#312 Online Tablebase Bug

[Joseph Huang]

Tested on 1.5.6.0 MAC64 and 1.5.4.0 MAC64

Steps to reproduce:

1. Copy the following FEN to your clipboard "7q/7P/1K6/1p6/8/k2Q4/8/8 b - - 28 80"
2. Open ChessX
3. In preferences, enable "Use online tablebase servers" = Syzygy-7
4. Edit > Paste into new game
5. Start Analysis
6. Make the move 80.. Ka2

Expected result:
Tablebase 81.Qg6 White Wins

Actual result:
Tablebase 81.Qg6 - Draw

Discussion:
If I visit Syzygy endgame tablebases website directly and click Ka2 it correctly shows "White is winning with DTZ 45 --> Qg6"

Other notes:
If I switch the "Use online tablebase servers" setting to Nalimov-6, everything works correctly.
Maybe there is some sort of parsing error when loading the remote result?

[Joseph Huang]

Reproduced on 1.5.0.0 WIN64 MSVC13 Qt5.12.3 also

(I had to restart ChessX after changing the "Use online tablebase servers" setting)

[Joseph Huang]

Okay, according to https://github.com/Isarhamster/chessx/blob/5e34a5508201078033453be8995c66394db35d5f/src/database/tablebase.cpp#L84 the Sygyzy API being used is https://github.com/lichess-org/lila-tablebase

Loading http://tablebase.lichess.ovh/standard?fen=7q/7P/1K6/1p6/8/3Q4/k7/8 w - - 29 81 directly in the browser correctly shows "win" which means the bug is somewhere in the ChessX receiving code

[Joseph Huang]

Inspecting the code directly
https://github.com/Isarhamster/chessx/blob/5e34a5508201078033453be8995c66394db35d5f/src/database/tablebase.cpp#L138
and comparing to
the API results: http://tablebase.lichess.ovh/standard?fen=7q/7P/1K6/1p6/8/3Q4/k7/8 w - - 29 81
we can expect...

jdtm.isNull()  ==  true
dtz  ==  45
bestScore  ==  45
(bestScore > 0) ? 1:-1  =  1
bestScore ? (bestScore > 0) ? 1:-1 : 0  ===  bestScore ? 1 : 0  =  1
result = 1
...
moves[0]:xjdtm.isNull()  ==  true
moves[0]:xdtz  ==  -44
moves[0]:xbestScore  ==  -44
(xbestScore > 0) ? -1:1  =  1
xbestScore ? (xbestScore > 0) ? -1:1:0  ===  xbestScore ? 1 : 0
moves[0]:m  ==  "d3g6"  // Qg6

So far, so good?

[Joseph Huang]

So then, let's see if it's a display issue
https://github.com/Isarhamster/chessx/blob/5e34a5508201078033453be8995c66394db35d5f/src/gui/analysiswidget.cpp#L510

Expecting...

score = 45 // from bestScore
dtm = false // because: -0x800 < score < 0x800
...

Notably, during the bug we're showing all the other alternative moves but they all have dtz = 0 and dtm = 0

So somehow,

int result = bestScore ? (bestScore > 0) ? 1:-1:0;

is evaluating to 0?

[Joseph Huang]

OKAY: I think then potentially this was fixed in Oct 21, 2021 here
https://github.com/Isarhamster/chessx/blame/0d19c71dafaa595670c15010befbafcadb15e574/src/database/tablebase.cpp

The old version relies on dtm which isn't always available in the Sygyzy tablebases
https://github.com/Isarhamster/chessx/blob/9086a00e2d433eb2ec3d1dd2c0bac2fdb0e46ad0/src/database/tablebase.cpp

[Joseph Huang]

Looks like it's working on chessx-1.5.7b.dmg 2022-05-26 15.5 MB

[Jens Nissen]

Fixed