Minimal Xboard

[jshriver]

tyvm. my engine (gasp) can finally play legal chess I hope w/o error. I'm a slow learner it seems as it took me a decade to get this far lol, but I can't wait till the day I can finally release it to everyone here for real grinding. All of these posts really help me.

-Josh

[Sven]

As an entry point: Sharper

Also: Roce where you can find the perfttestsuite.epd given above (also named perftsuite.epd by others hosting a copy of it).

Sven

[metax]

The quit command is not really needed; The GUI will kill the engine if it does not react to quit.

But engines that do not implement quit are really annoying in my opinion because most GUIs wait some time - up to 10 seconds depending on the GUI - before killing the engine.

[hgm]

Sure. Leaving out any command will in some way be annoying. But the question was what the minimal subset of essential commands was. Avoiding a 10-sec wait is not really essential. In long TC games you would not even notice it.

I don't want to encourage anyone to leave out implementation of anything. That in particular holds for the memory and cores commands!

[metax]

But if you want to test at fast TCs, it is _really_ annoying For example, in an 1+1 game the 10 secs would be about 8-10% of the time for one game...

[hgm]

Sure. But not having a level command is also really annoying, when you want to change the TC, and have to re-compile for it. Or not having a memory command is really annoying, when you want to change the hash size, and have to figure out how to edit an obscure ini file to do it...

[Sven]

But if you want to test at fast TCs, it is _really_ annoying For example, in an 1+1 game the 10 secs would be about 8-10% of the time for one game...

At least for WinBoard/xboard I am not sure whether the 10 seconds are the time the GUI waits for termination of the engine, or rather the default time the GUI waits until starting the next game of a match. I think WinBoard/xterm kills the engine with SIGTERM quite shortly after sending "quit". In the latter case you can simply influence it by using the "matchPause" option, e.g. /matchPause 500 sets the waiting time to 500 msec (whether half a second is sufficient mostly depends on the engine).

Furthermore, I think that the goal of testing at fast TCs is really an important one but is simply contradicting to the goal of supporting a minimal set of WB commands, which is the key point in this thread. To combine both would require to rephrase the question into something like: "What is the minimal set of WB commands in order to support testing at fast TCs?". In that case, "quit", "level", but also "setboard" (when using different start positions) would appear on the list, too.

Sven