Chess protocols and moves to avoid starting to search?

[phhnguyen]

When analysing a hard position / problem I usually see engines be stuck in a PV for very long time. As a human I can quickly guess there is no hope for that PV and want those engines to try other way / other PVs. In other words I need them to avoid starting their searches with some given moves (they are first moves of PVs I don't want those engines to search).

Is there anyways to do that, using chess GUI with Winboard / UCI protocols? If no should we add some new commands to protocols (your opinions)?

Thanks

[hgm]

You can exclude moves from analysis. For UCI there is the 'searchmoves' keyword in the 'go' command, which should be followed by a list of allowed moves. Many engines do not implement this, though (and thus are not truly UCI compatible).

In WB protocol there are the GUI->engine commands

exclude MOVE
include MOVE

which can be used to add or remove moves from the excluded set one by one. Or by use of the word 'all' for MOVE in/exclude all moves at once. (Changing the position clears this set). These commands are only sent to engines that announced at startup it supported them, through sending

feature exclude=1

From WinBoard/XBoard you can toggle the exclusion state of a move by playing it with the Crtl key pressed, or by grabbing the piece with a double-click. There will also be a header line in the Engine-Output window that gives an overview of excluded moves (prefixed by a '-' to indicate their exclusion) and re-included moves (prefixed by '+'). Right-clicking on those also toggles the state of the move. The word 'bestmove' is part of the list, and can be clicked to exclude the first move of the most-recently printed PV, while the word 'tail' symbolizes all moves not explicitly shown in the exclusion header.

[Ras]

Many engines do not implement this, though (and thus are not truly UCI compatible).

From a GUI point of view, the issue is that searchmoves is not optional in UCI, that's why it's not being announced in the options list after the "uci" command.

For broken UCI engines that don't implement this feature, the only way to know is to test it when starting the engine. The GUI can transmit a position where there is an easy and obvious move, like mate in 1, and then using searchmoves with another move. If the engine still reports the mating move as beginning of the PV, then it does not support searchmoves.

[hgm]

I consider it evil to hide defects of other software. (Erosion of standards, and such.)

[Ras]

I consider it evil to hide defects of other software. (Erosion of standards, and such.)

The alternative would be to shift the error handling to layer 8, which sucks even more.

[phhnguyen]

That is exactly what I need. Thanks. How silly I am to miss them from documents