H.G.Muller [Sat, 8 Dec 2018 11:00:48 +0000]
Let engine thread parse difficult commands
The design of the adapter is completely changed: the engine thread is
made responsible for parsing and execution of all commands that cannot
be processed during a search. To this end it alternates between reading
from the engine (while the engine is searching, or should respond to a
'uci' or 'isready' query) and from the command queue (which is emptied
before we dedicate to relaying ponder output). Searches are terminated
by the GUI thread when a command arrives that would interrupt the search,
so that the engine thread will start looking at the command queue again.
Syncing threads to wait for an engine reply is replaced by just doing
a (blocking) read for engine output in the same threat, and having the
processing routine return once the answer has arrived (and is processed).
Instead the Sync pipe is now used to block reading of the command queue
when this is empty.
This change breaks several of the less elementary commands.
Fix command queue
H.G.Muller [Sat, 8 Dec 2018 09:27:20 +0000]
Move Engine2GUI to a more convenient place
The main loop of the engine thread is moved to a place in the code
where future changes would require fewer forward calls to other routines.
H.G.Muller [Sat, 8 Dec 2018 09:21:03 +0000]
Launch searches at end of GUI loop
The loop for interpreting GUI commands is restructured by launching
searches at the bottom of it rather than at the top. As it was an
infinite loop anyway this only eliminates the first call to LaunchSearch(),
which is OK as the engine would never have to search before any commands
were received.
As this change brought the 'nomove' label to the top of the loop,
the 'goto's to it could be replaced by 'continue's, and the label deleted.
That for engines that cannot ponder the 'easy' and 'hard' commands now
skip LaunchSearch() is OK, as LaunchSearch() would not do anything in
that case.
H.G.Muller [Sat, 8 Dec 2018 09:12:13 +0000]
Eliminate continues from GUI loop
The 'continue' statements at the end of the code sections for handling
specific commands, which caused matching for the remaining commands after
it to be skipped, are now deleted, and the code to be skipped is placed
in an 'else' clause. This to make restructuring of the GUI loop easier.
The 'continue' statements for 'easy' and 'hard' commands were left,
as these were harmless: it doesn't matter whether we call LaunchSearch()
after them or not.
H.G.Muller [Sat, 8 Dec 2018 09:02:49 +0000]
Put launching of searches in a separate routine
The code section that tested if the engine should search, and would
start the required search if this was the case, is now put in a routine
LaunchSearch(). This is then called from the infinite loop that handles
the GUI commands (where it was located before).
H.G.Muller [Sat, 8 Dec 2018 08:54:37 +0000]
Let 'searching' also indicate thinking state
The flag 'think' is eliminated, and its function is now indicated by
giving 'searching' the value 3. This makes 'searching' now a complete
state indicator, which indicates if we are idle, pondering, analysing
or thinking.
H.G.Muller [Fri, 7 Dec 2018 22:20:36 +0000]
Combine function of 'searching' and 'pondering' flags
The vaiable 'searching', formerly used only to indicate if the engine was
doing an analysis search, now indicates pondering if it has value 1, and
analysis if it has value 2. This to simplify the state logic.
H.G.Muller [Fri, 7 Dec 2018 22:02:48 +0000]
Pass command to DoCommand through global command queue
Rather than passing the command to process as a parameter to DoCommand,
we now append it to a global 'command queue', from which DoCommand() then
reads the first line. This in preparation for letting the engine thread
handle the difficult commands.
H.G.Muller [Fri, 7 Dec 2018 21:35:01 +0000]
Parse 'difficult' commands in separate routine
The parsing of (most) commands that can only be executed when not searching
is now put in a separate routine DoCommand(). This in preparation for
having the engine thread do this.
H.G.Muller [Fri, 7 Dec 2018 14:02:14 +0000]
Put handling of engine output in separate routine
The reading, decoding and handling of the engine output is now put in
a routine HandleEngineOutput(), which is still an infinite loop that
wakes up a blocked GUI thread after a 'uciok', 'readyok' or 'bestmove'
command. This in preparation of a new design, where it would break out
of the loop and return after these commands, to let its caller process
any queued GUI commands.
H.G.Muller [Thu, 6 Dec 2018 20:23:49 +0000]
Do all input through one ReadLine() routine
Some duplicate code now is made into a subroutine, to which we pass
the stream from which it should read a line, and the buffer to hold
the result. It returns whether the read was succesful (i.e. no EOF).
As this can also easily fake engine output from a file, the special
routine GetChar()we had for this could be abolished.
H.G.Muller [Sun, 2 Dec 2018 10:40:18 +0000]
Implement option ' ponder always'
UCI2WB can now also be configured to let the UCI engine ponder even
in force mode. This is useful during OTB tournaments, when the GUI
is still in book (and thus keeps the engine in force mode). Because
there is no ponder move at this stage, the engine will be set pondering
on the opponent's position, and the ponder move is set to an empty
string to force a ponder miss afterwards.
H.G.Muller [Thu, 4 Oct 2018 09:06:52 +0000]
Implement extended UCI info string variant option
To allow UCI engines to also define variants with different board size
and parent variant, an extended version of the info-string-variant command
is recognized. This can also incluse 'files', 'ranks', 'pocket' and
'template' fields. The default values for these are 8, 8, 0 and "fairy",
so that we remain compatible with the old info-string-variant command when
the new fields are absent.
H.G.Muller [Wed, 31 Jan 2018 13:37:10 +0000]
Support Scorpio bitbases
The CECP EGT flavor 'scorpio' is now translated to the UCI option
'bitbases path'. This is a bit awkward because, unlike all other flavors,
the UCI option name does not start with the flavor name. So new code had
to be added to recognize this as an exception. In addition, the UCI option
is not universally used by engines supporting Scorpio EGBB.
H.G.Muller [Mon, 29 Jan 2018 10:43:02 +0000]
Add support for robbobases
The UCI options RobboTripleBaseDirectory and RobboTotalBaseDirectory are
now recogized as defining the egtpath for flavors 'robbtripple' and
'robbototal', respectively. For this the general EGT-handling code had to
be adapted to be able to append 'BaseDirectory' instead of '(Tb)Path' to
the flavor for setting the option, and clipping this off for extracting
the flavor name. This is achieved by using the part of the standard option
in the list of EGT formats up to the first capital as CECP flavor name,
and match the flavor name in the egtpath command with the list, to retrieve
the full name from there.
H.G.Muller [Fri, 27 Oct 2017 20:17:10 +0000]
Change UCI_Variant name from threecheck to 3check
This to stay in sync with a similar change in multi-variant Stockfish,
which now makes the UCI and CECP names equal.
H.G.Muller [Wed, 1 Feb 2017 16:47:29 +0000]
Suppress setup command also in Suicide ad Losers
Variants suicide and losers are added to the list of exceptions for which
the UCI 'info string variant' command should not result in a CECP 'setup'
command.
H.G.Muller [Sat, 24 Dec 2016 21:17:00 +0000]
Bump version to 3.0
H.G.Muller [Wed, 4 Jan 2017 17:06:29 +0000]
Disarm printing of 'uci' to the GUI in debug mode
A '#' sign is now printed when UCI2WB reports it sent the 'uci' command.
H.G.Muller [Wed, 4 Jan 2017 17:04:10 +0000]
Fix egtpath for Gaviota EGT
The standard option name for the Gaviota path in UCI is GaviotaTbPath,
ot GaviotaPath. So we have to strip off the 'Tb' part, and add it again,
when converting to and from the CECP name.
H.G.Muller [Tue, 3 Jan 2017 15:27:50 +0000]
Suppress printing duplicat PVs (for multi-PV)
UCI resends all PVs if one new PV arrives,possibly with different depths.
ow any PV of a depth less than the best depth seen from the current search
so far will be ignored. Duplicats of the same depth and score will also
be ignored.
H.G.Muller [Tue, 3 Jan 2017 11:59:46 +0000]
Diversify UCI string options into -string, -path and -file
Heuristics is applied to the option name, in order to guess whether UCI
string options are filenames or directory paths. In which case they are
translated into CECP -file and -path options, so XBoard will give them
the appropriate browse button.
H.G.Muller [Sat, 31 Dec 2016 19:37:53 +0000]
Fix strcasestr for Windows
The strcasestr definition, which we have to supply ourselves to make
the Windows version compilable, was not upto the task of having a
string constant as first argument, and a string that needs decapitalization
as the second argument.
H.G.Muller [Sat, 24 Dec 2016 18:39:57 +0000]
Print egt feature
Options of the form xxxPath for xxx = Nalimov, Gaviota of Syzygy are now
intercepted, and the EGT type name 'xxx' collected, for printing as a
single egt feature when 'uciok' is received.
H.G.Muller [Sat, 24 Dec 2016 18:41:11 +0000]
Fix string options with empty default
When sscanf finds nothing after 'default', it does not touch the target
array at all, rather than putting an empty string in it. So the target
is now initialized to an empty string before the sscanf.
H.G.Muller [Sat, 24 Dec 2016 14:47:13 +0000]
Process CECP egtpath command
The 'egtpath TYPE PATH' command causes the UCI option TYPEPath to be set
to VALUE.
H.G.Muller [Fri, 23 Dec 2016 15:17:58 +0000]
Fix -var option with UCI_Chess960
The variant specified in the -var option would be overruled by the
"normal,fischerandom" emitted later under the influence of a UCI_Chess960
option. This is now solved by always sending the variants feature on 'uciok'.
The varList is now initialized to the -var string, and overwritten by the
protocol defaults before sending 'uci' only when it is still empty.
A UCI_Variants option then gets the chance to overwrite this, while
a UCI_Chess960 option will append "normal,fischeradom" to the list.
H.G.Muller [Mon, 5 Dec 2016 18:47:05 +0000]
Make sure we resign when engine reports we are checkmated
When a UCI engine is set thinking in a checkmated position, the PV
necessarily is empty, and instead of printing an empty 'pv' field in
as pv info, it might omit it alltogether. So we even accept a 'mate 0'
score when there is no PV, so that a following 'bestmove 0000' will
be interpreted as a loss (sending 'resign'), rather than a draw
(sending 1/2-1/2).
H.G.Muller [Mon, 5 Dec 2016 08:29:13 +0000]
Display error message when engine dies
UCI2WB used to exit silently when reading from the engine produced an EOF
response (indicating the engine process died). This then evoked the standard
XBoard fatal error "... engine exited uexpectedly". Now it sends a
'tellusererror' command to the GUI to suppress the standard popup, and
to clearly indicate that it was because of a crash in the engine rather
than in the adapter.
H.G.Muller [Sat, 3 Dec 2016 10:42:41 +0000]
Improve efficiency
The commands usermove, time and otim are now matched first, because
these are the commands used during game play, when response time and
CPU usage are relevant concerns.
H.G.Muller [Thu, 1 Dec 2016 20:17:11 +0000]
Buffer 'setoption' commands until engine done thinking
UCI does not allow setting of options during search after all. So all
'setoption' commands are now first collected in a memory buffer. During
ponder or analysis the search is then stopped immediately to flush that
buffer to the engine, in a new routine Release(). This routine will
also (re)start a ponder search if one is called for. During thinking,
finishing the search has priority. So only Release() the buffer contents
when a search finishes (immediately after receiving 'bestmove').
To deal with race conditions we also flush the buffer in the GUI thread,
after waiting for thinking to finish, just in case.
The WB 'easy' and 'hard' command implicitly set the Ponder option,
and this is treated by the hack of injecting fake 'option' commands in the
input stream, which then are subjected to the same buffering scheme.
Because waiting for thinking termination might now not happen at all
(if we only receive commands that can be handled instantly), a test for
search termination was added immediately after reading the command,
so that every command can now be aware whether thinking is in progress.
H.G.Muller [Thu, 1 Dec 2016 13:53:48 +0000]
Fix instant commands during think after ponder hit
The new method of suspending command processing only after peeking at
the next command to see if it should terminate the search was not
yet working after a ponder hit turned a ponder search into thinking.
H.G.Muller [Thu, 1 Dec 2016 13:41:05 +0000]
Process all innocent commands immediately (fixes pause/resume)
To prevent that commands received during thinking will halt command
processing,eclipsing later instant commands, all commands that merely
alter a settings value in adapter or engine are now immediately executed.
UCI engines should be able to handle 'setoption' commands even during
thinking, so explicit or implicit (easy/hard) settings changes are
immediately relayed to the engine as well.
The 'pause' and 'resume' commands are now also treated immediately;
waiting with that until the engine is done thinking subverted the
intention of the 'pause' command.
Commands that would never be sent when an engine is thinking are still
left with the commands that require thinking to complete, as for those
thisdoes not matter.
H.G.Muller [Wed, 30 Nov 2016 13:17:55 +0000]
Also treat 'result' command immediately
WinBoard has the nasty habit of sending a 'result' command just before
'quit' when it is closed while the engine is thinking, without putting
the engine in force mode first. So this command has to abort thinking too.
To prevent this will start a ponder search, 'result' now puts us in
force mode.
H.G.Muller [Wed, 30 Nov 2016 11:09:00 +0000]
Implement UCI_AnalyseMode option
An engine option of the form "??I_AnalyseMode" is now assumed to be the
standard check option UCI_... or USI_... (OK, that is a bit flaky).
It will not be reported as option feature, and will be automatically set
during analysis.
H.G.Muller [Wed, 30 Nov 2016 08:22:55 +0000]
Treat EOF same as quit command
A EOF during reading from the GUI caused UCI2WB to exit through a different
path as a 'quit' command. Now the EOF detection just fakes a 'quit -1'
command was received, and then processes that as usual. This means it
also causes proper termination of the engine when it occurs during thinking
(sending 'stop'). The '-1' appendage to the command allows the 'quit'
handling to return an error code.
H.G.Muller [Wed, 30 Nov 2016 08:02:50 +0000]
Treat 'quit', 'force' and '?' commands immediately
The blocking of input during thinking made it also insensitive to 'quit'.
This would cause hanging engine processes when the GUI was closed and would
kill the adapter because it did not respond to 'quit' in time. The Sync()
call that waits for the engine's 'bestmove' is now deferred until after
reading input (executed conditionally through a 'think' flag, so that
it only happens during thinking), so that we can peek at the first command
that arrives during thinking. If it is one of the 'immediate' commands
(?, quit or force), a 'stop' command is sent to the engine before we
start waiting for the 'bestmove', which then never should take long.
Draw offers could potentially arrive during thinking as well, interefering
with this mechanism by eclipsing a following ?/quit/force. As these would
just set the drawOffer flag for processing during the next move, these
are now fully processed immediately.
H.G.Muller [Thu, 10 Nov 2016 10:38:08 +0000]
Print everything sent to engine in debug mode
Now the macro EPRINT makes this easy we make sure that everything sent to
the engine is also reported in a comment to the GUI, when the debug
option is on.
H.G.Muller [Thu, 10 Nov 2016 10:13:19 +0000]
Simplify debug printing
The duplicats of each print statement to the engine to (optionally) print
debug information are now generated automatically by a single macro EPRINT.
H.G.Muller [Thu, 10 Nov 2016 09:55:41 +0000]
Print newline after 'position moves' instead of before 'go'
The '\n' between position-moves and go was printed with the latter
to avoid a separate printf for it after the move list, but this obfuscated
the code, and was a show stopper for more elegant printing of debug info.
H.G.Muller [Thu, 10 Nov 2016 09:12:11 +0000]
Fix commands during analysis (MultiPV option!)
UCI2WB could not handle spurious commands while analyzing: these would
start a new search without first terminating the old one. Only the commands
allowed in analysis mode (exit, usermove, and in/exclude) were handled
correctly. Unfortunately XBoard sometimes sends spurious 'accepted' commands
during analysis, in violation of the CECP specs. All such non-compliant
commands are now ignored. And the 'option' command will also abort an
ongoing analysis search, to allow altering the MultiPV setting during
analysis. (Something XBoard also does.)
H.G.Muller [Thu, 10 Nov 2016 08:55:40 +0000]
Suppress 'setup' command for some standard variants
Receiving an 'info string variant' from the engine now not automatically
leads to emission of a 'setup' command to the GUI. Because currently such
a setup command contains defaults for info not supplied in the 'info'
command, such as holdings size and pieceToChar, we suppress this in some
standard variants where this would be harmful (such as crazyhouse).
Such that the GUI does not get wrongly configured when playing these
variants with legality testing off. Before, the setup command was
only suppressed in variant normal.
H.G.Muller [Tue, 8 Nov 2016 21:29:32 +0000]
Bump version to 2.3
H.G.Muller [Tue, 8 Nov 2016 21:27:45 +0000]
Define strcasestr for Windows
The MinGW compiler apparently does not have this function.
H.G.Muller [Tue, 8 Nov 2016 21:01:16 +0000]
Update README file (changelog) and man page
H.G.Muller [Sun, 6 Nov 2016 13:29:08 +0000]
Implement CECP 'setup' command
On reception of an "info string variant NAME startpos FEN" from the engine
UCI2WB will now send "setup (-) 8x8+0_fairy FEN" to the GUI, to make
engine-defined variants also work with UCI engines. Correct operation
required sending of an 'isready' command to the engine after setting
the UCI_Variant option (which evokes the 'info string' response),
and stalling processing of GUI commands until the 'readyok' response,
to prevent early processing of a following 'ping' before the engine
got time to submit the reply. This way the 'info string' response
can be optional. When variant ="chess" the 'info string' is ignored,
as such a setting of UCI_Variant only occurs in response to 'new',
and never to 'variant'.
H.G.Muller [Sun, 6 Nov 2016 11:29:42 +0000]
Configure XBoard to use UCI2WB as uxiAdapter in make-install
The install target of the Makefile now invokes XBoard to set its
-uxiAdapter option for the current user to a value suitable for
usig UCI2WB.
H.G.Muller [Sun, 6 Nov 2016 11:15:31 +0000]
Implement CECP 'computer' command
Remember if the 'computer' command was received from the GUI, and use
this info when setting the UCI_Opponent option to distinguish 'human'
and 'computer' as first word, instead of the uniformative 'none'.
The implemntation assumes 'computer' would be sent before 'name', which
XBoard does (but the CECP specs do not guarantee).
H.G.Muller [Sun, 6 Nov 2016 10:54:30 +0000]
Support UCI_Opponent option
The UCI_Opponent option is now recognized, and its option feature is
suppressed. Instead its presence is remembered, and will lead to its
setting on reception of a CECP 'name' command. The feature name=1
is emitted at startuptosolicitsuch a command.
H.G.Muller [Sat, 5 Nov 2016 12:54:10 +0000]
Do not transform FEN when UCI_Variant option is supported
For USI FENs have to be translated to SFEN by putting the holdings
after the side-to-move indicator. This was done for any FEN that specified
holdings (probably to accomodate the Seirawan Chess version of Komodo).
But in Crazyhouse for Stockfish this wrecks things. The announcement
of a UCI_Variant option now suppresses this transformation.
H.G.Muller [Sat, 5 Nov 2016 12:45:51 +0000]
Implement UCI_Variant option
A UCI combo option UCI_Variant is now recognized as special, and translated
to 'feature variant=...' rather than 'feature option="UCI_Variant ..."'.
The CECP 'variant' command now leads to setting of this option, while on
the 'new' command the option is reset to 'chess'.
The variant names are transmitted as given, except for 'chess',
which is traslated to 'normal', and 'threecheck' (CECP: 3check).
H.G.Muller [Wed, 21 Sep 2016 21:41:52 +0000]
Bump version to 2.2
H.G.Muller [Wed, 21 Sep 2016 21:41:19 +0000]
Update README file
H.G.Muller [Wed, 21 Sep 2016 21:20:02 +0000]
Add description of -var option to man page
H.G.Muller [Wed, 21 Sep 2016 21:03:39 +0000]
Update man page
The manual page is updated with
- mentioning of the Arena960 dialect and -f flag to trigger it
- discussion of byoyomi TC, and the options to force it
- replacing the defunct USI specs URL by a valid one on my own website
H.G.Muller [Wed, 21 Sep 2016 13:05:13 +0000]
Subtract byoyomi from wtime and btime
USI engines think byoyomi comes in addition to their time on the clock,
so the time passed from the GUI in CECP must be decreased by the byoyomi
to let the engine know how close it is to forfeit.
H.G.Muller [Wed, 21 Sep 2016 12:42:20 +0000]
Send game result to USI engines
The USI 'gameover' command is now used as translation of the CECP
'result' command, to relay the result to the engine. To decide how
the absolute game result (white/black) is translated to a relative one
(win/lose), the side last moved for is used as indication, and remembered
whenever the engine is set thinking on a move (as it could already be
switched to force mode at the time the 'result' command is received).
H.G.Muller [Wed, 21 Sep 2016 12:26:07 +0000]
Handle USI win claims
USI engines can emit 'bestmove win' to claim a win, e.g. because
of the point count in an impasse situation. This is traslated to
a CECP result claim in favor of the side to move, with REASON "claim".
H.G.Muller [Wed, 21 Sep 2016 12:22:36 +0000]
Split byoyomi options over lines
H.G.Muller [Wed, 7 Sep 2016 18:56:54 +0000]
Bump version to 2.1
H.G.Muller [Wed, 7 Sep 2016 18:44:25 +0000]
Add option Byoyomi
In USI (-s) mode, UCI2WB now announces an option Byoyomi, which can
be used to define a hard byoyomi when the the option Floating Byoyomi
is off. Setting it to N is equivalent to a third argument " byoyomi N".
With N < 0 the option is switched off.
H.G.Muller [Mon, 5 Sep 2016 19:48:22 +0000]
Add option 'Floating Byoyomi'
In USI (-s) mode an option 'Floating Byoyomi' will be announced, which
will force the 'suffix' to the 'go' command to specify a byoyomi equal
to a reasonable time to think (timeLeft/40 + inc). This can be used as
a work-around to let engines that ignore the time left and always think
they are playing in byoyomi (a common misery in USI engines) work reasonably
well under incrementalor sudden-death TC. This then removes the need to
specify the suffix on the UCI2WB command line.
H.G.Muller [Wed, 7 Sep 2016 18:52:48 +0000]
Make suffix an array
The suffix to the 'go' command is now declared as array rather than
pointer, so that it can be easily printed in. (Before it could be made
to point to argv[3] if needed, and be NULL otherwise, but we want to
add some options that would create a suffix dynamically.)
H.G.Muller [Mon, 5 Sep 2016 19:33:03 +0000]
Put LDFLAGS at end of compile command
Pointed out by Logan Rosen.
H.G.Muller [Tue, 14 Jun 2016 09:07:29 +0000]
Supply time and node count for engines that do not give it
By default thinking time and nodecount were set to 0, but this made
XBoard interpret Thinking Output with score 0 as an info string, which
was undesirable. So the node count is now 1 by default, while the time
could be measured by UCI2WB itself (as it was already recoding startTime
of all searches).
H.G.Muller [Sat, 21 May 2016 16:38:58 +0000]
Join name and version when in separate 'id' commands
If an engine issues an "id version" command when an "id name" was already
received, a new 'myname' feature is emitted that contains both name and
version number. This is supposed to overrule the bare name in the myname
feature emitted because of the "id name" command.
H.G.Muller [Tue, 12 Apr 2016 18:38:24 +0000]
Set suspected non-standard Chess960 check options
Any check-type option containing the substring "960" or "frc" in their
name will be set automatically when UCI2WB is run with the -f option.
They will also be reported to the GUI, but always as having default
setting 'true', no matter what their original default setting was,
to reflect the new reality. So the potential damage this heuristic can
do is limited to altereing the default.
H.G.Muller [Mon, 4 Apr 2016 15:46:34 +0000]
Fix setting hash-table size in UCCI
The keyword 'name' in "setoption name Hash ..." was hard-coded, but in
UCCI it needs to be left out. So the variable nameWord is now used instead.
H.G.Muller [Mon, 4 Apr 2016 11:32:06 +0000]
Support Chess960
AIf an engine annunces an option UCI_Chess960 UCI2WB will report to
the GUI it supports variant fischerandom. In addition it will request
the GUI to encode castling moves as KxR through feature oocastle=0.
A flag argument -f will make UCI2WB support FRC even when the engine
does not have the option. In that case the GUI is set to deliver castling
moves as O-O[-O] (feature oocastle=1), and any castling field in a
received FEN will be overwritten with KQkq. (This is the obsolete
'Arena standard' for FRC.)
H.G.Muller [Wed, 7 Jan 2015 16:50:20 +0000]
Fix blocking of input in ponder games
The blocking of input during thinking did break pondering, as a ponder
miss also used pause = 2 to schedule ignoring of the 'bestmove' reply
to 'stop'. This now uses pause = 1 again, while after a ponder hit the
input thread is blocked after turning pause into 2, indicating the
'bestmove' is for real.
H.G.Muller [Tue, 23 Dec 2014 21:13:50 +0000]
Send ucinewgame after isready/readyok handshake
H.G.Muller [Tue, 23 Dec 2014 20:32:36 +0000]
Block processing of CECP commands during thinking
The GUI thread is blocked during thinking until a move is printed,
so that ping can be instantly replied to when it is processed,
without any need for isready/readyok.
H.G.Muller [Fri, 21 Nov 2014 08:06:38 +0000]
Specify license and include GPL text
H.G.Muller [Thu, 30 Oct 2014 14:52:29 +0000]
Also use new start FEN in UCCI after Pawn push
Irreversible (= forward) Pawn moves now also are recognized as
irreversible moves, next to captures.
H.G.Muller [Thu, 30 Oct 2014 14:16:49 +0000]
Pack Windows icon with distro tar ball
H.G.Muller [Thu, 30 Oct 2014 11:46:51 +0000]
Add Linux Makefile
H.G.Muller [Thu, 30 Oct 2014 11:46:19 +0000]
Add Linux manual page
H.G.Muller [Thu, 30 Oct 2014 09:13:18 +0000]
Print feature sigint=0 sigterm=0
Better to not even let XBoard send the signals we are going to ignore,
as this would just waste system resources.
H.G.Muller [Wed, 29 Oct 2014 21:46:59 +0000]
Don't forget to quit engine when GUI dies unexpectedly
When we get an EOF signal from the GUI pipe, meaning that the GUI
died without sending a 'quit' command, we'd better shutdown the engine
before exiting ourself.
H.G.Muller [Wed, 29 Oct 2014 21:43:46 +0000]
Make UCI2WB resistent to SIGTERM in Linux
UCI2WB was already resistant to SIGINT, which XBoard sends with every
usermove, but must also ignore the SIGTERM that is sent with the quit
command in order to reliably relay this quit to the engine.
H.G.Muller [Tue, 28 Oct 2014 20:02:56 +0000]
Bump version to 2.0
H.G.Muller [Tue, 28 Oct 2014 19:45:23 +0000]
Work-around for move during ponder
When engines, in violation of UCI specs, moved during ponder without
waiting for 'stop' or 'ponderhit', UCI2WB would pass their move out
of turn to the GUI, and hang when the opponent would move, waiting
for an engine move that had already come. We now ignore such a
premature ending of the ponder search, like no ponder search was
made at all. And we give an error message in the engine output.
H.G.Muller [Tue, 28 Oct 2014 11:57:14 +0000]
Fix UCCI draw offers
The offer must not be sent with the 'go ponder' command, but with 'ponderhit'.
H.G.Muller [Tue, 28 Oct 2014 08:22:20 +0000]
Update README file
H.G.Muller [Tue, 28 Oct 2014 08:08:54 +0000]
Suppress isready for first ping
The first ping sent by the XBoard should not probe the engine with
the isready/readyok dialog, but be answered with 'pong' immediately,
as the 'uciok' already issued an 'isready'.
H.G.Muller [Tue, 28 Oct 2014 07:44:34 +0000]
Remove possible deadlock
Add some flush() calls to make sure we are not waiting for a
response to something we still have buffered.
H.G.Muller [Mon, 27 Oct 2014 23:21:20 +0000]
Implement UCCI draw offers
The XBoard "offer draw" command is recognized, and leads to sending of
a 'draw' keyword with the 'go'. The other way a 'draw' in 'bestmove'
will lead to sending of an "offer draw" command.
H.G.Muller [Mon, 27 Oct 2014 23:12:50 +0000]
Implement the UCCI usemillisec option
This option is set (and msec are used) when the engine announced it
with default value 'false'.
H.G.Muller [Mon, 27 Oct 2014 23:09:52 +0000]
Implement UCCI newgame option
In UCCI 'newgame' is an engine-defined option, which should only be sent
when the engine has announced it.
H.G.Muller [Mon, 27 Oct 2014 22:59:02 +0000]
Implement UCCI
The main framework of UCCI now works, and is activated by the option -x :
We send 'ucci' and expect 'ucciok', the alternative keywords for timing
are used (time and opptime, increment and oppincrement), and 'name'
and 'value' are suppressed when setting options. 'name' is also not
mandatory anymore in the received 'option' commands. The position-moves
command starts with the FEN for after the latest capture, and now
has the 'position' keyword that was missing in Cyclone UCI.
H.G.Muller [Mon, 27 Oct 2014 22:43:11 +0000]
Print UCI keywords from variable, so they can be easily changed
The keyword wtime, btime, winc, binc, name, and value are all defined
as variable, to facilitate implementing UCCI, which uses different
words for these.
H.G.Muller [Mon, 27 Oct 2014 22:35:31 +0000]
Compactify timing code by figuring out who is black first
A new variable 'black' is set when the engine plays black, to know
if the XBoard 'time' setting should be transmitted as 'wtim' or 'btim'.
H.G.Muller [Mon, 27 Oct 2014 22:32:11 +0000]
Print 'ponder' always immediately behind 'go'
Some engines do not like it if you print it otherwise.
H.G.Muller [Mon, 27 Oct 2014 22:23:00 +0000]
Add infrastructure for keeping track of board position
A Xiangqi board is added, with code to set it up from a FEN, convert it
to the board part of a FEN, and apply the moves from the move list to it,
keeping track of what was the last capture. The FEN for setting it up
is always taken from the "fen <FEN>" position command ('iniPos') of the
Cyclone dialect used in variant xiangqi.
H.G.Muller [Mon, 27 Oct 2014 22:07:39 +0000]
Create infrastructure for accounting time in other units
A variable 'unit', initialized to 1, indicates how many msec a unit is.
H.G.Muller [Mon, 27 Oct 2014 22:46:39 +0000]
Fix extracting side-to-move from setboard command
When white was to move, the stm was just keeping its previous value,
which could very well have been BLACK.
H.G. Muller [Thu, 20 Dec 2012 14:00:57 +0000]
Connect WB cores command to UCI option Threads
H.G. Muller [Sun, 16 Dec 2012 08:42:12 +0000]
Implement Ponder option
UCI2WB now makes the possibility to ponder subject to the engine reporting
the Ponder option, and sets this option according to the WB ponder state.
H.G. Muller [Wed, 12 Dec 2012 16:52:21 +0000]
Allow faking of options the engine fails to announce
A file DefectiveEngineOptions.ini will be read as if it is engine output
before any real engine output. This can be used to fake UCI 'option' commands
that the engine fails to announce, so that they will appear in the GUI's
Engine Settings dialog, and can be changed by user or GUI.
H.G. Muller [Sun, 9 Dec 2012 17:50:41 +0000]
Fix pause in ponder-off games