example:

game = pyspiel.load_game('rbc', {'fen': '2kr1n1N/pp2p3/8/2P2pp1/2P1p3/4B3/r1PK1P2/8 b - - 1 25'})
state = game.new_initial_state().child(0)
print([state.action_to_string(a) for a in state.legal_actions()])

results include both Raxd2 and Rxd2, but should instead include Raxd2 and Rdxd2, since Rxd2 is ambiguous.

(the board is

. . k r . n . N
p p . . p . . .
. . . . . . . .
. . P . . p p .
. . P . p . . .
. . . . B . . .
r . P K . P . .
. . . . . . . .

)

This is due to ChessBoard's Move.ToSan() using only (pseudo)-legal moves to disambiguate, of which Raxd2 is not one, so it thinks disambiguation is not necessary.

Sorry to raise yet another issue for reconchess! 😛 I can make the fix for this one, but I'm not sure how to do it: pass an extra argument to ToSAN that's a vector of all moves to disambiguate between?

Or is the better fix to just switch ActionToString to use ToLAN instead? This might be the better choice, since some SAN concepts fall apart in recon chess anyways: e.g. an action could be either Qd2 or Qxd2, but with LAN the action only has one valid representation.