ChangeLog.txt

-----------------------
CHANGES IN VERSION 0.50
-----------------------

o Temporarily made F12 the keypad enter key in the Win32 port, just so that
  we can change songs in Nucleus. :) Once 0.50 is out the Win32 adb-drv.c
  will be modified to support all the keypad keys properly.

-------------------------
CHANGES IN VERSION 0.49.3
-------------------------

o Finally fixed the loading of images under the Win32 port and probably
  a few Unix ports as well. I was assuming thet errno was always set to zero
  on a successful call, when in fact it's only meaningful if an error
  occured. The image opening code now only checks errno when it is sure an
  error occured.

o Added support for keypad keys to adb.c. Keypad keys are now represented by
  entries in ski_input_buffer[] that have the high bit set.

o Modified the X11 adb-drv.c to return keypad keys properly according to the
  new adb.c modifications.

-------------------------
CHANGES IN VERSION 0.49.2
-------------------------

o The size of 3.5" tracks are now much closer to what they need to be
  to exactly hold the number of sectors in the track. The new code also
  correctly makes the inner tracks shorter than the outer tracks, just
  like a real drive. This patch makes a lot more programs load from the
  emulated 3.5" drives.

o Corrected a typo in iwm.c in the array that gave the starting block
  number for each track of a 3.5" disk. It was causing the wrong data to
  be read and hence many images simply wouldn't load correctly.

o Implemented support for writing to 3.5" disks. The only thing left is to
  implement support for some sort of nibblized 3.5" disk images (like the
  .NIB images used for 5.25" disks).

o Loading of pre-nibblized images into the 5.25" drives is now supported.
  Writing back to nibblized images still does not work, however.

o Consolidated some of the big local buffers in the 3.5" and 5.25" drive
  routines into shared global buffers.

o Writebacks to 5.25" disks should work once again; I had screwed up the
  initialization of the iwm_from_disk_byte[] table while I was cleaning
  up the code.

o Writebacks to DOS-ordered images should now work. Previously it would
  write the data back in ProODS order.

o .NIB images are now loaded as read-only, because my current writeback
  code would try to write the data back in ProDOS-order format and trash
  the image file.

o I ran through a compile with -Wall passed to gcc and corrected as many
  of the warnings as I could. Those with picky compilers should be somewhat
  happier now. :) 

-------------------------
CHANGES IN VERSION 0.49.1
-------------------------

o Funky version because I want to save 0.50 for a stable release. :)

o The Win32 port sound code is finally included in the source code
  release thanks to David Ross.

o Semi-major overhaul and cleanup of Kent's IWM code in progress.
  Please watch your step and beware of falling code fragments. :)

o Implemented read-only 3.5" hardware emulation and most of the
  framework for full read-write emulation. Mucho thanks to
  Neil Parker for all the info and ROM disassemblies and Ed
  Armstrong for getting the info from Neil and pestering me to
  get the code written. :-)

o 5.25" emulation code now checks for and correctly handles
  DOS-order images. Still no .NIB handling, though.

o IWM code now dynamically allocates and deallocates track
  nib_areas. IWM_update automatically deallocates any tracks
  that aren't the current track as it goes through writing dirty
  tracks to disk.

o Moved nib_pos from Track structure to Disk525 structure in
  the IWM code. Theoretically, sync'd tracks should work now.

o Moved loading of drives at startup from the XXX_init routines
  into EMUL_run().

o Moved head selection for 3.5" drives into the Disk35 struct,
  since I suppose it's theoretically possible to be using
  different heads on both drives at once.
  
o Modified the XXX_loadDrive functions to do their job without
  printing any messages, and created equivalent XXX_unloadDrive
  functions to go with them.

o Moved the emulated SmartPort back to slot 7. Boy, that was
  a short trip...

o The emulated SmartPort now makes sure it only loads images
  of type IMAGE_IF_RAW_PO.

-----------------------
CHANGES IN VERSION 0.49
-----------------------

o Added Ian Schmidt's sound fix. The sound code _Finally_ works
  correctly now. (Some internal variables were being reset when
  oscillators halted themselves, but not when they were halted by
  a program. Thus a program that halted an oscillator would often
  be unable to get it to play correctly again.)

o Moved SmartPort from slot 7 to slot 5, until the day when the
  3.5" IWM code works. The -s5d1 and -s5d2 command line options
  are now equivalent to -smpt0 and -smpt1, respectively.

o Fixed problem found by Michael Bytnar in which VID_redrawScreen()
  could be drawing in the wrong palette. It now forces a total
  redraw by setting a nonsense video mode and calling VID_newMode().

o Applied Frederic Devernay's endianness fix to the super.c drawing
  code.

o Fixed a few more prototypes as suggested by Michael Bytnar.

-----------------------
CHANGES IN VERSION 0.48
-----------------------

o Added DSK_readUnivImageHeader and DSK_writeUnivImageHeader to
  disks.c. These routines read and write a 2IMG header while
  automatically handling all thse messy byte-swapping stuff on
  big-endian platforms. This concentrates most of the #ifdef's into
  these two routines, making the code cleaner.

o Added DSK_swapField, a universal byte order swapper, to disks.c.
  It can swap any field that is an even number of bytes in length.

o Rewrote imgutil.c and imgconv.c to use the routines in disks.c for
  reading/writing 2IMG headers for for doing byte swapping.

o Applied a patch sent to me six months ago to better handle systems
  which don't have random(). I saved the patch in my mailbox and
  it was lost in the mess until I cleaned things out today.

o Added patch by George Ruf [ gruof@pacificnet.net ] to the
  Win32 event loop to catch ALT-F4 and properly shut down the
  emulator.

o Rewrote all of disks.c into a set of generic image-handling
  utilities supporting 2IMG, raw, .DSK/.DO, .PO, .NIB, and
  DiskCopy images. Support for the old proprietary XGS image
  format is gone and the new primary format is 2IMG.

o Rewrote parts of iwm.c and smartport.c to direct all image-file
  access through the routines in disks.c. They should now work
  on any image types supported by disks.c.

o Rewrote imageutil.c to deal with 2IMG files instead of XGS
  image files.

o Added Michael Bytnar's improved super.c that does writes to the
  final video image in 32-bit chunks instead of 8-bit chunks.

o mem-main.c now zeroes out the fast, slow, and save RAM areas at
  emulator boot time using memset(). This seems to cure problems
  that appear when compiling XGS for DOS under DGJPP.

o Added memory deallocation at shutdown to several parts of the
  emulator.

o Consolidated mem-main.c and mem-switches.c into memory.c.

o Folded vid-switches.c into video.c.

o Crunched some of the filenames down to 8.3 format for DOS.

o Added Michael Bytnar's DiskCopy 'unconversion' into the
  image utilities plus the necessary changes to make it work
  with 2IMG instead of XGS images.

o The Win32 VID_outputImage() routine should now correctly handle
  a DD_SURFACELOST error by restoring the surface and refreshing the
  entire display. If the surface cannot be restored the video update
  is silently aborted. This lets you run XGS applications while XGS
  is minimized -- I often do this to listen to synthLAB songs while
  working in another Win95 app. :-)

-----------------------
CHANGES IN VERSION 0.47
-----------------------

o We now sacrifice the border color in SHR mode so we can use
  all 16 colors in all 16 palettes. This fixes the problem with
  opening the Control Panel CDEV in GS/OS.

o Consolidated all the font and image utilities into two
  programs called fontutil and imageutil. The fontutil
  is currently unfinished and is neither built nor installed
  by the standard Makefile.

o Added joystick support to the Win32 version. It works the
  same as the Unix version (ie, uses the mouse).

o Fixed keyboard support in the Win32 port and mapped the
  F3 and F4 keys as Command and Option, respectively.

o Sound data is now passed into the sound drivers as an
  array of structs. These structs contain the raw, un-scaled
  sound data from SND_scanOscillators organized as left,right
  sample pairs. Thus it is possible for the sound driver to
  output the sound in formats other than 8-bit unsigned mono.

o SND_scanOscillators() now produces stereo sound.

o Rewrote EMUL_doVBL() to not depend on gettimeofday(). It
  now uses the EMUL_getCurrentTime() function and does
  all of its calculations in milliseconds. Thus the new
  VBL code _should_ be fully portable.

o Added arch-specific EMUL_getCurrentTime(), which returns
  the current time in milliseconds as a long integer. The
  current time need not be the real time so long as it
  increments 1000 times per second (ie, it can be the
  real time, or the number of ms since boot, or whatever)

o Added initial code to the SVGAlib port. However it is
  not yet functional.

o Moved vid_buffer and vid_lines[] into the arch-specific
  video drivers.

o Fixed all branch instructions to only take the 1 cycle
  cross-page penalty if the branch is actually taken,
  as per the GTE 65816 documentation.

o Re-ordered the cycle time tables in dispatch.c. They
  were in the wrong order and messing up the cycle
  times; this would not have been discovered had it not
  been for the new IWM code, which needs the cycle times
  to be correct.

o Replaced all of iwm.c with Kent Dickey's new code.
  The new code is cycle-accurate and supports writing
  as well as reading of 5.25" disks.

o Added support for multi-platform compiles to
  Makefile.in and configure.in [ devernay@istar.fr ]

o Added ifdef'd include of sys/types.h to mem-main.c
  for the OS/2 Warp port.

o The SmartPort Format and Extended Format commands
  now return an error code of 0 (ie, operation
  completed). This allows images to be formatted
  from GS/OS.

o MVP/MVN now do it the real 65816 way and just keep
  looping the same instruction, instead of using a C
  loop. Otherwise a move instruction could mess up the
  sound because SND_scanOscillators wasn't called often
  enough.

o snd_enable is no longer cleared when the sound
  output device fails to initialize, because on systems
  w/o sound drivers it is still sometimes necessary to
  at least "go through the motions" of sound emulation
  in order to make some programs work.

o Partially implemented the SLOTC3ROM, INTCXROM,
  and ROMBANK switches. XGS ignores them, but if you
  change a switch and read its value back it at least
  returns what you think it should.

o Fixed behavior of the 80STORE switch in relation
  to memory shadowing. 80STORE was always overriding
  AUXRD/AUXWRT for hires page 1, when in fact it
  should only do so when HIRES is set.

-----------------------
CHANGES IN VERSION 0.46
-----------------------

o Removed ENABLE_SOUND. Sound is now on by default
  and can be disabled with the "-nosound" command
  line option.

o Added David Ross's new Win32 user interface. You
  can now change the XGS directory path and even
  select what images to mount before you run XGS.

o SND_updateOscillator now re-enables an oscillator
  (and its partner, if in sync mode) when bit 0 of
  its control register is cleared. Previously this
  was not being done, because I have at least one
  piece of Ensoniq documentation that says it can't
  be done, but apparently it CAN, and lots of programs
  do in fact do this.

  With this fix, the sound now works. It's slow as
  hell (at least on Unix) but it does work. I even
  listened to a couple of tracks in synthLAB...

o SND_scanOscillators now produces correct sound
  data. There were some glitches involving the volume
  scaling and the byte<-->float conversions, and the
  patched routines now scale the summed output sample
  based on the number of oscillators with non-silent
  samples instead of num_osc. With these patches in
  place the sound now sounds like a real IIGS. :-)

o Fixed sound routines to enable the correct number
  of oscillators. [ irsman@iag.net ]

o Added various missing function prototypes.

o Added APP_shutdown to main.c. For most system this
  routine called exit(0) (although for Win32 it does
  ExitThread(0)).

o Moved EMUL_expandPath to arch-dependant main.c file.

o All functions that work with binary files now use
  the "b" option to fopen(). Some ANSI C compilers
  like to translate line terminators if you don't
  do this (most Mac compilers do this).

-----------------------
CHANGES IN VERSION 0.45
-----------------------

o The OSS SND_outputWrite() now makes sure there is
  sufficient space available in the sound driver's
  output buffer before trying to write any data.
  Previously XGS was blocking on sound writes a lot
  and slowing down the emulation.

o Changed EMUL_hardwareUpdate() to run 31,440 times
  per second, making the update period a bit over
  31 ms, which is what is needed for proper Ensoniq
  oscillator updates.

o Added EMUL_expandPath() to expand any partial
  pathnames to include XGS_DIR. All the file loading
  routines in XGS now use this routine on filenames
  before using them, so the chdir(XGS_DIR) in emul.c
  was removed.

  On Unix, a path is partial if it does not begin with
  "." or "/". On MS-DOS, a path is partial if it does
  not begin with "\" and if the second character is not
  a ":" (ie, the path doesn't begin with a drive letter).

o Major overhaul of the Win32 port to move from WinG
  to DirectDraw as the graphics API. It now runs in a
  full-screen 800x600x8 mode and looks pretty cool. :-)

o Moved the joystick and mouse grab keys from F1/F2 to
  F5/F6. F1 is used as the escape key in the Win32 port
  and I wanted the joystick and mouse keys to be
  consistant across ports, so I moved them.

o Moved the main() function out to an arch-dependant file
  (main.c). This is primarily for the Win32 port, which
  doesn't have a main() function (it uses WinMain()
  instead). The startup code is now part of EMUL_init(),
  which no longer calls EMUL_run() on its own.

o Moved all of the code in VID_update() into the arch-
  dependant VID_outputImage() routine. This allows each
  architecture to do special things before and/or after
  the image updates.

o vid_lines[] setup now done by the arch-dependant video
  driver.

o the text and lores drawing routines no longer assume that
  the vid_buffer is contiguous in memory (ie, line n isn't
  necessarily right after line n-1). This allows the code to
  work on more architectures.

o Fixed the COP opcode emulation in opcodes.h. It was
  mangling the processor status register in emulation mode.

o Updated Makefile.in to remove the symlinks into the arch/
  directory when doing a "make distclean".

o Added some shell script to the end of configure.in so that
  the appropriate stuff from arch/ is symlinked into the main
  directory.

o Sprinkled some explicit typecasts around the code to make
  Visual C++ happy when compiling the Win32 port.

o the EMUL_delay() routine has been replaced by the new, more
  accurate EMUL_hardwareUpdate() routine. The new routine uses
  gettimeofday() to try to provide cycle-accurate timing at the
  scan line level. The speed display should be more accurate
  as well since we're no longer depending on signals for the
  timing.

o It is no longer necessary to explicitly set ROM 01/03 mode.
  The proper mode is now determined at runtime from the size
  of the ROM file, which must be either 131,072 or 262,144
  bytes long.

o Major reorganization of the source code, particularly for
  the video, sound, and ADB code. All platform-specific code
  has been cut out and placed in appropriate spots in the
  new arch/ directory (arch/x11 and arch/oss).

o Changed handling of IRQ/NMI/etc in the CPU_run() routine so
  that the generated code (at least on the x86) only performs
  a branch when an exception needs to be handled. This helps a
  bit on processors that favor the no-branch scenario.

o Removed cpu_in_irq flag. The dispatcher now simply checks the
  I bit in the processor status register before dispatching to
  an IRQ. I'm not sure why I wasn't doing it this way to begin
  with.

o Optimized MEM_writeMem2. It was checking explicitly for writes
  to banks $E0 and E1, when in fact these banks are already
  marked with MEM_FLAG_SPECIAL. Just checking for the flag does
  the same thing but does it faster.

o Added a .s target to the Makefile that will let you generate
  the assembler code for any of the C code files. This is
  primarily for my benefit; it lets me easily look at the code
  being generated so I can make optimizations.

o MEM_FLAG_INVALID is now handled by MEM_writeMem2. This means
  the MEM_readMem function is now faster since it isn't doing
  as much flag checking. This speed things up b a few tenths of
  a MHz.

o The mem_pagestruct structure was expanded from 12 to 16 bytes
  by adding two more flag fields which are currently unused. This
  speeds up the emulator by allowing the offset into the array to
  be calculated as a simple shift by four (x16) instead of the
  more complicated x12 calculation. On the x86 for example, this
  saves two instructions for every MEM_readMem or MEM_writeMem
  invocation.

o Added 128K of "scratch" RAM in banks $E8 and $E9. Pressing the
  NMI key will cause banks $00 and $01 to be dumped to banks
  $E8 and $E9. I did this so I could examine program code after
  a reboot (which normally clears bank $00).

o Mapped Ensoniq DOC RAM to bank $EA.

o The beginnings of the "live" Ensoniq code are in place inside
  the sound server. It doesn't quite work yet however.

o Temporarily turned off classic Apple II sound (it never worked
  well anyway).

o Sound code has been removed from the actual emulator and moved
  into a separate xgs-sound process. XGS spawns the sound server
  at startup time and communicates with it via shared memory.
  This greatly simplifies the XGS sound code because the timing
  problems are a bit easier to handle.

o The CPU emulator now correctly sets the B flag in the status
  register for emulation mode BRKs and IRQs.

o IWM emulation had the disk switched status bit inverted, making
  the system think there was always a disk switched error. This
  was causing Standard File dialogs to sit in an infinite loop
  and freeze the system.

o XGS now actually honors the slot register (mostly) by mapping
  out the ROMs for any slots set to "external." The exception is
  slot 7, which is always the emulated SmartPort device.

o Added IRQ counter (number of outstanding IRQ requests remaining)
  to the debug output. This makes it easier to tell if there is
  a problem with clearing a previously raised interrupt.

o Fixed overflow calculations in ADC/SBC _again_. The calculations
  were correct but I wasn't doing them in the right order.
  Mouse dragging in the Finder now works, as does BubbleGhost GS.

o The Data Bank register is now zero'd when a BRK, COP, ABORT,
  NMI, or IRQ occurs in emulation mode. I doubt this bug ever
  caused any problems, but it does make the emulator fit the
  behavior of a real 65816.

o PLB and PLD now set the N and Z bits correctly. Credit for
  finding this bug goes to Kent Dickey [ kentd@hpcuhe.cup.hp.com ]

o MVP and MVN should now work correctly. They no longer check
  to see if the accumulator is 8-bit, and they now copy bytes
  until the accumulator is $FFFF instead of $0000. This fits the
  behavior of a real 65816. The credit for finding this obscure
  bug also goes to Kent Dickey [ kentd@hpcuhe.cup.hp.com ]

-----------------------
CHANGES IN VERSION 0.40
-----------------------

o Updated documentation to describe saving the GS ROM in 16 KB
  pieces instead of 32 KB pieces. The 32 KB chunks must overrun
  a BASIC.SYSTEM buffer or something because it locks the machine
  (though it worked when I first tried it, I have no idea what
  version I used but I must've gotten lucky).

o No drawing is done if the main window isn't mapped. Thus if you
  iconify XGS during processor-intensive programs they'll run
  somewhat faster.  [ devernay@istar.fr ]

o The code in fault.c has been removed and folded directly into
  the MEM_readMem and MEM_writeMem functions. This provides a
  marginal speedup (a couple percent at most).
  [ devernay@istar.fr ]

o More working of the ADB logic. The Desk Manager key sequence
  is reported specially and does not interfere with any SKI data
  transfers currently in progress. This _seems_ to have helped
  smooth out some problems with accessing the control panel from
  GS/OS on a ROM 01, and doesn't seem to have broken ROM 03
  either.

o Put skeleton code in place to trap reads to the Mega II mouse
  registers at $C044 and $C045.

o The default memory size is now 2 MB. I got sick of typing out
  "-ram 2" to get stuff to run. :-)

o Reads to the Mega II horizontal video count now return a random
  horizontal counter value between 0 and 127. This isn't strictly
  correct, but some programs use the horizontal count as a random
  number seed. For example, this fix makes the FTA Nucleus demo
  randomly place its stars as it does on a real IIgs.

o Fixed problems relating to shadowing from bank $01 to $E1 when
  the writes were made to bank $00 with AUXWRT or 80STORE enabled.
  This fix allows the FTA Nucleus demo work.

o The mouse emulation code now moves the X pointer back to the
  center of the window after every pointer movement event. This
  fixes the problem where the X pointer would stop at the end of
  the window before the GS's pointer was at the edge of the
  screen. [ devernay@istar.fr ]

o Fixed overflow flag emulation in the ADC/SBC instructions
  (I think...) BubbleGhost GS almost works now.

o Fixed bug in XBA opcode emulation. It should have set the N and
  Z flags to reflect the final state of the low byte of the
  accumulator, but it wasn't. This fixes the problem with
  launching P8 applications from P16 and GS/OS. Thanks to
  Andre Horstmann [andre.horstmann@unifr.ch] for pointing this
  one out.

o Since USS is now known as OSS (Open Sound System), I changed
  the code to reflect this. This affects the configure script
  (--with-oss-sound) and sound.c/sound.h (the USS_SOUND define is
  now OSS_SOUND)

o Fixed ADB_readCommand() bug where the clearing of pending ADB
  interrupts would cause the SKI emulation to lose track of the
  number of bytes it had returned so far. This bug caused ROM 01
  to not work.

o Implemented code to accelerate applications which do page flipping
  (text, lores/double-lores and hires/double-hires).  Only areas
  which are actually different than the last page are redrawn now.
  [ devernay@istar.fr ]

o The window is now correctly redrawn in response to Expose events.

o Rewrote IRQ emulation to support multiple pending IRQs that must
  be cleared by the device which generated the interrupt (like on a
  real IIgs). Previouly interrupts were lost if multiple interrupts
  were generated at once (which happened a lot, since XGS tends to
  only generate IRQs at defined times). For example, you can now
  get into the Desk Manager menu from GS/OS (mouse interrupts used
  to prevent this).

o Scanline interrupts now work right. The CPU update period was
  being mangled by the EMUL_update() procedure, causing scanline
  interrupts to occur about once a second instead of every 64
  emulated cycles. As a result of this, the mouse FINALLY works in
  16-bit apps!

o The 80STORE switch now only takes precedance over RAMRD/RAMWRT in
  the $00/2000 - $3FFF range when the HIRES switch is on. Previously
  it always took precedance, causing some programs to access the
  wrong memory and crash.

o Implemented double high resolution graphics. It's not 100% working
  yet but it's almost there.

o Made some semi-major changes throughtout all the code to allow XGS
  to compile and run correctly on Alpha Linux (and, presumably,
  other 64-bit operating systems).

o Video code now only sends updates to the X server for the areas of
  the screen which have actually changed. The result is a rather
  signifigant speedup in many applications. [ devernay@istar.fr ]

o X11 support more robust now; it handles all the events a basic X11
  application should be handling (such as Expose events).
  [ devernay@istar.fr ]

o Implemented pointer grabbing for the mouse and joystick. F1 grabs/
  ungrabs the pointer for joystick emulation, and F2 does the same
  for mouse emulation. Only one can be active at a time.

o Much more robust MIT-SHM support added. It now queries the MIT-SHM
  extension before using shared memory, and will default back to the
  standard XPutImage() method if shared memory is not present or not
  fully supported. [ devernay@istar.fr ]

o Makefile reorganized a bit. Should be 99% friendly to non-GNU make
  now. [ devernay@istar.fr ]

o You should now be able to to truly compile XGS without USS support
  now (a few more #ifdef's were needed). [ devernay@istar.fr ]

o Removed line in MEM_setSlotReg() that was always forcing slots 5-7 to
  "external." This code is no longer necessary since slots 5-6 are now
  properly emulated at the hardware level. With this code removed, the
  3.5" BitCopy in Copy II Plus now loads and runs just fine.

o Fixed some subtle bugs in the CPU emulation dealing with operand
  address calculations. Under certain conditions data could be read from
  random banks instead of the correct ones.

o The sound GLU emulation now correctly auto-increments the address
  registers. There's still no sound generation code, but some programs
  would hang waiting for the address to be auto-incremented to a certain
  point (such as Hardball GS)

o Implemented the horizontal and vertical video counters. They are
  nowhere near accurate, but they do satisfy programs which watch for
  certain counter values.

o TCD now sets the N and Z processor flags. This was the last bug that
  prevented GS/OS System Disk 6.0.1 from booting.

o MVN/MVP now loop internally instead of through the opcode dispatcher,
  so memory moves run MUCH faster than any real 65816.

o Corrected minor bug in language card emulation.

-----------------------
CHANGES IN VERSION 0.30
-----------------------

o In the interest of speed and simplicity, XGS now supports _only_ 8-bit
  X servers. The new video code is now optimized for that bit depth, and
  takes advantage of the palette-changing ability of that mode to handle
  things like text color changes and super high res palette changes.

o ADB mouse hardware is now emulated (accessed through $C024 and $C027)

o The joystick hardware is now emulated. Currently, the joystick just follows
  the mouse.

o Wrote "raw2img" for converting raw disk dumps ("dd" output, or ProDOS-
  ordered .PO images) into XGS image files.

o Wrote "dsk2img" for converting DOS 3.3 format .DO/.DSK 5.25" images
  into XGS image files. This utility is useful for converting all those
  images on ftp.asimov.net into XGS images.

o Slots 5 and 6 are now emulated at the hardware level using the new IWM
  emulation code. Unfortunately it's still read-only, and slot 5 doesn't
  work at all yet.

o Slot 7 is now an Extended SmartPort device supporting up to 14 images
  online.

o Fixed a bug in the PAGE2/80STORE switch emulation that prevented some
  128K programs (such as ShrinkIT 3.4) from loading.

o That damn clock works, finally. Battery RAM emulation _seems_ to work, but
  the control panel still won't allow any settings to be changed. Hmm.

o Reverted back to the old subroutine-based CPU emulator. It turns out it's
  faster anyway (probably because the old one can't be compiled by GCC with
  optimization on).

o Endianness problem with super high res color has been fixed.

o The CPU emulator now counts cycles. It is (hopefully) an accurate count.

o VBLs are now emulated using a 100 Hz interval timer, with the VBL being
  triggered at half that frequency (50 Hz).

o XGS's window now includes a couple of lines of status information, updated
  once per second. A lot of it is for my debugging, but it does display a
  couple of useful items: namely the target speed the emulator would like to
  achieve (1.0 or 2.5 MHz, depending on the CYA speed register setting), and
  the actual speed it is achieving. My P166 benches in the 2.0 - 2.7 MHz
  range, depending on what I'm doing in XGS at the time.

o Memory map handling substantially rewritten. It now subdivides _all_ memory
  into pages, instead of just banks $00/$01/$E0/$E1. Fault handlers are gone;
  I/O space and shadowing are the only special cases and they are handled via
  bit flags in the memory page structures. Also I keep an array of long ints
  which mark which bytes in the slow RAM banks have been modified; each bit
  represents eight bytes (which is nice because then one long word = one page
  of 256 bytes).

o Shakey but sorta working support for classic Apple II sound. It sounds a bit
  distorted, and pops a lot, but it does work.

o The video code now picks the correct visual to use for the chosen display.
  This doesn't seem to matter on XFree86 systems, but it apparently does on
  others. XGS should now refuse to start up if it can't find a suitable visual.

o The video code now only rescans the areas of video memory that have actually
  changed since the last video update. This means that a time spent in video
  updates has been substantially reduced.

o Video updating is now done on 50 times per second instead of 4 times per
  second, because the new video code doesn't slow the emulator down as much.
  The result of this change is that video updates are much smoother.

o The source tree has been substantially reorganized. Everything is in
  one directory to make things a lot easier to work with, and the whole
  thing is now configurable with GNU autoconf. There's even a real
  Makefile now, with proper dependancies.

-----------------------
CHANGES IN VERSION 0.20
-----------------------

o Due to popular demand, 8-bit color X servers are now supported.

o XGS _should_ work now if you give it a ROM 01 image padded to 256 KB.

o Super high-resolution graphics supported though not fully tested. It _should_
  work in all modes including 640x200, 320x200, and 320x200 fill mode.

o Created virtual ProDOS block devices in slots 5 and 7. Each can support two
  disk image files up to 32 MB each. This means it's now possible to _boot_
  stuff on XGS (at least ProDOS 8 stuff). You can even write to the virtual
  disks!

o The arrow keys are now recognized by the emulator.

o You can now enter the Desk Manager menu (Control-Command-Escape), though
 it's still acting a bit flakey sometimes.

o Added the utils/ subdirectory. There's lots of neat stuff here for handling
  the XGS ".img" files and for viewing and rebuilding the XGS font files.

o In all bit depths, we now open a private colormap and allocate the colors
  using XAllocColor(). This was necessary for 8-bit mode, and for the other
  modes it should fix the problem XGS had on X servers that specified the RGB
  color levels in a different order than what XGS was using.

o Video handling substantially overhauled. The emulator no longer traps video
  memory accesses; instead, the video screen is simply periodically redrawn.
  Some things run quite a bit faster as a result.

o High-resolution graphics glitch fixed; the right border is no longer
  partially overwritten.

o Mixed text/graphics modes now work in 40 or 80 columns.

o Fixed some of the more annoying font glitches. The mousetext 'Z' character
  is now a right-justified vertical line as it should be, and the "1" and "l"
  (lowercase L) characters are now centered in the font matrix instead of
  being left-justified.

o The Makefiles have been modified to work on non-GNU versions of make and ar.

o Reorganized the link commands in the Makefile to specifiy the libraries last
  on the command line. It worked fine for me the old way but not for some other
  people.

o Some key symbols specific to X11R6 were replaced with symbols present on all
  versions of X11. This should allow XGS to compile on systems that use X11R5.

o 65816 emulation should be a faster now. Also some bugs with IRQs and NMIs
  were fixed (they weren't pushing the proper PC value onto the stack, plus
  the RTI instruction was incorrectly incrementing the value it pulled back
  from the stack. Used together they work fine, but some code in ProDOS 8
  apparently sets up a fake IRQ stack frame and does an RTI, and this code
  didn't work with the old 65816 emulation).

o Skeletal support for emulating the Ensoniq DOC has been added. It doesn't
  really _do_ anything yet other than allow reads and writes to the DOC and
  sound RAM (basically just enough to get the darn thing to work properly).

o VBL and quarter second interrupts are now supported...I think. Haven't fully
  tested them so I can't be 100% sure.