Amiga MA Magazine 1998 #3/amigamamagazinepolishissue1998.iso/szachy/uchess/uchesspatch/readme

Text File | 1994-08-04 | 16KB | 425 lines

v2.89
Enhancements to eval and search make for a stronger game, it
seems much more consistent against other strong computer pgms
now.  Also fixed the clearmem routine in 68020 assembler, it
had a bug.

v2.88
Enhanced clock management for >25 secs/move games
Also fixed problems with the hint move and contention between
the 2 processes, and made the mvstr large enough for algr, it
was too small before, resulting in corrupt data at times.

v2.87
Split the game into 2 AmigaDOS processes, one for search/eval another
for housekeeping, this allows stronger thinkahead with
far fewer side effects, which would damage play in games
where thinking was on.  Fixed a few draw/ttable bugs,
ttable aging is now better than in 2.85/2.86
Enhanced clock management, long games should be stronger.
Retuned some pawn evaluations.
All changes are transparent to the user.

v2.86
Fixed up the repetition code, it was causing mis-scoring
and interfering with play quality in many cases.

v2.85
Changed the hashtable/evalcache routines, changed search and
selectmove, incorporating modified versions of changes
from gnuchess 4pl70 Made many changes to eval as well.
This version should have improved playing strength vs
other versions, from improved eval and much improved
speed.  Some Black openings added.

v2.83
Fixes some bugs in loading games, and in toggling book, bugs
that would cause the hint mechanism to fail and always return
"I have no idea" as the hint.  Also fixed a bad bug that sometimes
happened when thinking was on, and you started with 1.e3
Changed the way flashing the screen to signal a move works.
Save game format is now changed to accomodate enpassant
information.  Old games should still load, but enpassant
will be messed up, possibly.

v2.81
Features more readable newlook menus and now allows you
to click drag multiple menu selections per standard Amiga
specs.

v2.79
Fixed menus so that they should work with MagicMenu

v2.78
Fixed some NULLMOVE and search bugs, that could cause
unusual performance, especially in endgames.

MAJOR Hash table performance boost, much faster eval
as a result.

Many enhancements to search/recapture extensions in search(),
result in deeper searches, better eval per ply as well.

Redid the white/black books, more variety, no real
stupid openings, still very small book, but seems
to be a safe one, and more variety than the stopgap
book I was using.

This version supports 3rd party graphics cards.
It supports them "natively" using the screenmode requestor.
DO NOT use any screen promotion utilities like ChangeScreen
or whatever, to promote the screens, just choose your
graphics modes native screen from the screenmode requestor.
For 15Khz monitor users, with AGA chipsets, the
64 and 256 color modes will now come up with
DEFAULT_MONITOR screens, rather than forcing PAL mode.
The screens will now autoscroll, so you can see the
whole 480 pixel high image, even on NTSC monitors.

3rd Party graphics notes:

You need OS version 3.0 or later for 3rd party graphics
support. GVP Spectrum owners note the NOBOBS CLI option
and the NO_BOBS=1 WB tooltype.

Do not use the RTG option on AGA/ECS modes, you can if you
want, but the displays will use more memory and
will be slower...
To start with 3rd party graphics cards there are 2 new
CLI options, RTG and NOBOBS:

UChess RTG

will start in RTG 3rd party graphics mode.  This will
bring up the screenmode requestor and you can
pick a screenmode for chess play.  It should be
640X480 to fit the bitmap...smaller will autoscroll
but 3rd party graphics software may not autoscroll
well...so use a 640X480 or larger screenmode
The corresponding TOOLTYPE to add in the WB icon
is :
RTG=1

******SPECIAL NOTE TO OWNERS OF GVP SPECTRUM EGS CARDS******
       (OTHER BOARDS MAY HAVE THIS TROUBLE AS WELL)
           (IF YOU HAVE LOCKUPS, TRY THIS)

The software with these boards will LOCK UP THE PGM
when it tries to use BOBS, to prevent this from
happening there is a new UChess argument.  From
the CLI use:

UChess 0 RTG NOBOBS

this will inhibit UChess from using BOBS to animate/move
pieces.  Meaning pieces you pick up with the mouse
will be invisible until you drop them onto a square.
This is due to deficiencies in the GVP software.
without the NOBOBS argument, the GVP lack of system
BOB support locks up the pgm entirely, so USE THIS OPTION!!

The corresponding TOOLTYPE in the WB icon is

NO_BOBS=1

v2.72
Fixed a slight time control problem, it was not adding
remaining time when # moves was completed.
Also fixed a nullmove problem that may have resulted
in wierd endgame checks.

v2.71
Fixed bugs in search causing selective deepening to be
too selective in many cases.
Redid the repetition code as per 4PL68 standards, can
now recognize perpetual check and other draw situations
more readily.
Many bug fixes which now make the ideas in 2.69 a reality
for all games.  This version is more efficient, and makes
fewer mistakes vs strong opponents than 2.69 did.
Endgame is improved as well.

v2.69
Fixed some endgame deficencies, in King-Knight-Bishop vs King
endgame.  Fixed some lookahead bugs, and some slight mouse vs
kbd display bugs.

v2.67
Totally new eval/search.  This is based on gnuchess 4PL67 with
many enhancements.  Searchs more nodes per ply, so will get
fewer ply on slow machines, but plays better.  Also support for
Super72 mode 800X600X64 or 256 colors.  New CLI arg Super72, new
Tooltype SUPER72=1.  The L version requires 9M or more
of FASTRAM (like before, but this was not clear to everyone).
This version plays much more competitively vs skilled players
and good computers, less daring against lesser opponents though.
May not be as well suited to slower systems (020/030's), but
in all tests it seems stronger, even on A3000/25's.
(in computer v computer tests this version went 7-1-2 vs
Sargon5 on a 486, v2.54 usually went about 3-5-2 or so)

v2.54
Fixed some bugs with resign, and a small bug in undo
which caused mate msg to be displayed if you
undid mate.
Plays the same as 2.53

v2.53
Finally after days of effort I managed to find and
fix a couple of bugs introduced in 2.50 with the new
search on opponents time code...
Bug fixes and changes to genmove (movelist) nullmove
routines, and especially to search on opponents time
logic.

v2.52
Bug fixes to load game, and to transposition table routines.
Some attempt to deal with eval == 20000 bug.
Major speedup of movelist routines, speed of search and
eval greatly increased.

v2.51
Slight enhancement to endgame eval routines.

v2.50
Many changes incorporated from the Gnuchess 4PL64 release.
Any changes that did not conflict with UChess specific changes,
were merged into this version.  Changes to endgame (major)
eval and search routines were made, to bring this
version up to Gnuchess 4PL64 standards.
Many changes to clock handling routine, UChess now
manages its clock MUCH better, and thinking on opponents
time is MUCH improved, making UChess far better than
before when this feature is enabled.
NOTE: Rate Pgm is now slower than before, it takes it
longer to calc some move situations because of a bug
fix to the history table.  The older way resulted
in missed moves, but could calc some situations faster,
but not correctly.  This new way results in slower calcs,
but correct calc of moves.

v2.35
Cleaned up Mating routines, UChess will now choose shortest
route to mate and not keep searching beyond that to come
up with strange endgame moves.
Fixed some eval and search buglets and fixed a side
effect in Take Back move.  Now take back move will
clear out history buffer, so re-eval of moves after
takeback will result in best move being made, not
necessarily a different move from the last choice.
Taking Back a move now takes much more time, but
results in steady gameplay.  Prior version often chose
alternate moves after take back a move was chosen.
Redid some of the positional modifiers in eval routines,
this version seems to play quite a bit stronger and
steadier than previous versions, when pressed.
Recompiled using SAS/C 6.51

v2.33
Exactly like 2.32 but will run on AmigaDOS 2.04 and 2.1 systems.
Calls to SetABPenDrMd were made unconditionally in v2.32 of UChess,
meaning v2.32 required AmigaDOS 3.0 or later to run..sorry.

v2.32 (recompiled using SAS/C 6.50)
New Display Option for AGA machines, at startup you can
specify FASTAGA (or 64) at the CLI or you can use the FASTAGA=1
or COLORS=64 arguments in the Workbench ICON, this opens UCHess on
a 640X480 or 640X512 64 Color Screen, for additional speed.
Especially helps display speed in the MultiScan Productivity mode.
It looks almost as good the 256 color screen and is much faster.

Rewrote some routines in 68020+ Assembly language, and
re-aligned many data items for more nodes/sec eval speed.
Redid the entire opening book code, uchess will now
look for uchess:uchesswhite.book when playing as
white and uchess:uchessblack.book when playing as
black, if it cannot find these it reverts to uchess:uchess.book.
Sample white/black/neutral books enclosed, but much
improvement/tuning of these books is needed.
In general the white/black should work best for now.
In the future I hope to come up with really optimal
white/black openings and include these, at least for
now the pgm will avoid openings that are real detrimental
to it by splitting the book into 2 parts.
Also made the edit request autoactivate the edit string gadget,
for more convenient edit game entry.
Also added a resign feature, if you have started the game
in a regular manner at advanced level, and you get over
9.5 pts ahead of UChess, it will offer to resign (once).
If you load a saved game, or edit the board, it will
not offer to resign.  It will only offer to resign the game
one time.

v2.31
Re-Arranged the new game feature so it now pays
attention to the skill level when setting up a new board.
All new games came up as Advanced before, regardless
of skill level chosen

v2.30
Added a Level feature for play vs the computer.
3 levels are available, Advanced (same as before)
Intermediate (Knight Odds), and Easy (Knight + Queen Odds)

Choosing these options will force a new game, with you
are white and UChess as black, UChess will have its
Knights removed on Intermediate, and its Knights+Queen
removed on Easy.  Advanced is an equal game, like
past versions


v2.28
fixed a bug in the initialization and re-initialization
of the cache and transposition tables.
Makes for more consistent play and a better start for
the computer when using the new game option.

v2.27
Just recompiled using __USE_SYSBASE for a bit more perfomance
of system exec routines.

v2.26
Fixed many bugs in the transposition table routines, which
would cause crashes during overnight analysis with 2.25, also
affected play somewhat, this version can handle overnight
analysis, long games with no problems.  2.25 and some
earlier versions could corrupt memory if they computed
too long (for hours).
Redid the compute rating equation, for a bit more realistic
rating.  With the major speed improvements of 2.25, the old
rating equation was a bit off, giving an inflated rating of the
pgm.  This new equation is somewhat better.

v2.25
Major speed enhancements, this version should play FAR stronger
thanks to rearrangement of Cache Tables.
Major endgame changes as in gnuchess 4 pl 61, for lone
king endings.

v2.18
Attempted to reduce the number of "wild" moves that
the program makes.  Under rare circumstances, the
pgm would make wild moves based on poorly considering
the users response move.  This should not happen much
any longer.  May take more time to come up with a move
in these cases though.  Should prevent it from throwing
games away.

v2.17
Fixed a minor bug in thinkahead move choice.

v2.16
Finally found a longstanding bug screwing up move selection with
thinking on.  Thinking on now helps a lot, it also does not
screw up judicious use of the clock by the program.

v2.15
Further refined conditions when thinkahead move is taken, or additional
thinking beyond the current ply is warranted.  Refined for more likely
choice of the thinkahead move in very short (<10secs/move) games

v2.14
Fiddled with timing when thinkahead move is guessed correctly,
it was taking its thinkahead move immediately if it was as deep,
or deeper, than the previous move it pondered, it now only takes the
move immediately if it is as deep or deeper and at least 4 plys in
depth.

v2.12
Added a rating feature, based on fact that the gnuchess 4pl60
core plays at about 2300 on an IBM/RS6000, and that each
doubling of processor speed results in a 75 pt ratings improvement,
the pgm will estimate its USCF strength on your machine.
On an A4000 the L version is about 2200 USCF or so.

v2.11
Fixed a bug in dialog window for depth control.

v2.10
Redid time control so that on long matches, it will not
try and search too far ahead, if it guessed right.
It was trying for real deep searches when it would
result in massive overtime for the last level, on some
occasions, hurting play later by forcing the pgm to
rush too much.
Also load/save game now updates clock display when it loads,
and sets move now accordingly.

v2.09
Fixed bug when depth mode was on, thinking was on, and show
thinking was on, it was disabling show thinking in those cases.
Cleaned up timing a bit.
IT IS NOT CLEAR TO ME THAT THINKING ON HELPS.
I am testing this, it may be that with thinking mode off, it
plays more consistent chess, I am not sure though.

v2.07
Added Depth Control, and Book Control Options
To use Depth Control, just set it to # of plys you
want the pgm to look ahead, it will then finish its move
when it finishes that ply, regardless of time.
Book Control just allows you to enable/disable opening book

v2.06
Fixes a problem with mem alloc which showed up under Mungwall, it is
now 100% enforcer and Mungwall clean.  Guards against loading improper
game files.  Other Minor fixes.

v2.05
Redoes the time control routine for thinking mode on.
In earlier 2.0x versions, the forced timeout mode when
thinking is on, could result in poor move selection
in some circumstances, this fix should allow fewer mistakes
in move selection with thinking on, when computer is forced
to stop searching a ply earlier than it would like.

v2.04
Fixes a display bug when castling with mouse (rare).
Asks for PAL monitor when opening a 15Khz 256 color screen.

v2.03
This version fixes a bug which caused the load game function
to crash on certain machines, when the pgm was run with
a priority > 0.  Also the Move Now feature is more responsive.

v2.02
Puts back Underpromotion of pawns, which disapeared briefly
in v2.00 and v2.01, Move Now item works again.

V2.01
Added an additional CLI arg and tooltype, from CLI
type "UChess 15Khz" if you have an AGA machine and want
to force a 640X480X256 color 15Khz screen, may not be enough
overscan for whole screen to be visible on monitor.
From WB you can use the MONITOR=15KHZ tooltype to force this.
Pgm will auto default to 15Khz 256 color screen if you have
and AGA machine with no multiscan monitor in devs:monitors,
to force 16 colors use 16 color arg.  The 16 color screen
will fit easily on a 640X400 display, the 256 color
screen will not fit on a 640X400 display.  People with
PAL machines should have no problems fitting it all on
a 640X512 15Khz screen.

This version fixes MAJOR time control bugs present in v2.0


V2.0
I widened the search for this version and it has defeated
version 1.01 head to head 10 out of 10 times.
I belive this version plays MUCH stronger chess than
version 1.01!

-Roger


V1.01
This is version 1.01, it cleans up the 16 color graphics and
makes them look much better than in version 1.0
That is the onluy change.
See Uchess.doc for details


-Roger
[e-mail]
ruzun on bix