mirror of
https://github.com/MarioSpore/Grinch-AP.git
synced 2025-10-21 12:11:33 -06:00
Core: Fill fix local logic conflict (#1271)
This commit is contained in:
48
Fill.py
48
Fill.py
@@ -24,7 +24,8 @@ def sweep_from_pool(base_state: CollectionState, itempool: typing.Sequence[Item]
|
||||
|
||||
def fill_restrictive(world: MultiWorld, base_state: CollectionState, locations: typing.List[Location],
|
||||
itempool: typing.List[Item], single_player_placement: bool = False, lock: bool = False,
|
||||
swap: bool = True, on_place: typing.Optional[typing.Callable[[Location], None]] = None) -> None:
|
||||
swap: bool = True, on_place: typing.Optional[typing.Callable[[Location], None]] = None,
|
||||
allow_partial: bool = False) -> None:
|
||||
unplaced_items: typing.List[Item] = []
|
||||
placements: typing.List[Location] = []
|
||||
|
||||
@@ -132,7 +133,7 @@ def fill_restrictive(world: MultiWorld, base_state: CollectionState, locations:
|
||||
if on_place:
|
||||
on_place(spot_to_fill)
|
||||
|
||||
if len(unplaced_items) > 0 and len(locations) > 0:
|
||||
if not allow_partial and len(unplaced_items) > 0 and len(locations) > 0:
|
||||
# There are leftover unplaceable items and locations that won't accept them
|
||||
if world.can_beat_game():
|
||||
logging.warning(
|
||||
@@ -252,11 +253,12 @@ def distribute_early_items(world: MultiWorld,
|
||||
fill_locations: typing.List[Location],
|
||||
itempool: typing.List[Item]) -> typing.Tuple[typing.List[Location], typing.List[Item]]:
|
||||
""" returns new fill_locations and itempool """
|
||||
early_items_count: typing.Dict[typing.Tuple[str, int], int] = {}
|
||||
early_items_count: typing.Dict[typing.Tuple[str, int], typing.List[int]] = {}
|
||||
for player in world.player_ids:
|
||||
items = itertools.chain(world.early_items[player], world.local_early_items[player])
|
||||
for item in items:
|
||||
early_items_count[(item, player)] = [world.early_items[player].get(item, 0), world.local_early_items[player].get(item, 0)]
|
||||
early_items_count[item, player] = [world.early_items[player].get(item, 0),
|
||||
world.local_early_items[player].get(item, 0)]
|
||||
if early_items_count:
|
||||
early_locations: typing.List[Location] = []
|
||||
early_priority_locations: typing.List[Location] = []
|
||||
@@ -280,42 +282,50 @@ def distribute_early_items(world: MultiWorld,
|
||||
for i, item in enumerate(itempool):
|
||||
if (item.name, item.player) in early_items_count:
|
||||
if item.advancement:
|
||||
if early_items_count[(item.name, item.player)][1]:
|
||||
if early_items_count[item.name, item.player][1]:
|
||||
early_local_prog_items[item.player].append(item)
|
||||
early_items_count[(item.name, item.player)][1] -= 1
|
||||
early_items_count[item.name, item.player][1] -= 1
|
||||
else:
|
||||
early_prog_items.append(item)
|
||||
early_items_count[(item.name, item.player)][0] -= 1
|
||||
early_items_count[item.name, item.player][0] -= 1
|
||||
else:
|
||||
if early_items_count[(item.name, item.player)][1]:
|
||||
if early_items_count[item.name, item.player][1]:
|
||||
early_local_rest_items[item.player].append(item)
|
||||
early_items_count[(item.name, item.player)][1] -= 1
|
||||
early_items_count[item.name, item.player][1] -= 1
|
||||
else:
|
||||
early_rest_items.append(item)
|
||||
early_items_count[(item.name, item.player)][0] -= 1
|
||||
early_items_count[item.name, item.player][0] -= 1
|
||||
item_indexes_to_remove.add(i)
|
||||
if early_items_count[(item.name, item.player)] == [0, 0]:
|
||||
del early_items_count[(item.name, item.player)]
|
||||
if early_items_count[item.name, item.player] == [0, 0]:
|
||||
del early_items_count[item.name, item.player]
|
||||
if len(early_items_count) == 0:
|
||||
break
|
||||
itempool = [item for i, item in enumerate(itempool) if i not in item_indexes_to_remove]
|
||||
for player in world.player_ids:
|
||||
player_local = early_local_rest_items[player]
|
||||
fill_restrictive(world, base_state,
|
||||
[loc for loc in early_locations if loc.player == player],
|
||||
early_local_rest_items[player], lock=True)
|
||||
[loc for loc in early_locations if loc.player == player],
|
||||
player_local, lock=True, allow_partial=True)
|
||||
if player_local:
|
||||
logging.warning(f"Could not fulfill rules of early items: {player_local}")
|
||||
early_rest_items.extend(early_local_rest_items[player])
|
||||
early_locations = [loc for loc in early_locations if not loc.item]
|
||||
fill_restrictive(world, base_state, early_locations, early_rest_items, lock=True)
|
||||
fill_restrictive(world, base_state, early_locations, early_rest_items, lock=True, allow_partial=True)
|
||||
early_locations += early_priority_locations
|
||||
for player in world.player_ids:
|
||||
player_local = early_local_prog_items[player]
|
||||
fill_restrictive(world, base_state,
|
||||
[loc for loc in early_locations if loc.player == player],
|
||||
early_local_prog_items[player], lock=True)
|
||||
[loc for loc in early_locations if loc.player == player],
|
||||
player_local, lock=True, allow_partial=True)
|
||||
if player_local:
|
||||
logging.warning(f"Could not fulfill rules of early items: {player_local}")
|
||||
early_prog_items.extend(player_local)
|
||||
early_locations = [loc for loc in early_locations if not loc.item]
|
||||
fill_restrictive(world, base_state, early_locations, early_prog_items, lock=True)
|
||||
fill_restrictive(world, base_state, early_locations, early_prog_items, lock=True, allow_partial=True)
|
||||
unplaced_early_items = early_rest_items + early_prog_items
|
||||
if unplaced_early_items:
|
||||
logging.warning("Ran out of early locations for early items. Failed to place "
|
||||
f"{len(unplaced_early_items)} items early.")
|
||||
f"{unplaced_early_items} early.")
|
||||
itempool += unplaced_early_items
|
||||
|
||||
fill_locations.extend(early_locations)
|
||||
|
Reference in New Issue
Block a user