It needs to be rank_attacks[8*64] at the minimum, to distinguish between the 8 files that the moving piece could be on. a1 vs a5 can be distinguished by shifting the result back. but a1 vs e1 are different attack sets for each occupancy set. the opposite is of course the case for file_attacks.

Oh and http://chessprogramming.wikispaces.com/ ... ce+Attacks provides a nice collection of the various options available for sliding piece attacks with bitboards. There's a lot of them, and personally i find rotated bitboards the most confusing of them.

A factor of 8 definitely matters for the sake of determining where your moving piece is in the occupancy set (you have different legal targets along the first rank from e1 than a1 if say c1 is occupied). You could probably shift it back along the other direction (though instead of multiply and ...

As you can see from looking at this game by Ivanchuk, this is extremely wild. But notice, 5.Qe2 would have stopped Black cold and forced Black on the defensive. All of that could have been given by analysis had both the Houska and Bauer books bothered to mention the Anderssen Counterattack in the ...

The precise way you have in mind requires a lot of state information be saved (since you want to potentially go back to the node at a later time), which is why going to full depth and then unwinding (refuting from the tail end as you see in the visualiser) is the typical process.

But saving energy ...

Uly, what you're describing closely resembles the idea of late move reductions/pruning present in all top engines right now. Once you've reached a certain place in the move list and suspect you're just going to fail-low anyways, further attempts are either ignored or searched at a reduced depth. If ...