Closed masterkni6 closed 3 years ago
crashing at real TC 1sec+0.1
It seems that if a search exits immediately (e.g. time exhaustion) then the BestMove is null and this can be dereferenced. However no replication sequence of nodes is obvious. But Ceres should be more defensive in this situation. Related, when presented with a terminal position Ceres aborts upon a "go" command (albeit with a clear message). Consider how to handle this. LC0 responds by outputting an invalid move "bestmove a1a1" which does not seem idea either.
This entire issue may just be a cutechess bug? The current position should have been adjudicated as draw but given to ceres anyways.
This issue is long since resolved.