Support syzygy tablebases

[codecov[bot]]

Codecov Report

Merging #29 into master will increase coverage by 0.21%. The diff coverage is 97.32%.

@@            Coverage Diff             @@
##           master      #29      +/-   ##
==========================================
+ Coverage   90.67%   90.89%   +0.21%     
==========================================
  Files          31       34       +3     
  Lines        1459     1549      +90     
==========================================
+ Hits         1323     1408      +85     
- Misses        136      141       +5

Impacted Files	Coverage Δ
include/search.hpp	100% <ø> (ø)	:arrow_up:
include/movegenerator.hpp	100% <ø> (ø)	:arrow_up:
include/protocol.hpp	100% <ø> (ø)	:arrow_up:
include/goldfish.hpp	100% <ø> (ø)	:arrow_up:
include/value.hpp	57.14% <ø> (ø)	:arrow_up:
src/searchmanagement.cpp	85.55% <100%> (+0.32%)	:arrow_up:
include/position.hpp	100% <100%> (ø)
src/notation.cpp	77.84% <100%> (+0.55%)	:arrow_up:
src/goldfish.cpp	92.06% <100%> (-0.17%)	:arrow_down:
src/search.cpp	96.42% <100%> (-0.87%)	:arrow_down:
... and 9 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more Δ = absolute <relative> (impact), ø = not affected, ? = missing data Powered by Codecov. Last update 07030c5...162ae6b. Read the comment docs.

[coveralls]

Coverage increased (+0.3%) to 91.925% when pulling 162ae6b957e6b1ba134938fe0b0f164f371c65e3 on syzygy-tb into 07030c51d30a2576da11c8cc98d9577f5bc32c70 on master.

[bsamseth]

About failing checks:

• LGTM fails due to LGTM. It's not Goldfish, it's just some crap.
• Codecov fails (rightfully) because coverage decreases.
  • But, this is only because the CI doesn't have a tablebase available, and so all the tests that would run are skipped.
  • Should consider making a table available and cached at travis
  • In the meantime, I promise it is covered by the tests when run with access to a tablebase. Cross my heart.
• Coveralls is same.

[bsamseth]

Results from playing 500 x (1+0), 500 x (1+1) and 100 x (3 + 2) games (minutes+increment).

It plays exactly equal in strength in total, with a ~7 elo drop for bullet, and corresponding increase for longer games. This should mean that the extra overhead is small, and that it pays for it self in the occational precise end game. I've also noticed a slight tendancy that v1.11.0 wins faster, and loses slower, when comparing the same opening played from both sides. This could be confirmation bias though, and should be aggregated over all the games to be certain.

Anyway, since strength is not measurably decreasing (.3 elo is exactly equal w.r.t the uncertainty we have), this will be merged and put on the Lichess bot server.

Note: All testing is for 5-man tables. Performance difference to 6-man has not been considered, and probably will not be included due to the large files neccesary.

2) Goldfish v1.11.0 2352.9 :   1100 (+371,=357,-372),  50.0 %

    vs.                     :  games (   +,   =,   -),   (%) :    Diff
    Goldfish v1.9.0         :   1100 ( 371, 357, 372),  50.0 :    -0.3

#	PLAYER	RATING	POINTS	PLAYED	(%)
1	Goldfish v1.9.0	2353.2	831.0	1530	54
2	Goldfish v1.11.0	2352.9	549.5	1100	50
3	Goldfish v1.7.0	2251.6	877.5	1699	52
4	Goldfish v1.7.1	2243.8	244.0	477	51
5	Goldfish v1.8.0	2237.1	324.0	650	50
6	Goldfish v1.8.1	2234.2	485.5	1000	49
7	Goldfish v1.8.2	2234.1	174.0	429	41
8	Goldfish v1.6.0	2218.8	434.5	797	55
9	Goldfish v1.7.2	2218.1	69.5	150	46
10	Goldfish v1.5.1	2168.9	460.5	970	47
11	Goldfish v1.5	2158.6	554.5	1145	48
12	Goldfish v1.4	2153.6	487.0	970	50
13	Goldfish v1.3	2130.8	162.5	325	50
14	Goldfish v1.2	2112.2	107.5	230	47
15	Goldfish v1.1	2055.2	93.0	232	40
16	Goldfish v1.0	2000.0	13.5	32	42