In [1]:
import spot
from buddy import bddtrue
spot.setup()

Support for games

The support for games is currently quite rudimentary, as Spot currently only uses those games in ltlsynt.

In essence, a game is just an ω-automaton with a property named state-player that stores the player owning each state. The players are named 0 and 1. The player owning a state can decide what the next transition from this state should be. The goal for player 1 is to force the play to be infinite and to satisfy the acceptance condition of the automaton, while the goal for player 0 is to prevent it by either forcing a finite play, or forcing an infinite play that does not satisfy the acceptance condition.

The support is currently restricted to games that use:

  • t acceptance: all infinite run are accepting, and player 0 can only win if it manages to force a finite play (this requires reaching states without successors).
  • parity acceptance of any form (max odd, max even, min odd, min even): player 0 can win if the maximal (or minimal) value seen infinitely often is even (or odd)

Creating games from scratch

Games can be created like any automaton.
Using set_state_players() will fix the state owners.

In [2]:
bdict = spot.make_bdd_dict();
game = spot.make_twa_graph(bdict)
game.new_states(9)
for (s, d) in ((0,1), (0, 3), 
               (1, 0), (1, 2),
               (2, 1), (2, 5),
               (3, 2), (3, 4), (3, 6),
               (6, 7),
               (7, 6), (7, 8),
               (8, 5)):
    game.new_edge(s, d, bddtrue)
spot.set_state_players(game, [True, False, True, False, True, True, True, False, False])
game.show('.g')  # Use "g" to hide the irrelevant edge labels.
Out[2]:
t [all] 0 0 I->0 1 1 0->1 3 3 0->3 1->0 2 2 1->2 3->2 4 4 3->4 6 6 3->6 2->1 5 5 2->5 7 7 6->7 7->6 8 8 7->8 8->5

The set_state_players() function takes a list of owner for each of the states in the automaton. In the output, states from player 0 use circles, ellispes, or rectangle with rounded corners (mnemonic: 0 is round) while states from player 1 have a losanse shape (1 has only straight lines).

State ownership can also be manipulated by the following functions:

In [3]:
spot.get_state_players(game)
Out[3]:
(True, False, True, False, True, True, True, False, False)
In [4]:
spot.get_state_player(game, 4)
Out[4]:
True
In [5]:
spot.set_state_player(game, 4, False)
game.show('.g')
Out[5]:
t [all] 0 0 I->0 1 1 0->1 3 3 0->3 1->0 2 2 1->2 3->2 4 4 3->4 6 6 3->6 2->1 5 5 2->5 7 7 6->7 7->6 8 8 7->8 8->5

Solving a game

Solving a game is done my calling solve_game(). This function actually dispatches to solve_safety_game() (if the acceptance is t) or to solve_parity_game() (if the acceptance is any parity condition). You may call these functions directly if desired, but using solve_game() makes it possible for future version of Spot to dispatch to some better function if we add some.

These functions will attach two additional vectors into the game automaton: one vector stores the winner of each state, and one vector stores (memory-less) strategy for each state, i.e., the transition that should always be taken by the owner of this state in order to win.

The return value of those functions is simply the winner for the initial state.

In [6]:
spot.solve_game(game)
Out[6]:
True

Calling the highlight_strategy() function can be used to decorate the game automaton using the winning regions and strategies. Below, green represent the winning region/strategy for player 1 and red those for player 0.

In [7]:
spot.highlight_strategy(game)
game.show('.g')
Out[7]:
t [all] 0 0 I->0 1 1 0->1 3 3 0->3 1->0 2 2 1->2 3->2 4 4 3->4 6 6 3->6 2->1 5 5 2->5 7 7 6->7 7->6 8 8 7->8 8->5

Input/Output

An extension of the HOA format makes it possible to store the state-player property. This allows us to read the parity game constructed by ltlsynt using spot.automaton() like any other automaton.

In [8]:
game = spot.automaton("ltlsynt --ins=a --outs=b -f '!b & GFa <-> Gb' --print-game-hoa |");
game
Out[8]:
Fin( ) & (Inf( ) | (Fin( ) & (Inf( ) | Fin( )))) [parity max odd 5] 4 4 I->4 10 10 4->10 !a 11 11 4->11 a 0 0 5 5 0->5 !a 6 6 0->6 a 5->0 1 1 1 6->1 1 1->6 a 7 7 1->7 !a 7->0 1 2 2 8 8 2->8 1 8->2 1 3 3 9 9 3->9 1 9->2 !b 9->3 b 10->0 !b 10->3 b 11->1 !b 11->3 b

In the graphical output, player 0 is represented by circles (or ellipses or rounded rectangles depending on the situations), while player 1's states are diamond shaped. In the case of ltlsynt, player 0 plays the role of the environment, and player 1 plays the role of the controler.

In the HOA output, a header spot-state-player (or spot.state-player in HOA 1.1) lists the owner of each state.

In [9]:
print(game.to_str('hoa'))
HOA: v1
States: 12
Start: 4
AP: 2 "b" "a"
acc-name: parity max odd 5
Acceptance: 5 Fin(4) & (Inf(3) | (Fin(2) & (Inf(1) | Fin(0))))
properties: trans-labels explicit-labels trans-acc colored complete
properties: deterministic
spot-state-player: 0 0 0 0 0 1 1 1 1 1 1 1
--BODY--
State: 0
[!1] 5 {1}
[1] 6 {1}
State: 1
[1] 6 {1}
[!1] 7 {1}
State: 2
[t] 8 {1}
State: 3
[t] 9 {1}
State: 4
[!1] 10 {1}
[1] 11 {1}
State: 5
[t] 0 {3}
State: 6
[t] 1 {4}
State: 7
[t] 0 {4}
State: 8
[t] 2 {3}
State: 9
[!0] 2 {3}
[0] 3 {4}
State: 10
[!0] 0 {3}
[0] 3 {3}
State: 11
[!0] 1 {3}
[0] 3 {3}
--END--

Here is the solution of this particular game.

In [10]:
spot.solve_parity_game(game)
Out[10]:
True
In [11]:
spot.highlight_strategy(game)
Out[11]:
Fin( ) & (Inf( ) | (Fin( ) & (Inf( ) | Fin( )))) [parity max odd 5] 4 4 I->4 10 10 4->10 !a 11 11 4->11 a 0 0 5 5 0->5 !a 6 6 0->6 a 5->0 1 1 1 6->1 1 1->6 a 7 7 1->7 !a 7->0 1 2 2 8 8 2->8 1 8->2 1 3 3 9 9 3->9 1 9->2 !b 9->3 b 10->0 !b 10->3 b 11->1 !b 11->3 b