound drivers supported (which will be fun).
- Merged changes from WinUAE 0.8.24
- Fixed bug in libscg driver which was causing segfault on reset

20040219 (not publicly released)
- CD led display now works.
- Assorted fixes for building akiko (CD32) emulation.
- Implemented block device driver using cdrtools 2.0.x's libscg (this is
  largely based on the old scsidev driver). CD/DVD devices only are
  currently supported.
- Quick fix for the Unix bsdsocket emulation, so that it no longer directly
  requires POSIX threads, semaphores and thread-local storage. This fixes
  many problems when UAE is built against SDL. It should also now work
  on OS X with SDL now, but it still doesn't on BeOS due to limitations with 
  the BeOS socket API. Many problems remain, and a re-write is still needed.
- Support for IPF (CAPS) floppy images added.
- Use SDL timers for calibration in md-i386-gcc when building with SDL.
  Using POSIX timers directly can screw threading. This is currently less
  accurate, and requires a longer calibration period. 
- Fix for building on older SDL versions without mousewheel support.
- Fixes for SDL sound which were causing threading problems.
- Implement the Gtk+ LEDs updates as a separate callback. This lets us
  schedule the events callback more frequently.
- Make GTK+ GUI thread the main thread which launches the UAE thread (rather
  than vice-versa). Makes the code slightly cleaner, and allows the GUI to
  be opened after the emulation is started. The use_gui config file option
  now means that the GUI is not opened at start-up, but may be opened later.
  Added hotkey combination F12 + Enter to SDL and X11 targets to for this
  purpose.
- Work-around that allows filesystems to be mounted from host ISO filesystems
  when flagged as read/write in the config file. On systems with statvfs() and
  on Linux (where statvfs() is not used because of broken implementations) UAE
  will now mount such filesystems as read-only. More work needed here.
- On md-ppc-gcc target, make m68k CPU flags a bitfield. This generates better
  code.
- SDL gfx driver was broken on 8-bit host displays. Fixed for both native and
  P96 screens. Try running UAE on the Linux framebuffer device at 8-bit if
  you want to see some serious speed.

20040129
- Ta da! JIT now works on BeOS/x86 (albeit without direct memory support).
  Still, it gives 10x the speed of pure interpretation. I finally sussed
  what was causing the stack problems - cpuopti! Solution: don't run cpuopti
  on platforms that can't handle the x86 assembly magic in newcpu.c for
  calling opcode handlers - e.g. BeOS. Obvious when ya think about it. ;-)
- Merged and updated Axel's BeOS sound driver from BeUAE. Sound quality is
  still poor on my system - but somewhat less so. ;-)
- Config file support for running xres/mimeset on the compiled binary when
  targetting BeOS. Added Axel's resource file to the dist.
- Tidied up BeOS floppy image dialog. If invoked in full-screen mode, UAE
  will now try and switch to windowed mode before opening the dialog.
 
20040126
- Dist was missing a file from src/gfx-x11. Fixed.
- Typo in OS x raw key driver. Fixed.

20040125
- In Amiga target, set the UAE processes's pr_WindowPtr to ensure requesters
  appear on UAE's screen.
- Tidy up Amiga gfx target. Remove Graffiti support. Add new Amiga-specific
  options to configure the type of screen UAE runs on (still more to do).
- Allow comments again in cfgfiles. Lines beginning with '#' will be ignored.
- Remove hacks in sleep.c. 
- Add support for popping up floppy dialogs in Gtk+ UI in response to
  gui_display() callback from UAE.
- Fix raw keyboard handling with SDL target. If the SDL window loses focus
  when keys are held down, SDL generates key-up events with 0 keycode for
  the keys that are pressed. UAE couldn't generate proper key-up events and
  keys were 'sticking'. Respond to loss of focus and force keys to be
  'released'.
- Initial implementation of main.cpp for BeOS.
- Break out gui stubs from beos.cpp into separate file gui.cpp. Implemented
  gui_message() via a BAlert and implemented file dialog for inserting disk
  images.
- Fix fsusage.c on BeOS. The K&R declaration for adjust_blocks() was
  screwing up the compiler for some reason.
- Fix to fsusage.c to build on systems with a working SVR4 statvfs() (many
  non-glibc Unices). I had forgotten to revert back to using STAT_STATVFS
  rather than HAVE_STATVFS. Thanks to Ian Chapman for reporting the problem
  and testing the fix.
- Comment out JIT debugging messages in the Linux segv handler with
  #ifdef JIT_DEBUG like the win32 handler does. The user doesn't need to be
  worried by all those scary messages.
- Fix GCC-specific abort() macro to call exit() rather than abort(). This
  ensures that any atexit() routines are called -  and the user will get
  their X display back.
- Fix AmigaOS POSIX semaphore emulation so that the semaphore thread is always
  killed at exit (previously it wouldn't be unless all sempahores had been
  deleted).
- Move disk image requester code from ami-win.c to ami-gui.c. Now used in
  response to an appropriate gui_display() call. 
- Re-implemented gui_message() in AmigaOS target to display messages as
  EasyRequest requesters.
- Re-implemented and cleaned up keyboard handling in AmigaOS target to fully use
  inputdevice.c and new hotkey key code. AmigaOS now supports a similar range
  of hotkeys as X11 and SDL targets - all starting with Ctrl+LAlt rather
  than F12.
- Implemented os-target main.c support (e.g., for start-up and support code).
  Factored out break handling from the gfx targets and implemented it in the
  various os targets.
- Added option sdl.map_raw_keys to allow the choice between mapping cooked
  or raw keys with SDL gfx target.
- Added support for parsing gfx-target options.
- Separated gfx target from OS target. Unix target is now called 'unix'
  rather than the previously confusing 'x11'. Gfx targets include 'x11',
  'sdl', etc.
- Inputdevice handling overhaul (work-in-progress)
  * Got rid of the input-dummy.c hack and implemented the necessary code in
    each  of the gfx drivers.
  * Cleaned up and simplified input handling in X11 and SDL targets, now
    re-using more code from inputdevice.c.
  * Implemented mouse wheel support in X11 and SDL targets.
  * Initial implementation of target-independent hotkey handling.
  * Initial support for target-independent raw key mapping (currently used in
    SDL only).
- Screen mode tweaks to SDL gfx target. SDL not supplying precisely the mode
  you asked for should cause fewer problems now. UAE also now asks for a hardware
  surface (which works with, e.g., the dga and fbcon SDL video drivers).
- Implemented simple on-screen P96 RectFills and Blits using SDL blitting
  functions (this can double the scroll speed of CON: windows even when
  running  in a SW surface).
- Lots of changes to the P96 emulation bringing in new code from Win32
  version - fixes the annoying scroll bug with CON: windows. (The big-endian
  red/blue swapping bug which can occur with some types of blits on 32-bit
  screens remains - and I'm convinced it's a bug in P96 itself. ;-)
- GUI tweaks to go with the --disable-natmem option.
- Added new config option --disable-natmem to build the JIT without direct
  memory access. This should let the JIT build on other x86 Unices and
  Unix-likes (JIT direct memory is currently dependent on x86/Linux segv
  signal handling) until we can make it more portable. It'll also now
  build on BeOS, although it bombs at run-time. Something strange seems to
  be going  on with the stack there, I think.
- Tweaks to let compemu_support.c build without sys/mman.h (e.g. on BeOS)
- Merged changes from my 0.8.23 branch. This means we're now based on WinUAE
  0.8.23.
- Hacked in support for building GTK+ GUI with GTK+2.x (the config script
  will prefer GTK+2.x over GTK+1.x if both are available).
- Began importing new GTK+ GUI elements, starting with new CPU prefs page.
  This'll now let you play with idling (if you're JIT-enabled) and my
  busy-wait switch. (I'm not happy with the layout of this page, however, so]
  it'll change again for the next release.)  
- Bumped AmigaOS default stack to 32k (it uses a lot more when run on
  AmigaOS < 3.9 for some reason).
- Tweaks for building on MorphOS. It now builds against ixemul, but not
  libnix (there appears to be a lot missing from MorphOS's implementation of
  libnix). I cannot test it, however, 'cos I ain't got MorphOS.

20040107
- Fix for floppy data not being initialized to 0 when allocated on the heap 
  (when SAVE_MEMORY is defined).
- Fix for typo in gfx-x11/xwin.c which was causing compilation to fail
  (thanks to Giacomo Magnini for spotting this one).
- Add __stack declaration to AmigaOS target to enable automatic enlargement
  of stack with libnix's swapstack module if the stack size is too low on
  start-up. Set min size to 16384 for now, which stack snooping reveals is
  erring heavily on the side of caution. I don't want to use stack
  checking/extension code because UAE is already slow enough on a classic
  Amiga. ;-)

20040106
- Fix for libnix auto-opening libraries on AmigaOS target and thus asking
  for inflated version numbers (it asks for v37 of everything by default).
- Fix typos in BeOS implementation of fsusage.

20040104
- Use sample8_handler() rather than ULAW handler for 8-bit mono audio in SDL
  sound driver.
- Fix switching of audio interpolation mode in x11 and SDL gfx targets.
- Define PICASSO96_SUPPORTED declaration in targets/t-beos.h when building
  with SDL so that P96 support will be included.
- Renamed set_alarm function in md-i386-gcc/support.c since it conflicts
  with a BeOS kernel function.
- Added a -version command-line option. 
- Added full package version string to about page in Gtk UI (if defined).
- Fixed HD LED on no-gui, AmigaOS and BeOS targets.
- Re-did my implementation of POSIX semaphores for AmigaOS target. My
  previous version didn't work, since it was pretty much a straightforward 
  mapping of POSIX semaphores onto Exec semaphores - but, since POSIX sems
  don't have an owner whereas Exec sems do, this obviously wasn't going to
  do the job. Doh! Should read docs more carefully in future. The new
  implementation uses message-passing with a semaphore server thread to
  achieve locking - and, as an added bonus, locking now obeys break signals.
  Filesystem threads and hardfile support now both work on AmigaOS. Yay! 
- Implemented fsuage for BeOS.
- Implemented new BeOS sound driver. 
- Detect libossaudio in config script to enable use of OSS audio on
  NetBSD/OpenBSD. Swap order of OSS and Sun audio in config script, so that
  Sun audio is preferred over OSS on NetBSD etc.
- Port od-beos/beos.cpp to new inputdevice API.
- Disable BeOS sound driver. Needs to be rewritten.
- Break out BeOS joystick code to separate file (joystick.cpp) and port to new
  inputdevice API. Now uses enhanced joystick API (although currently
  support only 1 stick).
- Tweaks to od-beos/beos.cpp to get it closer to compiling.
- Use snooze() for my_usleep() on BeOS
- Fix x11 gfx target so it'll build without P96 support.
- Don't define NO_MAIN_IN_MAIN_C when compiling against SDL on BeOS. This'll
  need to be changed later if we want the BeOS GUI to work with SDL target.
- Fixed P96 support in x11 target
- Added a log statement when retrying a hardfile mounted as read/write but
  without write permission as read-only.
- Fixed hardfile-unix.c to obey read-only flag.
- Fixed expansion.c to build without P96
- Fixed display of FPS counter on AmigaOS, GTK+ and no-gui targets.
- Merged new autoconf 2.5x/automake build system. Still very much a
  work-in-progress. Advanatges: more flexible, more config options, better
  support for cross-compiling and for building in separate directory.
- Fixed a serious typo: sleep was being used instead of usleep (if nanosleep
  was not available).
- Merged updated sleep code from my 0.8.23 branch. For systems with rpt, UAE
  will now test the chosen sleep routine and decide whether busy waiting is
  necessary or not. (If it takes 10ms or more to do a 1ms sleep, then busy-wait
  by default. The user will be eevntually able to turn this off with a config 
  option.)
- When building with the SAVE_MEMORY option, allocate floppy data at
  run-time. This shaves MBs off the BSS size and means that the AmigaOS
  binary doesn't need such a large unfragmented piece of memory to run.
- Make memory.h include osdep/memory.h to pick up OS-specific options.
  AmigaOS version now picks up the SAVE_MEMORY option correctly.
- Fixed AmigaOS gfx driver not clearing screen properly
- Implemented fsusage for AmigaOS
- Implemented frame timing for classic (m68k) AmigaOS using the EClock.
- Implemented AmigaOS threading/semaphore code (not complete).
- Implemented AmigaOS joystick driver using gameport.device
- Applied patch from Giuseppe Ghib<69> to re-enable the -h command-line option
  to show information about config options.
- Renamed the bsdsocket.c to bsdsocket-posix.c and imported the old
  bsdsocket.c file (as used by WinUAE). Hopefully one day there can be some
  reconciliation between the two. ;-)
- Modified the bsdsocket emulation so that a trap is installed even when the
  emulation proper is not built. This stops UAE crashing when the Amiga side
  of the emulation is used, but no emulation is present in UAE itself.
- Fixed bsdsocket emulation so that it really (this time) obeys the bsdsock
  config file option.

20031215
- If options file not found in home directory or current directory, then
  use home directory as default path. This enables the GTK+ UI to save
  a newly generated config file in a sane place.
- In SDL target, changed full-screen mode to always grab mouse input. This
  works around mouse motion problems in full-screen mode in OS X.
- Kludged on a widget for selecting cycle-exact CPU mode in the gtk UI. I
  don't want to add much to the GUI just now because it's in the process of
  being rewritten - but this was cheap to do and it doesn't make sense to
  build in cycle-exact mode if it cannot be easily used.
- Tweaks to allow building without cycle-exact CPU (CPUEMU_12 undefined).
  There's no config file support for this as yet, but if you want to save
  buckets of time when compiling, load up src/Makefile and remove that
  define in CFLAGS and cpuemu_*_6.o/compemu_*_6.o from OBJS.
- Fixed a race condition which caused the name of a newly selected kickstart
  image or key file sometimes not to be displayed.
- Added a scale factor to read_processor_time() on x86 target. People were
  reporting timing problems on fast machines, and I believe the problem is due
  to the number of ticks per second of the processor clock (syncbase) not
  fitting in a signed int (on 2GHz+ machines). By scaling down the clock
  (currently by a factor of 16), hopefully this problem is solved. Also
  modified the RDTSC call to return the processor time as a full 64-bits. This
  is needed so that scaled down clock still rolls over in 32 bits (and indeed
  read_processor_time() still returns time as 32 bits).
- Added a patch from Stephen Anthony to make the floppy file dialog in gtkui
  remember the last path used (saves all that searching through folders to
  find the right ADF file).
- Initial implementation of hardfile support on Unix (no LFS or Gtk GUI
  support yet). RDB hardfiles cause problems with recent versions of HDToolBox
  which needs to be further investigated.
- Some minor tweaks to AmigaOS and BeOS targets (AmigaOS target should now
  build out of the box, but still a lot of work to do on BeOS).
- Lots of tidying doing trivial syncs with WinUAE 0.8.22r9.
- Cleaned up sleep_millis() and sleep_millis_busy(). They're now in sleep.c
  and will call my_usleep() - which will be replaced with an apropriate libc
  or OS call depending on what's available (e.g., nanosleep(), usleep() or
  SDL_Delay()). sleep_millis() will now busy wait if possible when asked to
  sleep for less than 10ms (the granularity of most system sleep routines will
  not allow sleeping for shorter periods - according to the docs).
  sleep_millis_busy() will now only sleep if busy waiting is not required (for
  sub-10ms sleeps).
- Cleaned up MAX_PATH nonsense. Arbitrarily assigned a value of 512 if it's
  not defined. (We could use MAXPATHLEN from POSIX headers, but this value
  is just as arbitrary and rather wasteful. E.g., it's 2048 on Linux. Anybody
  really need paths that long?)
- Commented out unused routines createinputfile(), etc. in zfile.c which were
  there to support external dms tool but are no longer required (and anyway
  don't build on AmigaOS without ixemul due to the lack of the dubious
  tempnam() function).
- Tidied up usage of strcmpi() and stricmp(). Replaced with strcasecmp() and
  added configure support for substituting strcasecmp() for one of these if
  it isn't available. (Shouldn't be necessary. strcasecmp() is a BSD standard,
  but the other two are aren't always available - e.g. Linux doesn't have them.)
- Fixed configure script to build bsdsocket emulation by default only on
  Unix platforms.
- Fixed misreporting of bsdsocket emulation not being built when it has been.

20031202
- Modified gencpu (and build scripts) to generate cpuemu.c in three pieces
  like it does in WinUAE. This should require less memory to build the CPU
  emulation and should thus speed up compilation on GCC3 (it also opens the
  door to compile-time configuration of the CPU emulation).
- Modified configure script to omit X86_ASSEMBLY option when building for
  x86 BeOS (it doesn't work there).
- Modified configure script to allow building with GCC 3 on x86 (this was
  already possible on other archs). The cpuopti step is simply omitted
  (since cpuopti won't currently work with GCC 3 output).
- Updated Amiga target. Closer to building now (with GCC at least). Sound and
  joystick support still to do, and config options probably need tweaking.
- Add updated Solaris sound driver from Elmar Plischke.
- Added patch from Elmar to allow bsdsocket emulation to compile on Solaris.
- Removed GNU varargs macros throughout (this should allow UAE to be compiled
  without GCC once more). Thanks to Elmar for pointing this out.
- Freeing memory in mapped_free() was broken when JIT was compiled in but
  direct memory access disabled. Fixed
- Imported support for using DMS floppy images from WinUAE.
- Support for using zipped or gzipped ADF files was broken. Fixed.
- Accept __POWERPC__ and __ppc__ as a synonyms of __powerpc__ when
  testing for PPC architecture in configure script (GCC on MacOS X uses
  these).
- Cleaned up custom.c and drawing.c in preparation for merge with WinUAE. (This
  breaks P96 in SDL and X11 targets. I've fixed SDL, but X11 fix is still to
  do).
- Added support for switching between windowed/full-screen mode on SDL target,
  even when SDL doesn't support SDL_WM_ToggleFullScreen (e.g. MacOS X).
- Added hack to allow the use of F11 as the hot-key modifier in SDL target.
  F12 doesn't work as other keys under MacOS X it seems. Key up/down events
  are generated only when the key is released - so it's no good for a hot-key
  modifier.
- When changing from a P96 screen mode to an Amiga screen mode, a segfault
  could occur. Fixed.
- JIT causes segfault on start-up on 2.6 kernels (and some 2.4 kernels).
  Problem is due to messing about with the processes's LDT (Local Descriptor
  Table) to accelerate access to VM's memory. Disabled this for now until I
  find a solution.
- Partial support for Danish keyboards in SDL target (some keys aren't
  currently supported by SDL keysyms).

20031118
- Added scripts to build debian package
- Added Carl Drougge's bsdsocket.library and source to package (this handles
  the Amiga side of his bsdsocket emulation, which is already included in this
  tree).
- Added this changelog to package
- Updated docs dir from uae-0.8.22 (I still had 0.8.21).
- Added 1280x960 screen mode to SDL and X11 targets - I prefer square pixels!
- Changing Amiga/P96 screen mode in SDL target while mouse was grabbed would
  screw mouse motion because UAE would forget mouse mode. Fixed.
- Changing Amiga/P96 screen mode in SDL target would cause full-screen mode
  to be lost. Fixed.
- Changing P96 screen mode to a larger mode in SDL target caused a crash. Fixed.
- Re-did video mode selection in SDL target. 15-bit modes now work for P96.
- Ensured that files which include the SDL headers do so with the proper
  path (That is, no path. It's provided by sdl-config).
- Lots of clean-ups in preparation for getting stuff merged in the WinUAE tree.
- Removed code for the Windows target from this tree. Eventually I want to
  get it building with MinGW, Cygwin and maybe even OpenWatcom - but we're
  some way from that goal, so for right now we'll just save some bandwidth.
- Removed code for Acorn, pOS, DOS, OS/2 and NeXT targets. I'm sure that these
  don't work now, and I don't have the facilities to test them anyway.
- Recoded source files to ensure the use of Unix-style end-of-line markers.
  GCC on BeOS was barfing on the DOS CR/LF codes that accidentally got imported
  from the WinUAE tree.

20031113
- Added mapping of left and right "super" keys (Windows keys on a Windows
  keyboard) to left and right Amiga keys in SDL and X targets.
- Swapped mapping of apostrophe and back quote keys in SDL target

20031106
- Silly me. Timehack was demanding clocks to be exactly in sync with 1us
  accuracy.  Made it less strict and only update amiga time if it is out by
  more  than 1ms.
- Added dummy set_thread_priority() function in sdl and posix thread
  wrappers (require if using filesystem threads).
- Tweaks for configure script:
  * Better detection of when thread support is required (currently
    filesystem code requires threads - even when not building with
    UAE_FILESYS_THREADS option - due to dependency on hardfile code).
  * Filesystem threads will now be compiled in by default
  * Configure now works again on BeOS (whether it will build is another
    story.)
- Fixes to re-enable building without a GUI.
- Made bsdsocket emulation a compile-time option.
- Really fix dummy joystick driver in od-generic.
- Fixed joystick counting in Linux joystick driver.
- Fixed mouse button mapping in X target.
- Fixed moused movement in DGA mode and when grabbed in X target
- Fixed X pointer being displayed in UAE window after grabbing and
  ungrabbing mouse with X target.
- Fixed various functions which should have been exported from
  gfxutil.c but were declared static.

20031011
- Fixed UAE side of timehack which was translating Unix Epoch UTC
  time to Amiga Epoch UTC+2.
- Added support in Amiga side of timehack for setting clock to local time
  using offset from UTC read from environment variable TZ. Added 
  break handling, so you can now kill the damn thing.
- Ported dummy sound driver in generic target to new sound API. 
- Put back big-endian fixes for OSS sound target (which for some reason 
  never made it into this tree).
- Got X11 target compiling again. Still needs work on mouse handling.
- Linux joystick driver now works. (Did it ever work before?) Not tested
  on BSD, though. This uses the old and crap v0.x Linux joystick API.
  Need to add a driver for the new API.
- Lots of of clean-up in input layer. Ported all the joystick drivers
  to the new inputdevice API and thus got rid of some of my hacks
  in input-dummy.c (this will eventually go away all together).
- GTK+ GUI now works on Darwin/MacOS X with SDL threads. POSIX semaphores
  appear to be broken there, however.
- Cleaned up threads targets. Implemented the sleep_millis() function
  in each of the thread.h targets (previously, SDL implementation was
  in sdlgfx.c). Now you can build with SDL threads without SDL gfx.
- Added debug logging to GTK+ GUI.
- Lots of clean-up in GTK+ GUI. Reworked pausing and quitting code.
  Verified that no GTK+ widgets are directly accessed from main UAE
  thread - hopefully clearing up GUI lock-up problems that have been
  reported. Moved more code into gtk_gui_thread() and added a semaphore
  to sync with UAE thread on GUI start-up.
- Started documenting GTK+ code.
- Fixed font problem in About page of GTK+ GUI. The code was changing
  the default font style, not just applying style to widget (although
  this bug didn't seem to effect any of my Linux boxes, it did show up on
  Darwin).
- Fixed sizing policy of GTK+ main window. When the window's vertical
  size was increased, the Snaphots gadgets would expand - which looked
  very ugly. Now they don't.

20031005
- First stab at a GTK message box implementation that is used to report
  certain warnings to the user via a GUI dialog rather than just dumping a
  message on the console. (It's not used much, yet, but is used to report
  problems in loading a Kickstart image, etc.)
- More mucking about with memory.c. Verified that differences with WinUAE
  are either bugs in WinUAE or differences in the way Windows does things.
  Did some more tidying up and started doing some documentation.
- Fixed a bug that would cause a segfault if the ROM keyfile (for encoded
  Cloanto ROMs) was unable to be read.
- Hacked a solution to the problem that most config changes effected in the
  GTK GUI wouldn't be saved out to a config file unless the VM was running
  (to accept the changes).
- Fixed a bug with the new input subsystem and the GTK UI that that meant
  changes in joystick port preferences didn't take effect and couldn't be
  saved out in the config file.
- Added config file support for 8-bit sound back in. 
- Started bringing back some of the snaphost code (still not ready to go yet,
  though).

20031003
- Fixed problem with SDL not cleaning up if UAE dies horribly or is killed.
  Now, when running full-screen and UAE dies, you should get your X display
  back.
- Finished merging the new config-file code from WinUAE (many of the new
  config options aren't actually used yet, though). New filesys config code
  may cause problems with config files from my earlier versions and cause
  duplicate disks to be mounted. Just go into the hard drive section of the
  config and remove any duplicates. (In hindsight, I should have merged all
  the new config code in one go.)
- Disabled the load and save snapshot buttons in the GTK+ UI. The snapshot
  function isn't actually useable yet. I'll re-enable them when it is.
- Merged latest ShapeShifter hack from WinUAE and did some tidying up
  of memory.c.
- Switched a lot of the debugging output in bsdsocket.c to be generated only
  with a compile-time option (cut down the rubbish dumped to the console).

20031001
- Fixed endian problem with P96 screens on SDL on big-endian hardware such as
  the PPC. This is a quick fix and just chooses the right byte-order for pixel
  encoding depending on the host machine's byte-order. A better fix would take
  account of the display's byte-order rather than the host's (similar applies to
  the previous AGA fix). This and a fix for plain X11 target to do. There
  appears to still be an occasional problem with P96 on 32-bit screens where red
  and blue guns are swapped. Possibly a bug somewhere in the blitting code
  in the P96 emulation. Need more time to investigate.
  
20030930 (mistakenly tagged the release as 20030330)
- Fixed endian problem with AGA screens on big-endian hardware such as the
  PPC.
- Fixed bug in the GTK+ UI which stopped the AGA option in the chipset page
  being selected even when it was selected in your config.

20030928-3
- Fixed problem with processor time calculations on the ppc-gcc target. This
  fix will have wide-ranging effects on UAE's performance on the PPC, since
  the processor timer (the PPC's timebase counter) is used to
  synchronize m68k emulation with custom chip emulation. 'Adjustable' cpu mode
  now works.

20030928-2
- Binary only release due to hosed PPC binary. Ooops.

20030928
- First release to the A1g3dev mailing list.
- Loads and loads of changes and fixes to plain old UAE-0.8.22 including:
  * Latest CPU, x86 JIT and custom-chip emulations from WinUAE 0.8.22r9
  * Latest filesystem code from WinUAE 0.8.22r9 (notification on a virtual
    filesystem should now work).
  * New input device code merged from WinUAE. Needs testing and attention.
  * Several SDL fixes, particularly with clean-up and exiting.
  * Fixed 16-bit SDL and OSS sound for big-endian machines
  * Loads of GTK+ UI fixes, particularly with threading and exiting.
  * Fixed bugs in floppy section of GTK+ UI (as a plus the LEDs now work).
  * Replaced hard drive section of the GTK+ UI. A lot more useable.
  * Assorted config and make script tweaks
  * Assorted fixes to clean up code and cut down on compiler warnings.
  * Possibly more that have I forgotten about . . .