cdani wrote:As the bug happened in only one move situation,
...
If this is indeed what happened, then the solution is to reset the various thread-specific data structures in the main thread before starting the actual search.
Bravo, Ron !
Seems the craziest bugs can be corrected through logical thinking...
Thank you!! The source code release makes Andscacs eligible for participation in the next season of FOSCEC. Moreover, CCRL should change the engine name color from green to orange!
tpoppins wrote:Günther, the point of my post was the color -- something your carefully formatted text table fails to convey.
Here is what I see:
a) 4-CPU direct comparison between 0.921 and 0.92 is not available (ver. 0.92 not having been tested with 4CPUs)
b) However, a comparison between single CPU performance indicates that ver. 0.921 (based on some 300 games) is 23 ELO *below* that of 0.92.
c) Therefore the fact that 0.921 is leading the charts for 4CPUs is misleading, and probably only because 0.92 was not tested with 4CPUs.
This would support the hypothesis that 0.921 may have have introduced an issue, irrespective of the illegal move played in one of the TCEC games.
kasinp wrote:... irrespective of the illegal move played in one of the TCEC games.
The bug of the illegal move was there from very old versions. Only that seems that a lot of cores and bad luck are necessary to manifest it.
Got it, thank you for the clarification. I suppose this still leaves the question of a direct comparison between 0.92 and 0.921, for which I suppose 300+ games is probably still a small sample size.