Decision concerning board representation.

[Evert]

Indeed. Bob claimed that Fruit also became stronger when he took out the history stuff.

Then I have some serious doubts about that claim. Removing history from Fruit also means removing history pruning.

Indeed.

I think the full claim is that the history condition on the reduction makes no difference, and you might as well reduce everything (except for the obvious exceptions) and be no worse. Bob can probably clarify.

- move ordering of quiet moves is important. it's not just a binary problem of putting the best first and anything else after, for which perhaps killers would be partially sufficient. at PV nodesordering well the quiet moves is very important.
- move ordering of quiet moves is even more important that with LMR you reduce quiet moves more as they have lower and lower history scores.

Yes, but not having history scores of course does not mean that you have no way to sort quiet moves...

[PK]

just for the record: I do not recommend doing away with history move ordering, Rodent uses it and gains because of it, ditto for history restriction on LMR. I'm only saying that for me using different bitscan orders for generating white and black moves was a useful tie-breaker in case of equal history scores (and just after starting the engine they are obviously equal). using piece lists alongside with bitboards loses this benefit, and I have been unable to get sufficient speedup to compensate for that fact. for that reason I would discourage mixing these board representations.

[lucasart]

Indeed. Bob claimed that Fruit also became stronger when he took out the history stuff.

Then I have some serious doubts about that claim. Removing history from Fruit also means removing history pruning.

Indeed.

I think the full claim is that the history condition on the reduction makes no difference, and you might as well reduce everything (except for the obvious exceptions) and be no worse. Bob can probably clarify.

That makes much more sense indeed. But history allows better search reduction, and Fruit 2.1 did not exploit that. It came with Glaurung, which refined Fruit's history pruning to reduce by an amount depending on the move count (or equivalently on the history score since we're talking quiet moves only here).