Give each game its own random.Random

[taziksh]

Problem

Game.__init__ calls random.seed(self.seed). The stream is shared by every game in the same process, so games running concurrently corrupt each other's draws and break reproducibility of seeds.

Change

• Game creates self.random = random.Random(seed) and passes it to State, which uses it everywhere the module global was used before.
• State.copy() and Game.copy() share the rng by reference, so simulations on copies advance the same stream, matching current behavior
• Bots now draw from game.state.random, instead of the same global random
• For backward compatibility with external callers, we include an optional rng parameter in roll_dice() and initialize_tiles that defaults to random
• Multiprocessing playouts now give each playout its own deterministic child seed. Previously, every pool worker unpickled the same rng state, so all playouts were identical

Tests

We tested by replaying seeded games with AlphaBeta, MCTS, and random bots, and checking the logs are identical to main. We also added test_seeded_games_are_isolated_from_each_other (fails on main, passes here) and a playout-diversity test, and fixed test_gym_reproducibility and the discard sequence test, which relied on the old global seeding.

[netlify]

👷 Deploy request for catanatron-staging pending review.

Visit the deploys page to approve it

Name	Link
🔨 Latest commit	89e9397

[bcollazo]

This is really cool! I'm a bit busy these days, but will try to review ASAP. If someone else can review / test that'd be great too!

Do games serialize nicely still to the Database? Does the GUI work nicely still?