SjaakII 1.2 RC1

[Evert]

I have now uploaded Sjaak II 1.2.0, available from http://www.eglebbk.dds.nl/program/chess-download.html

Source, Linux and OS X versions are up, Windows version will follow.

Changes since RC1:

• Add an option to change the way castling is sent to the GUI: Sjaak II normally sends these as O-O/O-O-O, but it can now be changed to e1g1/e1c1 or the equivalent in other variants. Normally this is not needed, but it may help in cases where Sjaak II and XBoard don't interpret "O-O" in the same way. Switching this on will break shuffle variants though, and it is normally not needed.
• Respond immediately to changes to the hashtable size in UCI mode, rather than waiting for a new game to start.
• Change "Marshall" and "Cardinal" to "Chancellor" and "Archbishop" in Great Chess, for compatibility.
• Add an option to change the variant that corresponds to variant "normal" (default: "chess"). This can be done from the engine options dialog, or it can be passed as a command line option ("-normal XXX"). This is not recommended, but it can be used as a workaround to play against other engines that play a variant under the name "normal" (UCI engines would fall under this category). This option is not persistent, so it resets when you restart the engine.
• Fix a number of bugs in the definition for Musketeer Chess
• Two new variants: "Spanish Chess"/"Archer Chess", which features pieces that move forward as a rook and backward as a bishop (fRbB) and "Enlarged and Improved Chess", which features pieces that move as a rook and capture as a bishop and vice versa (mRcB and mBcR).
• Minor tweaks and code cleanup.

Benchmark signature 4858390.

As usual, please let me know about missing or broken features.

[hgm]

I suppose changing the hash size clears the table? Or do you make an attempt to preserve the contents?

Spanish Chess indeed also looked funny. Perhaps I should add it to Fairy-Max too. What piece glyph did you choose for the Archer?

[Evert]

I suppose changing the hash size clears the table? Or do you make an attempt to preserve the contents?

Spanish Chess indeed also looked funny. Perhaps I should add it to Fairy-Max too. What piece glyph did you choose for the Archer?

I used the Dragon King (rook with a funny hat).
No particular reason for that, it seemed "rook-like". I could have gone for "bishop-like" instead.

Of course what I really wanted was the spear image, but that doesn't work correctly in the version I have.

EDIT: I called it "spanish_archer", for no other reason than "spanish" seemed a bit obscure...

[Alexander Schmidt]

As usual, please let me know about missing or broken features.

Hello Evert,

In FRC mode I get some illegal moves like this:

[Date "2016.01.09"]
[Round "1"]
[White "sjaakii"]
[Black "glaurung"]
[Result "0-1"]
[FEN "rnkbbrqn/pppppppp/8/8/8/8/PPPPPPPP/RNKBBRQN w KQkq - 0 1"]
[PlyCount "20"]
[SetUp "1"]
[TimeControl "240+2"]
[Variant "fischerandom"]

1. e4 {+0.10/17 11s} Ng6 {-0.11/14 9.3s} 2. Ng3 {+0.08/18 12s}
e5 {-0.13/14 8.1s} 3. Nf5 {+0.11/18 7.6s} Ne7 {-0.03/14 9.8s}
4. Nc3 {+0.25/16 5.9s} Nxf5 {+0.01/15 9.4s} 5. exf5 {+0.26/16 8.7s}
Nc6 {+0.07/14 13s} 6. f4 {+0.23/16 5.4s} f6 {+0.03/13 8.7s}
7. Qe3 {+0.23/14 8.3s} d5 {+0.35/12 8.9s} 8. Qg3 {-0.16/13 4.0s}
Be7 {+0.60/13 14s} 9. fxe5 {-0.17/13 3.5s} fxe5 {+0.56/13 7.9s}
10. d3 {-0.18/12 4.4s} O-O-O {+0.60/12 10s, White makes an illegal move: @@@@}
0-1

played with cutechess-qli.

Best wishes,
Alex

[Evert]

Odd. Do you happen to have a logfile of the engine <--> GUI (or cutechess, I suppose) communication?

SjaakII can make its own log if you start it with the "-log" option. Was it running in CECP or UCI mode?

[Evert]

Ok, I think I can actually reproduce the issue here, but I'm not actually sure what happened.

This is the position just before the end of the game:
[d]r1k1brq1/ppp1b1pp/2n5/3ppP2/8/2NP2Q1/PPP3PP/R1KBBR2 b KQkq - 0 10

Black played 10. ... O-O-O. Now, SjaakII's move list for this position gives

Code: Select all

...
 23/ 38 Qg8-f7     Qf7        g8f7      
 24/ 38 Qg8-h8     Qh8        g8h8      
 25/ 38 O-O-O      O-O-O      @@@@      
 26/ 38 Kc8-d7     Kd7        c8d7      
 27/ 38 Kc8-b8     Kb8        c8b8      
...

because source and destination are the same it sees the castle move as a null-move ("@@@@"), but it should send it as O-O-O in XBoard mode or as c8a8 in UCI mode (the standard long-algebraic would be "c8c8", but that gets replaced by a null move). Of course, it didn't play the move in this position, it received it. I'm guessing it was running in CECP mode and received O-O-O, otherwise it should have rejected the move as illegal - but then, on its move SjaakII apparently sent the long algebraic "@@@@" because it wanted to castle - but it shouldn't have done that in CECP mode. Even curiouser, the move list for that positions doesn't include castling, although castling rights are set correctly - so I don't know how it ever sent a castling move at all.

Anyway, I at least have something to go on. Thanks for testing!

[Evert]

Windows compiles by Martin are up.

There seems to be an issue with node-counts not matching up that is under investigation, and Alexander Schmidt found an issue with castling in FRC. Please let me know if anything else comes up.

[hgm]

I used the Dragon King (rook with a funny hat).
No particular reason for that, it seemed "rook-like". I could have gone for "bishop-like" instead.

Of course what I really wanted was the spear image, but that doesn't work correctly in the version I have.

In the latest XBoard the new pieces (when Pawn = 1) are:
22 Lion (already in 4.8; was King before)
23 Wolf
24 Camel
25 Zebra
26 Wizard
27 Amazon (Lance duplicat without Pawn character)
28 minor Lion (Lion duplicat without trading restriction)

After that the Chu-promoted series start:
29 Tokin (was 23 in 4.8)

A slash in the pieceToCharString fast-forwards to the Chu-promoted series, to make it possible to refer to pieces in this series in a way that is insensitive to adding more pieces at the end of the unpromoted series.

Unfortunately here is a compatibility issue here: making the pieceToChar so long that it uses the new pieces would select pieces from the Chu-promoted series in 4.8, and would be rejected (ignored) completely in earlier XBoard's. I don't see a good solution for this, other than not using old XBoards. Perhaps future XBoards should have a recipe for reassigning the end of overly long pieceToChar strings to unused pieces, so that they would be able to handle engines that are written for interfaces with even more pieces, but existing old XBoards do not have that.

I often abuse the Cannon glyph for pieces with Lance-like motion (e.g. Chu Shogi Reverse Chariot). The ideal glyph would be the Dolphin (because its shape with dorsal fin resembles the move), but it is deep in the Chu-promoted series.

[Evert]

I used the Dragon King (rook with a funny hat).
No particular reason for that, it seemed "rook-like". I could have gone for "bishop-like" instead.

Of course what I really wanted was the spear image, but that doesn't work correctly in the version I have.

In the latest XBoard the new pieces (when Pawn = 1) are:
22 Lion (already in 4.8; was King before)
23 Wolf
24 Camel
25 Zebra
26 Wizard
27 Amazon (Lance duplicat without Pawn character)
28 minor Lion (Lion duplicat without trading restriction)

Thanks, I'll add those to the sample piece-to-char I have at the top of the document. Do they have default characters associated with them (I guess certainly not all of them, since there aren't enough letters in the alphabet)?

Unfortunately here is a compatibility issue here: making the pieceToChar so long that it uses the new pieces would select pieces from the Chu-promoted series in 4.8, and would be rejected (ignored) completely in earlier XBoard's. I don't see a good solution for this, other than not using old XBoards. Perhaps future XBoards should have a recipe for reassigning the end of overly long pieceToChar strings to unused pieces, so that they would be able to handle engines that are written for interfaces with even more pieces, but existing old XBoards do not have that.

Well, you already lose functionality by running SjaakII on anything older than 4.8; I don't have a real issue with recommending 4.9 instead, but I should probably document what will break if you run it on something older.

[Evert]

I often abuse the Cannon glyph for pieces with Lance-like motion (e.g. Chu Shogi Reverse Chariot). The ideal glyph would be the Dolphin (because its shape with dorsal fin resembles the move), but it is deep in the Chu-promoted series.

Perhaps I should include that one too, just for the fun of it.

Did I ever send you the Omega Chess wizard/champion icons (a half moon and a fancy helmet) that I adapted from Wikipedia? Although I don't suppose you really need to cram too many additional glyphs in there...