path: root/ChangeLog

4.9c -- Sun May 14 17:54:25 PDT 2006 -- Tim Mann

* Fixed the new -e flag on import/cmd and export/cmd to actually
work.  Oops.  Tested on Model I DOSPLUS 3.5.

4.9b -- Sat May 13 11:54:26 PDT 2006 -- Tim Mann

* Bundled in the latest version of Roland Gerlach's CP/M utilties for
xtrs and their documentation (as of 5/13/2006).

* Added -emtsafe feature to turn off ability for emts to write to
unexpected places in the host filesystem.  Thanks to Joe Peterson.

* Added -e flag to import/cmd and export/cmd.  Dropped DOSPLUS patches
for those programs, which were outdated anyway.

* Dropped old method of importing/exporting files using fake I/O
ports and deleted import/bas and export/bas.

* Small cleanups in load_cmd.c.

* Renamed Z-80 source files to .z80 instead of .z to prevent Linux
utilities from thinking they're compressed files.

* Moved key queuing code from trs_interface.c to trs_keyboard.c and
cleaned it up a little.

* Applied patches from Branden Robinson to add watchpoints and the
zbxinfo command to the debugger and clean up a few things.  Thanks!

* Added some signed/unsigned type casts; thanks to Andreas Jochens.

4.9a -- Mon Apr 25 22:35:51 PDT 2005 -- Tim Mann

* Changed the DISKDEBUG options in trs_disk.c to be runtime flag bits
instead of #ifdefs.  Added diskdebug command in zbx to set the bits,
with terse documentation in zbx's help command.  2 (FDC commands) is
the most useful.  1 (FDC register I/O) is extremely verbose but
sometimes useful.

* Added code to export.cmd and import.cmd to ignore the TRSDOS/LDOS 6
advisory error code when we open a file with the "wrong" LRL.

* Fixed a typo in the usage message for mkdisk.

* Increased the maximum accepted track length for DMK to the hard
limit of 0x4000.  Previously it had been 0x2940, but cw2dmk is writing
longer tracks than that by default for 8" disks now!  Oops.

* Small cleanup patches from Branden Robinson for cassette.sh.

4.9 -- Sat May 11 13:42:06 PDT 2002  Tim Mann

This release has various bug fixes and tiny features that have
accumulated since the last one.

* Added a hint about floppy drive BIOS settings to the man page.

* Fixed a bug in the cassette shell script.  The cassette image type
wasn't being determined from the file extension due to a shell
programming error.

* Got rid of #ifdef HRG1B" and just put all the code for that in-line.
Moved documentation to where most of the implementation is.

* Added -iconic option.  Thanks to Jean-Marc Le Peuvedic.

* Bugfix: check_endian would crash if it failed; it was calling
fatal() before program_name was set.  Thanks to Sot Aliadis for the
bug report.

* Bugfix: the SIGUSR1 feature wasn't turning off the needchange flag
after doing the disk change, so it would do the change over and over.
Thanks to Jean-Marc Le Peuvedic for the bug report.

* Bugfix: The change that was made in version 4.6 to get the 500bps
loader that came on a 250bps tape to work turned out to break writing
1500bps tapes.  The tape-based 500bps loader needed a special
heuristic, because it does a write just after turning on the motor but
before its first read.  But the 1500bps writer does a read after all
its writes but before turning off the motor.  This was erroneously
activating the heuristic for the loader case, causing us to switch to
read mode without flushing the last transition from the write.  The
heuristic is now changed so that it won't activate if more than one
write has been done since the motor was turned on.  This makes both
cases work.

* Changed the sound code to be less picky about sample format on
/dev/dsp.  Needed because some hardware doesn't do the 8-bit formats.
We still prefer to use 8-bit mono for game sound and 8-bit stereo for
Orchestra-85/90, but we can also use 16-bit signed little endian for
either, and stereo when we want mono (but not vice versa).

* Fixed a bug in FDC emulation that was stopping CopyCat from booting.
After starting a read, the CopyCat boot loader reads from the data
register once immediately, before DRQ has come high.  This read must
not capture the first byte of data from the sector.

* Fixed a bug in FDC emulation that was keeping CopyCat from seeing
sectors on a track other than the first.  When client software used
the Model III/4/4P feature that lets you put the Z-80 into a wait
state until the FDC raises DRQ, the amount of time waited was not
being added to the T-state counter.  Thus repeated Read Address
commands that used the wait state feature would read the same sector
ID over and over instead of rotating to the next one.

* The change in version 4.2 to wait a few T-states after a floppy Read
or Read Address command before raising DRQ was only implemented for
floppy image files.  It's now implemented for real floppies too.

* The change in version 4.2 to set the track and sector registers on a
Read Address to the track and sector numbers read from the ID field
was implemented only for real floppies.  It's now implemented for
floppy image files too.

* A few minor cleanups to floppy emulation code.

* Bugfix: When a real Z-80 takes an interrupt or NMI from a halt
instruction, it pushes the address of the next instruction after the
halt as the return address (and later returns there, thus breaking out
of the halt).  Previously xtrs was erroneously pushing the address of
the halt instruction instead.  If no interrupt or NMI is taken, xtrs
still repeatedly executes the halt instruction; you can see this if
you use zbx to single-step or trace a halt.  A real Z-80 supposedly
executes nops internally instead of re-executing the halt, but the
difference is not visible to Z-80 programs running in xtrs.

4.8 -- Fri Aug 31 11:39:49 PDT 2001  Tim Mann

* This release has quick fixes for some troublesome bugs.  Better
fixes may be forthcoming in a later version.  Thanks to Jean-Marc Le
Peuvedic and Frederic Vecoven for problem reports and suggestions.

* Changed my email address, as I have left Compaq.

* Bug fix: Output to the X server was not always being flushed in a
timely manner.  It is now flushed at the frequency of the timer
interrupt on the machine being emulated (30 to 60 times per second).
This fixes problems with the display not being updated when it should
have been in certain games and operating systems.

* Added a heuristic to flush the key transition queue if the Z-80
program does not poll the keyboard often enough.  Right now this is
set to go off if more than 2 timer interrupt ticks go by without a
poll.  This change should get rid of the unwanted typeahead effect
previously seen in certain situations, such as when a game does some
screen flashing or sound effects between levels without polling the
keyboard.

* Fixed a compile error on non-Linux platforms.

4.7 -- Sun Jul  8 16:37:14 PDT 2001  Tim Mann

* Fixed a silly bug that caused a crash if no -diskdir was specified.
I had inadvertently deleted the initialization.

4.6 -- Sat Jul  7 13:56:33 PDT 2001  Tim Mann

* The value of -diskdir can now begin with "~/" (or just be "~") to
make it relative to the user's home directory.  Suggested by Jean-Marc
Le Peuvedic.

* Fixed the heuristic that decides whether the TRS-80 is loading or
saving to tape to work with the 500bps loader that came on a 250bps
cassette with the tape Editor/Assembler.  This code does an OUT to the
cassette port first thing after turning on the motor, which was
confusing us.  Now if the TRS-80 does an IN from the cassette port
with the motor on, we switch to reading even if we initially saw an
OUT and guessed it was writing.

* Added feature: sending SIGUSR1 to xtrs causes a disk change (as if
F7 were pressed).  This is for use by external disk-manipulating
programs.  Suggested by Jean-Marc Le Peuvedic.

* Changed to a new keystretch algorithm that stretches for a fixed
number of T-states instead of a fixed number of keyboard matrix reads.
This should hopefully eliminate the need to tweak the keystretch
setting to different values for games.  Inspired by suggestions from
Jean-Marc Le Peuvedic.  To make this work, also had to change the
KBWAIT algorithm to wait only after the ROM wait-for-key routine has
been called several times with no keyboard state change.

* I noticed that the X server on my development machine is now
reporting Shift+F1 as F13 instead of F11, so I changed the table in
trs_keyboard.c to match, but I left an #if setting that can
accommodate the old behavior.  It looks like newer versions of XFree86
don't do either translation, thankfully.

4.5a -- Fri May  4 23:17:46 PDT 2001  Tim Mann

* Corrected some typos in xtrs.man and Makefile.  Thanks to Branden
Robinson.

4.5 -- Wed May  2 20:48:10 PDT 2001  Tim Mann

* All undocumented instructions and almost all undocumented flag
settings from http://www.msxnet.org/tech/Z80/z80undoc.txt are now
emulated.

* Added code to parse resources from $HOME/.Xdefaults, $HOME/Xtrs, and
/usr/X11/lib/X11/app-defaults/Xtrs.  Normally Xt would handle this,
but xtrs doesn't use Xt.  Branden Robinson noted the absence of this
feature back in 1998; it's finally fixed.

* Added -shiftbracket command line option, inspired by mail from
Jean-Marc Le Peuvedic.  Defaults to true on Model 4/4P.

* Added man pages for hex2cmd and cmddump; thanks to Branden Robinson.

* Raised JV1/JV3 cylinder limit from 96 to 255 after someone sent me a
JV1 with 143 cylinders.  I wonder which DOS that was made with!

* Fixed syntax errors in debug.c that gcc had been allowing through.

* Corrected some man page errors and unclear passages.

4.4 -- Fri Feb 16 23:52:32 PST 2001  Tim Mann

* Added some DOSPLUS patches for import/cmd and export/cmd from
Knut Roll-Lund.

* Removed kludge from MSM5832 clock chip emulation; now it provides a
plain 2-digit year (i.e., "01" for 2001) as the real hardware did.

* Cleaned up a bit and eliminated C compiler warnings.

* Added break command to debugger as a synonym for "stop at".

* Modified the debugger's reset and run commands, and added softreset.
Now reset does a powerup initialization of the Z-80 and all devices,
while softreset just pushes the TRS-80's reset button.  The run
command does a hard reset and starts execution at address 0000.  On
the Model I and III, softreset resets external devices and gives the
Z-80 a nonmaskable interrupt, but on the Model 4 and 4P, it is the
same as reset.

* Added emulation of the Alpha Products joystick, mapped at port 0
only.  For the moment it doesn't work with a real joystick; instead,
it uses the numeric keypad keys with NumLock off.  Keys 2, 4, 6, 8
(KP_Down, KP_Left, KP_Right, KP_Up) are the main directions; keys 1,
3, 7, and 9 (KP_End, KP_Page_Down, KP_Home, KP_Page_Up) work as
diagonal directions by activating two main directions at once; and key
0 (KP_Insert) or 5 (KP_Begin) is the fire button.  At some point I may
acquire a real PC joystick, read about the X input extension, and make
that work too.

4.3 -- Fri Jul 21 20:44:48 PDT 2000  Tim Mann

* Added emulation of the serial port for all models.  Note: Model I
LDOS 5.3.1 has a bug in the RS232 driver (RS232R/DVR).  You must set
the break parameter to a nonzero value for it to work.  Try "SET *CL
RS232R (BREAK=1)", for example.

* Added emulation of the Radio Shack graphics card for Model III.
This is just like the Model 4 card, except that there is no overlay
mode, there are no scrolling registers, and the graphics are always
640x240 (never 512x192).  Changed default graphics card from
-microlabs to -nomicrolabs.  Changed the -resize flag to be meaningful
in Model III mode, and changed the default to be off for Model III
mode, but still on for Model 4/4P mode.

* Added emulation of the Micro Labs graphics card for Model III.  This
is quite different from the Model 4 card.  I deduced how it works by
looking at what the software for it was trying to do, so there could
be bugs in the emulation.

* Enable the hard disk controller at powerup.  NEWDOS is said to need this.

* Minor changes to compile on Solaris.  Thanks to Frederic Vecoven.
	
* Bugfix: The MSM5832 clock chip returns a leap year indication in
bit 2 of port base+8.

* Bugfix: When running an emulated model 4/4P in model 3 mode, characters
were being drawn only 10 scaled pixels high instead of 12.

* Added a unified debug() routine to print all debugging messages.
	
* Patch to fix zbx listing up to 0xffff, from Ulrich Mueller.

* Patch to correct disassembly of DD and FD prefixes that have no effect,
from Ulrich Mueller.

* Patches to remove unneeded cast and fix indentation, from Ulrich Mueller.

4.2 -- Sat Jun 24 01:05:13 PDT 2000  Tim Mann

* Emulation of the Model I HRG1B 384*192 graphics card, sold in
Germany for Model I and Video Genie by RB-Elektronik.  Contributed by
Ulrich Mueller.
	
* Fixed a bug that would crash xtrs if the Z-80 SP pointed into
keyboard memory.  Contributed by Ulrich Mueller.

* The -scale command line option now allows independent scaling in the
X and Y dimensions.  The scaling mechanism is also used to eliminate
the need for the bitmaps in trs_chars.c to be double-height, and to
eliminate the need for the separate double-width bitmaps formerly in
trs_wide_chars.c.  The German characters in trs_chars.c were corrected
to make this work.  Contributed by Ulrich Mueller.

* Fixed an obscure bug in the heuristic to identify whether a disk
image is JV1, JV3, or DMK.

* In trs_disk.c, reading a sector now waits a while before raising
DRQ.  The old behavior broke the boot loader for self-booting Zork.

* In trs_disk.c, Read Address now sets the track register to the track
number read and the sector register to the sector number read.  The
1771 data sheet says that Read Address sets the sector register to the
sector number, but nothing about the track register.  The 179x data
sheet says that Read Address sets the sector register to the track
number, and nothing about the track register, which seems wrong.  "The
Institute" game loader assumes that Read Address will set the track
register to the track number on a 1771.

* Minor accuracy tweaks to trs_hard.[ch].

4.1 -- Fri May 19 00:40:29 PDT 2000  Tim Mann
	
* Fixed a Y2K problem in settime/cmd (as assembled from settime.z).
For some random reason, Model I/III LDOS 5.3.1 stores a 2-digit year,
so 2000 => 0, but Model 4 LS-DOS 6.3.1 stores an offset from 1900,
so 2000 => 100.  Previously we were getting it right for 5.3.1 but
wrong for 6.3.1.
	
* Fixed a bug in the floppy drive motor timeout interrupt.  It was not
being cancelled immediately when a drive was reselected.  This
sometimes made the floppy driver in the Montezuma CP/M boot sector
fail, and could have caused problems in other software too.

* Now only Models I/III/4, not Model 4P, are emulated as having no
floppy controller when they have no disks mounted.  A Model 4P always
has an FDC, and it can boot off the hard drive with suitable software
(such as Montezuma CP/M), so it is possible to boot an operating
system with no floppies mounted, then mount and access some later.
This isn't possible on the other models.

* Bugfix: If a JV3 or DMK floppy was not writable, but did not have
its internal write-protected flag set, we were not emulating a write
protect tab as intended.

* F9 will now get you into the debugger (zbx) even if you didn't start
xtrs with the -debug flag.  Also, starting with -debug no longer
causes the emulator to run more slowly; it runs at full speed as long
as no traps are set.

* Formerly, executing a halt instruction caused entry into the
debugger (zbx) if xtrs had been started with the -debug flag.  This
feature has been removed.  As a replacement, executing the new
emt_debug emulator trap (ed2f) always causes the program to enter the
debugger.

* Fixed the debugger reset and run commands to reset the whole
machine, not just the processor.

* Installed a small patch that makes the emulator control its window
size in the correct way, by passing a size hint to the window manager
instead of noticing when its size has changed and changing it back.
Thanks to Joseph Wang.

* Added emulation of real Radio Shack hard drive controllers (WD1010).

* Changed xtrshard/dct to set the directory cylinder field in the Reed
header automagically.  (Heuristic: when a byte is written to cylinder
0, sector 0, byte 2, assume that's the right value.)  The WD1010
emulation does this too.  There is now no need for mkdisk to have a
flag to set this field, or for the user to remember to use the right
value with FORMAT, so that stuff is deleted from the man pages.

* Deleted references to cassette "autodetect" format, since I'm not
going to get around to implementing it.  The current autodetection
in the cassette shell script based on file extension works well
enough.  Rephrased the caveats about reading and writing real
cassettes, since I just tried it again and got fairly good results
this time.

4.0 -- Sat Apr  1 20:52:53 PST 2000  Tim Mann

* Fixed bugs in switching between 80x24 and 64x16 screen sizes in
-usefont mode.  Thanks to Al Petrofsky for noticing back in 1998 that
the code looked wrong; I finally got around to cleaning it up for 4.0
and found the bugs.

* Worked around an X problem.  If you start xtrs with the -resize flag
and a Model 4 operating system is already mounted, the window gets
resized very soon after being created.  This was triggering some
sort of X race condition that would make the window get confused about
how big it was.

* Fixed obscure bugs in the Model 4 hi-res graphics card emulation.

* Added -scale option.  Thanks to Denis Leconte for some of the code.
	
* Emulate software-switchable 3.54Mhz clock speedup kit on Model I.
Useful with Orchestra-85 emulation.
	
* Changed -samplerate default to 44,100 for better sound.

* Added emulation of Orchestra-85 and Orchestra-90.  Thanks to David
Keil for some information on how Orchestra-90 works.

* Fixed a bug that sometimes caused the F10 reboot key to hang the
emulator until the next X event.

* Added emulation of TimeDate80, TChron1, TRSWatch, and T-Timer, based
on code from Joe Peterson and information in LDOS Quarterly 1-6.  I'm
not sure it's completely right; see the comments in trs_io.c.  If you
have software that uses these devices, let me know how well it works,
especially with regard to the Y2K issue.

* Added support for Level 1 Basic .cas files.

* Worked around a problem that kept Model I Level 2 Basic from being
able to read .cas files.  When Basic records a 500bps cassette, it
puts an extra delay after the initial 0xA5 sync byte.  This delay is
required when reading a cassette because Basic executes quite a bit of
code right after reading the sync byte.  As a kludge, I put in an extra
delay whenever a 0xA5 is read from a 500bps cassette; ugly, but it
works.

* Updated documentation for EXPORT.COM in xtrs.man; this change should
have been in 3.9.

3.9a -- Sun Nov 28 14:17:49 PST 1999  Tim Mann

* Fixed a bug in the GENIE support and added one more GENIE feature:
printer interface at port 0xFD.

3.9 -- Fri Nov 26 11:07:24 PST 1999  Tim Mann

* Fixed a problem in the sound support which (for reasons I don't
fully understand) was making xtrs hang when playing sound on Linux
kernel 2.2.12.  Thanks to Roland Gerlach for the bug report.  Version
3.8 of xtrs worked on kernel 2.2.5, but not 2.2.12.  Also fixed a bug
where too many samples could be generated if the Z-80 program
repeatedly wrote the same value to the sound/cassette port; this makes
the CP/M bell sound much better, and also seems to fix the remaining
problem with the Talking Adventure 1 title screen.

* Corrected a small error in emulating Model I floppy disk controller
interrupts (a feature that's unused by any known software anyway!).

* Installed a pile of changes from Branden Robinson, which he says are
"mostly to satisfy the vagaries of Debian's packaging policy."  They
are as follows:

  * main.c: canonicalize program_name here, not in trs_xinterface.c, so
    that error messages are consistent
  * Makefile:
    - clean rule deletes compile_rom binary, and keeps deroffed manpages
    - veryclean rule deletes deroffed manpages
    - use Debian-canonical file modes when installing
    - install cmddump, hex2cmd, and mkdisk
  * Makefile.local:
    - compile-in expected paths for ROM files
    - build using readline and ncurses (for zbx internal debugger)
    - Debian-canonical flags to cc
    - LDFLAGS and IFLAGS that make sense with our X directories
    - create a PREFIX variable that gets prepended to BINDIR and MANDIR
  * trs_xinterface.c:
    - remove canonicalization of program_name, this is now in main.c
    - change logic for finding romfiles; check command line options, X
      resources, compiled-in path to rom files, then finally fall back on a
      built-in ROM image
  * cassette.sh: rewrote csh cassette script in Bourne
	
I undid or modified a few of Branden's changes: I left the old
cassette script as the default since I'm not confident that every Unix
system has a new enough version of /bin/sh to support cassette.sh.  I
left -Wall out of the C flags, since I see no reason to worry users
with the small number of harmless warnings I've left in the code.  I
put the default installation locations back to places that are
appropriate for add-on code in most non-Debian systems: /usr/local/bin
and /usr/local/man.

* Merged in support for the GENIE, a German Model I clone, from Jenz
Guenther.  I simplified it a bit along the way; I hope I haven't left
out anything essential.

* Fetched updated cpmutil.dsk from Roland Gerlach's Web page,
http://www.rkga.com.au/~roland/trs-80/cpm-xtrs/.

3.8 -- Sat Jul 17 19:25:57 PDT 1999  Tim Mann

* Made several tweaks to the sound support.  The title screen of
"Talking Adventure 1: Forbidden Planet" for Model III now works
somewhat better, at least at -samplerate 44100, but there still seems
to be a problem.  Also, it still looks like the speed boost when a
sound begins (trs_suspend_delay) may be making games run too fast when
a sound plays.

* Lost Data is now set if a read or write is not completed in 500ms.
This is not an exact emulation, but is enough to make DOSPLUS 3.3 and
Floppy Doctor work.
	
* Seeking and stepping slowed down a little to make Floppy Doctor
notice that the Busy bit does get set.

* Head Engaged bit emulated correctly now.  Was needed for Talking
Adventure loader (!).

* DMK writing emulation more accurate, handles case of writing to a
sector that was formatted with no DAM.  Thanks to David Keil.

* zbx "dump" command prints T-state counter and delay value.
"diskdump" prints T-states to next motor timeout, not T-state counter
value at timeout.

3.7 -- Sat Jul  3 15:57:04 PDT 1999  Tim Mann

* Completely rewrote the cassette support.  Instead of trapping the
Model I ROM routines, it now observes I/O requests to the cassette
port and uses xtrs's T-state counter to measure (or generate) the
correct signal timing.  Real cassettes can even be read/written using
the OSS sound driver or .wav files, though this could use some work.

* Completely rewrote the sound support to use the OSS sound driver.
Most code is shared with the cassette support.  The old code is still
there too, but #if'ed out.
	
* Modified import/cmd and export/cmd to work on Model III TRSDOS 1.3.
Tested and verified that setdate/cmd works there too.

* Recompiled cd/ccc, mount/ccc, pwd/ccc, truedam/ccc, umount/ccc, and
unix/ccc with the LDOS I/O redirection option turned off.  This makes
them work on Model I/III TRSDOS and NEWDOS too.

* Added -l (force lowercase) to cd/cmd, mount/cmd, and unix/cmd.
Useful with Model I/III TRSDOS and NEWDOS.

* Fixed a bug in printing exit status from unix/cmd.

* Added -p/-u flags to mkdisk for turning write protection on/off.

* Added display of IFF1, IFF2, IM, and undocumented F register bits to
the zbx dump command.

* Fixed a 1-byte memory smash in emt_opendisk.

* Fixed a bug that could crash xtrs when accessing an empty real
floppy drive.
	
* Made error() and fatal() more capable and cleaned up some stray
places where fprintf(stderr, ...) was being used instead.

3.6 -- Sun Jun 20 16:13:05 PDT 1999  Tim Mann

* Implemented the DMK emulated floppy format.  This format supports
essentially everything the original hardware could do, including
"protected" disk formats, non-IBM sectors on the Model I, partially
reformatting a track to achieve mixed density, Read Track, and more.
It is compatible with David Keil's Model 4 emulator; thanks to David
for designing and documenting it.

* Added support for the Motor Off NMI on Model III/4.

* Added emulated CRCs on Read Address.

* Added diskdump command to zbx; dumps the state of the emulated
floppy disk controller and drives.

3.5 -- Sun May 16 21:55:38 PDT 1999  Tim Mann

* Changed physical disk handling to report the correct WD17xx error
bits upon reading (or writing) a sector with a CRC error in the ID
field.  In this case both CRC and RNF bits should be set.  Fixing this
lets us boot physical Scott Adams Adventure disks that use the Kim
Watt loader.  Unfortunately, the WD17xx behavior with "read address"
on such a sector cannot be emulated, because a 765-style FDC's "read
id" command ignores IDs with CRC errors rather than returning an error
when it sees one.  Also, of course, .DSK format has no way to
represent a sector with a CRC error in the ID field.
	
* Always use a recalibrate if seeking a real disk to track 0.  This
should help us recover from confusion about what track the disk is
really on.  Such confusion should not arise unless two copies of xtrs
are both seeking the disk, or the disk is mapped twice to the same
xtrs and both virtual drives seek -- but there are sometimes reasons
for doing that.
	
* Updated xtrsrom4p.README file with material inadvertently left out
in 3.4 release.
	
3.4 -- Sun Apr  4 22:27:46 PDT 1999  Tim Mann

* Corrected the Model 4 reset switch to send RESET instead of NMI.
Thanks to Lamar Owen for pointing out the error.
	
* Corrected the default charset for Model III to be katakana.  Thanks
to Gary Shanafelt and others for pointing out the error.

* Fixed minor bugs in cassette support.  F10 (Reset button) now gets
you out of cloading something that isn't there, even if you aren't
using the SIGIO support.  Also, autodelay measurement is turned off
while cloading.

* Added a free minimal Model 4P ROM, written by Pete Cervasio (with a
few hacks of my own).  Thanks, Pete!  The free ROM can boot only Model
4 mode operating systems; it doesn't know how to load MODELA/III.
	
3.3 -- Sat Jan 23 15:42:36 PST 1999  Tim Mann

* Added an emt_misc code to query/set the -truedam flag, and added a
Misosys C program to invoke it.  Made -truedam more rigorous for real
floppies, and clarified (I hope) DAM documentation in the man page.

* Much improved behavior when accessing a real floppy drive with no
disk in it.  Previously this would generate a raft of error messages
to stderr, subsequent accessses to the drive would not work until the
user pressed F7 to reset the error condition, and the TRS-80 program
would always see the problem as a sector not found or a write fault
instead of an empty drive.  Now we check whether a disk is present
whenever the drive is selected, unless we have already checked in the
last T seconds (including the implicit check that occurs when you read
or write to the drive).  If the drive is empty, the TRS-80 program
sees this correctly, there are no stderr messages, and F7 is not
needed.  F7 will still force a check before the timeout T.  T is
currently set to 3 seconds.

* Drive motor timeout is now emulated.  Formerly drive motors were
emulated as never timing out; now they time out after two seconds
(measured in T-states).  This is useful for a few programs that fail
to handle error cases properly unless the motor eventually times out
and causes Not Ready to be asserted.  Model I SuperUtility has been
observed to do this, at least in some unusual cases.  Minor bug: the
T-state counter doesn't run during HALT instructions or when the
emulator has detected that the TRS-80 program is blocked waiting for
keyboard input (but it does run while the TRS-80 program is handling
timer interrupts out of those states).  

* Fixed bugs in maintaining the track position on real floppies.  Part
of the fix causes a restore (recalibrate) on each real floppy drive
when xtrs is first started, and when an emulated drive is changed from
a .dsk file to a real drive and F7 is pressed.  The other part avoids
forgetting the head positions when F7 is pressed.

* Detect attempts to format a track with a sector ID that is not
followed by any sector data (i.e., no data address mark).  We can't
represent this on either a .dsk file or a real floppy, so we print an
error message.  In the JV3 .dsk file case, we record it as an
intentional CRC error instead, even though that is not the same thing.

* Made -truedam affect reading real floppies.  Without -truedam,
single density F8 is translated to FA upon reading; with it, F8 reads
as F8.  On writing to real floppies, it is still the case that all
non-FB DAMs are written as F8, but with -truedam you now get an
non-fatal error message for F9 and FA.

* Changed delayed event scheduler to work in terms of T-states, not
instructions.

* Used delayed events to emulate semi-correct timing for the Read
Address command on emulated floppies.  Unfortunately, there were so
many problems making this work with real floppies that I had to give
up.  The timing emulation does make the HyperZap "A" command produce
reasonable-looking results on emulated floppies, for what that's worth.

* Tweaked the FDC emulation so that the method for formatting mixed
density tracks that Model III and 4 Super Utility uses will work on
emulated floppies (but not real floppies) -- they toggle the MFM input
to the FDC during formatting.  I had no idea that would work on real
hardware, but evidently it does.  The method that Model I SU and
HyperZap use still does not work -- they format the track first in one
density with a very large gap at the beginning, then reformat over the
gap in the other density, aborting the format in the middle using the
Force Interrupt command to avoid overwriting the first set of sectors.
It's hard to see how to do this reliably with a .DSK file, which
carries no information about gaps; how do you decide which sectors (if
any) from the old track should remain valid?  It also would be a pain
to implement.  Neither method works on real floppies; it does not seem
feasible to implement mixed density formatting on a PC-style FDC,
especially not working at arm's length through the Linux driver.

* Fixed a few small problems with FDC emulation that were revealed by
HyperZap.  I still can't really say that xtrs supports HyperZap well
enough to be useful.

3.2 -- Sat Jan  2 20:43:42 PST 1999  Tim Mann

* Reworked JV3 DAM support to be more accurate by default, or fully
accurate with the new -truedam flag.

3.1 -- Tue Dec 15 19:37:30 PST 1998  Tim Mann

* Added cd/cmd, pwd/cmd, unix/cmd, mount/cmd, and umount/cmd.  They
are compiled with Misosys C and therefore probably work only on LDOS
and LS-DOS.  (They definitely don't work on Model I TRSDOS 2.3 or
Model I Newdos/80.)

* Changed import/cmd and export/cmd to default the destination
filename.

* Added emt_getddir and emt_setddir.  Changed MC names of some emts in
xtrsemt.h because MC's linker requires names to be unique in the first
seven characters.  Fixed several bugs in xtrsemt.h and xtrsemt.ccc.

3.0 -- Sat Dec 12 11:19:58 PST 1998  Tim Mann

* Fixed a bug in the retry loop that's used to read real floppies with
sector sizes other than 256.  It would screw up if the sector could
not be read at all.

* Fixed a bug in reflecting CRC errors on real floppies to the
emulated controller.  Formerly this was done in an incorrect way that
would typically hang the floppy driver in the Z-80 DOS.

* Minor portability changes.

2.9 -- Fri Nov  6 23:14:20 PST 1998  Tim Mann

* Fixed a nasty bug in formatting emulated floppies with sector sizes
other than 256 bytes.  Sector data would be written to the wrong
offset in the file during the format and during any subsequent writes,
until the next F7 disk change or the next time the emulator was run.

* New version of cpmutil.dsk from Roland Gerlach with a bug fix.
IMPORT.COM would crash if the CP/M filename you specified already
existed and you gave a different name in response to the error
message.

2.8 -- Sat Oct 31 15:00:52 PST 1998  Tim Mann

* The default XFree86 keymap reports the shifted F1-F10 keys as
keysyms F11-F20, so I changed xtrs's mapping from keysym to TRS-80 key
to compensate.  Shift+F6 now works as TRS-80 Shift+0, as documented.
The remapping entailed some small user-visible changes to seldom used
keys: (1) F11-F14 no longer have distinct functions of their own. (2)
Home is no longer a synonym for Clear; instead it takes over the
obscure duty of letting you activate the Model III/4 RightShift
position even if you're in Model I mode.

* Replaced the incorrect DAA implementation with a correct one, fixed
a few obscure bugs in documented Z-80 flag settings, and implemented
all the undocumented flag settings.  I used a test program included
with the yaze CP/M emulator to check for errors, and I read the yaze
code for guidance on fixing them.  Thanks to the author of yaze for
checking his implementation so carefully and making his code available
to others under the GPL.
	
* Bug fix: due to a race condition between the main thread and the
SIGALRM handler, autodelay could cause xtrs to hang when the delay
changed from nonzero to zero.

* Bug fix: trying to do a Write Track on an emulated drive with no
disk in it could crash xtrs. Some other operations also failed to
check for an empty drive, but with less dire consequences.

* We now compute a reasonable gap3 size on real floppies instead of
using a small constant value (0x0c). The small gap3 seemed to cause
problems at times. The formulas might still need tweaking.

* Added code to measure the gap sizes that Z-80 software is trying to
format. These are not actually used for anything, but can be printed
if debugging is turned on. It's not possible to feed them directly to
a PC FDC due to interface differences.

* Formatting a real floppy now tries to use the fill data byte given
by the Z-80 program. Since the WD17xx let you fill with (almost) any
data, but PC FDCs let you choose only one byte, we can't fully emulate
this. We just pick the last fill byte given in the last sector and use
that throughout. This is needed for CP/M, which looks for its E5 fill
byte to find free directory space!

* Changed the real floppy sector-not-found error handling heuristic
for reads to try each possible sector size once internally before
returning an error to the Z-80 program.  Of necessity, however, a
sector-not-found error on write is still returned immediately to the
Z-80 program; on the retry with a different sector size, we will
need to run around the Z-80 program's loop more or fewer times to pick
up the number of bytes we are trying with.

* Fixed a bug in reading/writing 1024-byte sectors on real floppies:
the error handling heuristic was not trying that size.
	
* Minor bug fix: executing a HALT instruction would pause the emulator
even if a disk interrupt was scheduled to happen a few cycles
later. We now check the event schedule before pausing.

* Fixed a bug in doing inverse video with -usefont.

* Changed the default font for -usefont to have fewer *'s, to eliminate
problems where the wide version didn't scale the intended font.

2.7 -- Sat Oct 24 16:21:17 PDT 1998  Tim Mann

* Fixed a bug that would have prevented the Z-80 emulation from working
on machines where char is unsigned by default.

* Implemented the alternate characters for Model III/4, corrected all the
character generator fonts and added several options.  Decommissioned the
.bdf fonts; they are inaccurate and (for unknown reasons) they don't work
with the font scaler in XFree86.  Decommissioned the -trsfont feature;
now graphics blocks are always generated at runtime, not taken from the
font. Thanks to Todd Cromwell III for supplying real Model III character
generator ROM data obtained using a ROM reader.

* Added a fake boot ROM to use as a default. It just prints a message that
you don't have a ROM installed for this TRS-80 model. Maybe in the future
it could be beefed up to do more: the Model 4 (when not in Model III mode)
and the Model 4P don't require a great deal from the ROM.

* Updated xtrsemt.h and xtrsemt.ccc.

* Added a bunch more functions to emt_misc. 

* Cleaned up code to remove most warnings.

* Added -sizemap and -stepmap.

* Added signal blocking around calls to readline in debug.c, allowing it
to work again.  See Makefile for details.

2.6 -- Wed Oct 14 23:05:01 PDT 1998  Tim Mann

* Sector sizes other than 256 are now supported! 128, 256, 512, and
1024 all work, on both JV3 floppy images and real floppies. This
feature uses a backward compatible extension to the JV3 format; xtrs
JV3 floppy images that use only 256-byte sectors will still work with
other emulators. I could find very little TRS-80 software to test this
feature with. Super Utility crashes when reading sectors larger than
256 bytes. Trakcess works well, though.

* Fixed a bug in delivering FDC completion INTRQs. Previously an
interrupt could be lost if Z-80 software read the status register just
at the moment the interrupt was to be delivered.

* Fixed bugs in the FDC "read address" command on real floppy drives.
	
* Added the -doubler option, and added a more accurate emulation of
having two separate FDC chips in Model I mode.  Super Utility now
works properly with double density if -doubler is set to either
"percom" or "radioshack" (not the default value "both").

* Bug fix: The floppy controller's BUSY bit was not set while the FDC
completion INTRQ was being delayed.  That was the real reason why
increasing the delay caused problems (as noted under 2.5 below).

2.5 -- Mon Oct 12 14:59:14 PDT 1998  Tim Mann

* Fixed a bug in graphics-only mode. When there is no text overlay,
graphics should always be 640x240. Graphics should be 512x192 only
when 64x16 text is actually overlayed. Thanks to Mark McDougall for
the bug report.

* In the fix for CP/M described under 2.3 below, I increased the delay
for disk INTRQ in too many cases. The increased delay for seek
commands was causing Model III NEWDOS/80 to fail to boot. For now, I
reduced the delay on seek (and related commands) back to the pre-2.3
level, while leaving the increased delay for reads on a blank track
(and related commands) in place. Thanks to Mark McDougall for the bug
report.

2.4 -- Sun Oct 11 08:14:36 PDT 1998  Tim Mann

* Oops. The fixes for Montezuma CP/M mentioned below were inadvertently
left out. They are included in this version.

2.3 -- Sun Oct 11 01:05:17 PDT 1998  Tim Mann

* Added import and export utilities for Montezuma CP/M, written by Roland
Gerlach. Thanks!

* Fixed bugs in 512x192 graphics mode, reported by Mark McDougall.

* Fixed a problem that prevented Montezuma CP/M from formatting
diskettes.  The number of instructions of delay between trying to read
a sector on a blank track and getting a disk INTRQ interrupt was too
short.  Also, reading the disk status register would cancel a pending
interrupt even if it was still being delayed (i.e., had not yet
actually happened).

* Fixed bug in Model 4 @ICNFG chaining in xtrshard/dct. If you have
the old version of xtrshard sysgened, you should remove your
config.sys and re-sysgen with the new version.  The bug could prevent
other sysgened drivers from initializing at boot time, and could even
cause crashes while booting due to an address being executed as
instructions.

* Fixed Expose event support to do fewer redraws. This reduces the
problem of xtrs hanging for a long time (doing repeated redraws) after
the window is moved, which shows up on window managers that display
the coordinates on top of the window while it's moving.

* Fixed bugs in -borderwidth, and changed default to 2.

* Added emulation of the Radio Shack hi-res graphics card, including
undocumented ports. This card is similar to the Micro Labs card but
not identical.

* Added mouse support and a mouse driver.

2.2 -- Sun Sep 27 19:59:00 PDT 1998  Tim Mann

* Added hints on configuring for games to the man page.

* Improved the -autodelay algorithm.  It now finds the correct speed
much faster.

2.1 -- Sat Sep 26 14:06:49 PDT 1998  Tim Mann

* Added emulation of the Micro Labs Grafyx Solution Model 4 hi-res
graphics card. Thanks to Mark McDougall for typing up a summary of the
documentation for me!  I checked the behavior on doubtful points
against Matthew Reed's emulator -- thanks to Matthew for pioneering
here.

* Added -autodelay. This dynamically adjusts the -delay setting to try
to make programs run at about the speed they would have on a real
machine.  It works by monitoring how many T-states are executed during
each heartbeat clock period and adjusting the delay value up or down
by one on each tick.  It doesn't track very precisely, but is good
enough to make realtime games a lot more playable.

* Deleted the -spinfast option, which is no longer needed.

* Disk index holes are now emulated in virtual time, locked to the
rate at which T-states would be executed by a real TRS-80. Thus
programs that measure the disk rotational speed with timing loops will
get the answers they expect. In particular, this will correct problems
with the FORMAT programs in TRSDOS 2.3, VTOS 3.0, and at least some
NEWDOS versions.

* Added a T-state counter.

* Keyboard repeat is automatically turned off when focus enters the
window and restored to the previous state when it leaves.  Thanks to
Marcelo Fernandes Vianna for the idea and an initial version of the
code.

* Partial support for 16-byte non-IBM sectors in the 1771 FDC
emulation, just enough to support VTOS 3.0's copy protection
scheme. Bit position 0x04 in the JV3 sector flags field is now used to
indicate a non-IBM sector.  Thanks to D. J. Dubay for inspiration.
	
* Parameters to the keystretch algorithm can now be controlled from
the command line. Tweaking them might be of help in getting games or
other software that polls the keyboard in unusual ways to work well.
	
* Bugfix: in Model 4 mode, we could try to access the SoundBlaster
hardware even when the -sb switch was not given, typically resulting
in a crash.  Thanks to Roland Gerlach for the bug report.
	
2.0 -- Sun May 24 18:14:52 PDT 1998  Tim Mann

* Added -delay flag, a crude speed control.

* Added Model 4P emulation.  Please don't pirate Model 4P ROMs.  Frank
Durda IV has expressed interest in creating a freely redistributable,
simplified 4P ROM for use in emulators, but it is not yet available.

* Added a somewhat more general mechanism for scheduling events to happen
several instructions in the future.  This supports software that sometimes
starts an I/O operation that will trigger a completion interrupt in the
near future, but relies on the fact that the real I/O device could not
interrupt until the CPU had had time to execute a few more instructions.
Previously I had a more kludgey case-by-case solution for this problem,
but that method didn't work for the Model 4P ROM.

* The cmddump program (a TRS-80 DOS load module analyzer) is now included
in the release, but there is no man page.  See the top of cmddump.c for
brief documentation.

* Added ability to use a TRS-80 load module file directly as the ROM.
In particular, the MODELA/III file is in this format.

* mkdisk now puts the original filename in the header of a .hdv file.

* Added missing code to initialize the SoundBlaster DSP and turn on the
DAC speaker.  This was required to make sound work properly on my machine,
which has the ESS1688 AudioDrive sound chip, a SoundBlaster clone.  It's
probably needed for older true SoundBlaster models too.

* Changed sound resource to have same name ("sb") as command line option.

* Moved sound init code to fix bug I introduced in 1.10.

* Older Linux kernels don't have <sys/io.h>; 2.0.29 does not, 2.0.32 does.
Added comment saying to delete the include if it causes an error.

* Fixed bugs in assembly version of settime: it did not work on an
emulated Model III/4/4P in Model III mode using MODELA/III as its ROM, and
it caused JCLs to abort on Model 4 LDOS. 

* Fixed bug in emulating the SLA instruction: the Z flag was not getting
set when 0x80 was shifted left to yield zero.  The bug was affecting the
MultiDOS 4 keyboard driver, causing problems with key repeat.

1.10 -- Thu Apr 16 19:20:30 PDT 1998  Tim Mann

* Fixed bug in setting data rate for real floppy drives that spin at 300
RPM (3.5-inch and 360KB 5-inch).  This bug caused disks written by xtrs in
such drives to be unreadable on a real TRS-80 (and vice versa).  Disks
written by old xtrs versions in 1.2MB 5-inch drives (which spin at 360
RPM) are OK.

* Finished implementing emulation of 8-inch floppy drives.  Added
xtrs8/dct, an LDOS/LS-DOS driver for the emulated 8-inch floppy hardware.
Corrected index hole emulation for 8-inch drives to run at 360 RPM.  Added
support for real 8-inch drives (untested) and emulation of 8-inch drives
by real 5.25-inch and 3.5-inch drives (tested and working).

* Added emt_system.  Updated xtrsemt.ccc.

* Added sound emulation based on code from Fabio Ferrari.  Linux with
SoundBlaster only.

* Fixed bug in returning error (or success) code from emt_strerror.
Reported by Roland Gerlach.

* Give an error message and don't crash if diskM-U is linked to a floppy
device that we don't have read/write permission for.

* Undocumented equivalents for documented instructions are now commented
in disassemblies.

* import.z and export.z modified to also work with Newdos/80.  settime.ccc
rewritten in assembly language and made to also work on Newdos/80.  Thanks
to Ulrich Mueller for this code!

* Corrected disassembly of ld a,r.  Added borderwidth resource.  Minor fix
to import.bas for Newdos/80.  Thanks to Ulrich Mueller for this code!

1.9 -- Thu Feb 12 20:44:02 PST 1998  Tim Mann

* Added a hard disk emulation driver, xtrshard/dct.  It basically uses the
plain file I/O emts, but I added minor variants of open and close to allow
the names to be relative to -diskdir and keep rebooting from causing file
descriptor leaks.  Renamed mkfloppy to mkdisk and added support for making
an empty hard disk.  Hard disk format is compatible with Matthew Reed's
emulators; thanks to Matthew for supplying documentation on his format.

* Distinguished between fatal and nonfatal errors.  Now an unimplemented
opcode is a nonfatal error.  It's treated as a no-op.

* Added the raw code bytes to disassembler printout. Fixed bugs in
disassembling ld (ix+dd),nn and ld (iy+dd),nn.  Corrected disassembly
of rlca and rrca.

* Fixed a fatal bug in emt_lseek.  Added emt_ftruncate.  Fixed portability
bug in emt_open that affected import/cmd and export/cmd.  Added disk
change count on emt_misc.

* "next" in debugger now will step over RSTs.  Note this won't work
right with the funny RST 8 calling convention in the Model I/III Basic 
ROM, which puts a 1-byte argument after the instruction.

* Added import, export, and settime for Model 4 TRSDOS/LDOS 6.

* Fixed problems compiling trs_disk.c on some Linux versions.

1.8 -- Mon Dec  1 16:00:52 PST 1997  Tim Mann

* Moved queueing of keys to after the translation from X keysyms to
TRS-80 keyboard transitions.  This lets shift transitions that are
forced by other keystrokes (where X and TRS-80 keyboards have opposite
shift states to get the same character) be queued separately and thus
separated in time from the basic key transitions.  This separation is
required by the Montezuma CP/M keyboard driver, which sometimes misses
a shift transition if another transition occurs at exactly the same instant.

* Bug fix: an FDC restore command was erroneously zeroing the sector
register instead of just the track register.  Montezuma CP/M tickled this
bug.

* Bug fix to mem_bank from Roland Gerlach.  Switching banks in the
low-order half of the address space was switching in the wrong memory.
Montezuma CP/M tickled this bug and crashed.

* Removed use of pre-POSIX signal code; thanks to Al Petrofsky.

* Added the delayed_intrq_kludge in more cases in trs_disk.c.  This fixes
additional problems with SuperUtility and Trakcess that I didn't notice
earlier.  Also fixed some problems with error handling on real floppies.

1.0.tpm1 to 1.7 -- Tim Mann

* Added floppy disk emulation, timer interrupt emulation, a kludge to
import and export data to the host system, Model III mode, Model 4 mode, a
rewrite of the keyboard handling, support for most of the undocumented
Z-80 instructions, and a bunch of bug fixes and minor improvements.

* Fixed bugs in the instructions rrc, inc ix, inc iy, sra, and neg.  I
didn't notice the sra bug until I tried to get Model III mode working;
the Model III LDOS disk driver uses it.  The neg bug was keeping the
break key from working right under LDOS in xtrs-1.3 and earlier.

* Got the emulator running on both Linux and Digital Unix.  Now it
might have minor problems elsewhere, on less ANSI- and POSIX-compliant
systems.

* Fixed two bugs in the disassembler.  The ld (bc),a instruction was
erroneously disassembled as ld bc,a, and instructions that start "dd
cb" or "fd cb" were not disassembled; for example: bit 7,(iy+03h).

* Fixed some bugs in emulated cassettes.  Formerly if you tried to
append more stuff to an emulated cassette, it would smash whatever was
already there.  If you tried to read from a cassette with nothing on
it, it would lock up so hard that even the emulated boot button
wouldn't work.  Now at least the boot button works.

* SIGIO is now used in a much safer way than before, and the emulator
will work without it if needed.

* Reworked the keyboard support substantially.  A full state vector is
now kept, including all 64 possible TRS-80 keys.  Key queueing is now
always used, and a "stretching" feature prevents the emulated keyboard
state from changing too often, to avoid losing keystrokes.  Keyboard
response is quite snappy and reliable now.  I added enough key
mappings that it should be possible to do about everything you could
do with a real TRS-80 keyboard (and more than on a Model I, because
modern keyboards have N-key rollover).  More key mappings are easily
added by changing tables; there is very little special-case code.

* Did some work on the KBWAIT feature, which tries to avoid burning
host CPU cycles when the emulated program is sitting in a loop polling
the keyboard.  It now works in a lot more cases.

* More accurate character bitmaps are now used, contributed by Al
Petrofsky.  The Model I bitmaps are still not the real ones, which
used a 5x7 matrix, not 7x8.

* The halt instruction is now emulated correctly.  Formerly it caused xtrs
to exit.

1.0 -- David Gingold and Alec Wolman

* Initial release.  Emulated a TRS-80 Model I with 48K of RAM, cassette
drive, and uppercase-only screen.  See README.