Py-Hanabi/src/hanabi/live/compress.py

224 lines
8 KiB
Python

from typing import List, Union
import more_itertools
from hanabi import hanab_game
from hanabi.live import hanab_live
# use same BASE62 as on hanab.live to encode decks
BASE62 = "abcdefghijklmnopqrstuvwxyz0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZ"
# Helper method, iterate over chunks of length n in a string
def chunks(s: str, n: int):
for i in range(0, len(s), n):
yield s[i:i + n]
# exception thrown by decompression methods if parsing fails
class InvalidFormatError(ValueError):
pass
def compress_actions(actions: List[hanab_game.Action]) -> str:
min_type = 0
max_type = 0
if len(actions) != 0:
min_type = min(map(lambda a: a.type.value, actions))
max_type = max(map(lambda a: a.type.value, actions))
type_range = max_type - min_type + 1
def compress_action(action):
# We encode action values with +1 to differentiate
# null (encoded 0) and 0 (encoded 1)
value = 0 if action.value is None else action.value + 1
if action.type == hanab_game.ActionType.VoteTerminate:
# This is currently a hack, the actual format has a 10 here,
# but we cannot encode this
value = 0
try:
a = BASE62[type_range * value + (action.type.value - min_type)]
b = BASE62[action.target]
except IndexError as e:
raise ValueError("Encoding action failed, value too large, found {}".format(value)) from e
return a + b
return "{}{}{}".format(
min_type,
max_type,
''.join(map(compress_action, actions))
)
def decompress_actions(actions_str: str) -> List[hanab_game.Action]:
if not len(actions_str) >= 2:
raise InvalidFormatError("min/max range not specified, found: {}".format(actions_str))
try:
min_type = int(actions_str[0])
max_type = int(actions_str[1])
except ValueError as e:
raise InvalidFormatError(
"min/max range of actions not specified, expected two integers, found {}".format(actions_str[:2])
) from e
if not min_type <= max_type:
raise InvalidFormatError("min/max range illegal, found [{},{}]".format(min_type, max_type))
type_range = max_type - min_type + 1
if not len(actions_str) % 2 == 0:
raise InvalidFormatError("Invalid action string length: Expected even number of characters")
for (index, char) in enumerate(actions_str[2:]):
if char not in BASE62:
raise InvalidFormatError(
"Invalid character at index {}: Found {}, expected one of {}".format(
index, char, BASE62
)
)
def decompress_action(action_idx: int, action: str):
try:
action_type_value = (BASE62.index(action[0]) % type_range) + min_type
action_type = hanab_game.ActionType(action_type_value)
except ValueError as e:
raise InvalidFormatError(
"Invalid action type at action {}: Found {}, expected one of {}".format(
action_idx, action_type_value,
[action_type.value for action_type in hanab_game.ActionType]
)
) from e
# We encode values with +1 to differentiate null (encoded 0) and 0 (encoded 1)
value = BASE62.index(action[0]) // type_range - 1
if value == -1:
value = None
if action_type in [hanab_game.ActionType.Play, hanab_game.ActionType.Discard]:
if value is not None:
raise InvalidFormatError(
"Invalid action value: Action at action index {} is Play/Discard, expected value None, "
"found: {}".format(action_idx, value)
)
target = BASE62.index(action[1])
return hanab_game.Action(action_type, target, value)
return [decompress_action(idx, a) for (idx, a) in enumerate(chunks(actions_str[2:], 2))]
def compress_deck(deck: List[hanab_game.DeckCard]) -> str:
assert (len(deck) != 0)
min_rank = min(map(lambda card: card.rank, deck))
max_rank = max(map(lambda card: card.rank, deck))
rank_range = max_rank - min_rank + 1
def compress_card(card):
try:
return BASE62[rank_range * card.suitIndex + (card.rank - min_rank)]
except IndexError as e:
raise InvalidFormatError(
"Could not compress card, suit or rank too large. Found: {}".format(card)
) from e
return "{}{}{}".format(
min_rank,
max_rank,
''.join(map(compress_card, deck))
)
def decompress_deck(deck_str: str) -> List[hanab_game.DeckCard]:
if len(deck_str) < 2:
raise InvalidFormatError("min/max rank range not specified, found: {}".format(deck_str))
try:
min_rank = int(deck_str[0])
max_rank = int(deck_str[1])
except ValueError as e:
raise InvalidFormatError(
"min/max rank range not specified, expected two integers, found {}".format(deck_str[:2])
) from e
if not max_rank >= min_rank:
raise InvalidFormatError(
"Invalid rank range, found [{},{}]".format(min_rank, max_rank)
)
rank_range = max_rank - min_rank + 1
for (index, char) in enumerate(deck_str[2:]):
if char not in BASE62:
raise InvalidFormatError(
"Invalid character at index {}: Found {}, expected one of {}".format(
index, char, BASE62
)
)
def decompress_card(card_char):
encoded = BASE62.index(card_char)
suit_index = encoded // rank_range
rank = encoded % rank_range + min_rank
return hanab_game.DeckCard(suit_index, rank)
return [decompress_card(card) for card in deck_str[2:]]
# compresses a standard GameState object into hanab.live format
# which can be used in json replay links
# The GameState object has to be standard / fitting hanab.live variants,
# otherwise compression is not possible
def compress_game_state(state: Union[hanab_game.GameState, hanab_live.HanabLiveGameState]) -> str:
if isinstance(state, hanab_live.HanabLiveGameState):
var_id = state.instance.variant_id
else:
assert isinstance(state, hanab_game.GameState)
var_id = hanab_live.HanabLiveInstance.select_standard_variant_id(state.instance)
out = "{}{},{},{}".format(
state.instance.num_players,
compress_deck(state.instance.deck),
compress_actions(state.actions),
var_id
)
with_dashes = ''.join(more_itertools.intersperse("-", out, 20))
return with_dashes
def decompress_game_state(game_str: str) -> hanab_live.HanabLiveGameState:
game_str = game_str.replace("-", "")
parts = game_str.split(",")
if not len(parts) == 3:
raise InvalidFormatError(
"Expected 3 comma-separated parts of game, found {}".format(
len(parts)
)
)
[players_deck, actions, variant_id] = parts
if len(players_deck) == 0:
raise InvalidFormatError("Expected nonempty first part")
try:
num_players = int(players_deck[0])
except ValueError as e:
raise InvalidFormatError(
"Expected number of players, found: {}".format(players_deck[0])
) from e
try:
deck = decompress_deck(players_deck[1:])
except InvalidFormatError as e:
raise InvalidFormatError("Error while parsing deck") from e
try:
actions = decompress_actions(actions)
except InvalidFormatError as e:
raise InvalidFormatError("Error while parsing actions") from e
try:
variant_id = int(variant_id)
except ValueError:
raise ValueError("Expected variant id, found: {}".format(variant_id))
instance = hanab_live.HanabLiveInstance(deck, num_players, variant_id)
game = hanab_live.HanabLiveGameState(instance)
# TODO: game is not in consistent state
game.actions = actions
return game
def link(game_state: hanab_game.GameState) -> str:
compressed = compress_game_state(game_state)
return "https://hanab.live/shared-replay-json/{}".format(compressed)