home
***
CD-ROM
|
disk
|
FTP
|
other
***
search
/
Games of Daze
/
Infomagic - Games of Daze (Summer 1995) (Disc 1 of 2).iso
/
x2ftp
/
msdos
/
gamesrc
/
arasan_s
/
changes
< prev
next >
Wrap
Text File
|
1994-08-15
|
5KB
|
96 lines
8-16-94: Arasan 1.2a
1. Fixed a bug in the Options module: caused assert failure if
arasan.ini was not present on startup.
2. Fixed two busg in computation of attack info. One could
occur after an en-passant capture. The other case in which
attack info could be incorrect was following a castling move.
Added some debug code (-DDEBUG_ATTACKS) to test attack updating.
Also corrected a bug in Search::draw.
3. Added some more test suites.
8-6-94: Arasan 1.2
1. Fixed a time control bug in module Search. This was causing
the program to fail Bratko-Kopec test #12.
2. Added a positional scoring term to penalize the program for
placing pieces in front of unadvanced center pawns. Also
added a knight mobility score and made some changes to the
computation of bishop mobility. Center control scoring
parameters have also been tweaked, for better performance in
the opening.
3. TESTSRC has been converted to accept a subset of the Extended
Position Description (EPD) format. All the test suites have
been converted to this format, which allows storing the key
move and position together in one file.
4. Two new types of time control, "Tournament" and "Game", are
now supported. A secondary time control can be specified.
5. Arasan now retrieves the predicted game continuation from the
hash table. The code used in earlier versions to keep track
of the predicted continuation was unreliable.
6. A couple of improvements have been made to the search extensions.
The search is extended if a capture move is made and the capturing
piece is "backed up" by another piece of the same color. Also,
the constant Check_Depth has been changed to 2.
7. Since the whole point of trying the null move is to cause cutoff,
Arasan no longer repeats a null move search with a wider window
if cutoff fails, as it used to.
8. The log was not being cleared before a "Load Game" command
was executed, so if this command was executed twice, the second
time the game would not load correctly. This has been fixed.
9. The opening book has been expanded to approx. 10,000 half-moves.
10. The hash table stores the repetition count for a position.
11. The Notation module had a bug. A move might be incorrectly flagged as
ambiguous if two pieces had possible moves to the destination square,
but one of them happened to be pinned. This is now handled.
4-27-94: Arasan 1.1.
This version has no user interface changes. However, there are
a number of bug fixes and internal improvements:
1. In saving a board using FEN, the en passant square was not
being written out when an en passant capture was possible.
2. In reading in a board from a position file, the en passant
square in the board class was set incorrectly.
3. Arasan would prevent the user from escaping check by making
an en passant capture. This has been fixed.
4. When saving a game, the month was being written incorrectly
in the PGN date field. This has been corrected.
5. In the hash table used by the Search class, the ply was
being stored instead of the depth (limit-ply). This caused
some nodes to be evaluated by table lookup when they should
have been searched. This a couple of other hash table
bugs have been fixed.
6. Castling and en passant status is now stored in the hash
table and must match that of the current board for a
position to be retrieved from the table. To be 100% correct,
the repetition count should be stored too, but that is
still not done.
7. The move ordering calculation in class Move_Ordering has
been improved. In most cases, this will make searching
faster (however, this is partly masked by the hash table
fix, which makes some searches slower).
8. At each iterative deepening, the search used to try moves in
the order of scores returned by the previous search. Now it
tries the highest-scoring move first, but calls Move_Ordering
to sort the remaining moves. This is another speed enhancement.
9. The search module now always starts with a 1-ply search -
it used to begin with a 2-ply search; this could cause
the program to miss a one-move mate and play instead a
mate in 2.
10. A search now terminates when the side to move detects that
its opponent has a forced mate in n, where n <= search
depth. Further searching will not change the score.
11. In some cases, the search routine was not performing the
forced move extension.
12. Additions and improvements have been made to the endgame
evaluation in module Scoring. These changes significantly
improve the program's play in simple king and pawn endgames.
However, the program is still fairly dumb about endgames.
13. The "TESTSRC" program (distributed with the source) has been
extended so that it can read >1 position from a single file.
The test suites distributed with version 1.0 have been
consolidated into single files. Also, a file has been added
(WINATCHES.FEN) containing the 300 "Win at Chess" problems.
14. Rebuilt .EXE with Borland C++ 4.0.
3-8-94: Arasan 1.0 initial release