mirror of
https://github.com/LmeSzinc/StarRailCopilot.git
synced 2024-11-30 03:16:08 +00:00
620 lines
27 KiB
Python
620 lines
27 KiB
Python
import itertools
|
||
import os
|
||
import re
|
||
import typing as t
|
||
from collections import defaultdict
|
||
from functools import cache, cached_property
|
||
from hashlib import md5
|
||
|
||
from module.base.code_generator import CodeGenerator
|
||
from module.config.utils import deep_get, read_file
|
||
from module.logger import logger
|
||
|
||
UI_LANGUAGES = ['cn', 'cht', 'en', 'jp', 'es']
|
||
|
||
|
||
def text_to_variable(text):
|
||
text = re.sub("'s |s' ", '_', text)
|
||
text = re.sub('[ \-—:\'/•.]+', '_', text)
|
||
text = re.sub(r'[(),#"?!&%*]|</?\w+>', '', text)
|
||
# text = re.sub(r'[#_]?\d+(_times?)?', '', text)
|
||
return text.strip('_')
|
||
|
||
|
||
def dungeon_name(name: str) -> str:
|
||
name = text_to_variable(name)
|
||
name = re.sub('Bud_of_(Memories|Aether|Treasures)', r'Calyx_Golden_\1', name)
|
||
name = re.sub('Bud_of_(.*)', r'Calyx_Crimson_\1', name).replace('Calyx_Crimson_Calyx_Crimson_', 'Calyx_Crimson_')
|
||
name = re.sub('Shape_of_(.*)', r'Stagnant_Shadow_\1', name)
|
||
name = re.sub('Path_of_(.*)', r'Cavern_of_Corrosion_Path_of_\1', name)
|
||
if name in ['Destruction_Beginning', 'End_of_the_Eternal_Freeze', 'Divine_Seed']:
|
||
name = f'Echo_of_War_{name}'
|
||
return name
|
||
|
||
|
||
def blessing_name(name: str) -> str:
|
||
name = text_to_variable(name)
|
||
name = re.sub(r'^\d', lambda match: f"_{match.group(0)}", name)
|
||
return name
|
||
|
||
|
||
nickname_count = 0
|
||
|
||
|
||
def character_name(name: str) -> str:
|
||
name = text_to_variable(name)
|
||
name = re.sub('_', '', name)
|
||
return name
|
||
|
||
|
||
class TextMap:
|
||
DATA_FOLDER = ''
|
||
|
||
def __init__(self, lang: str):
|
||
self.lang = lang
|
||
|
||
def __contains__(self, name: t.Union[int, str]) -> bool:
|
||
if isinstance(name, int) or (isinstance(name, str) and name.isdigit()):
|
||
return int(name) in self.data
|
||
return False
|
||
|
||
@cached_property
|
||
def data(self) -> dict[int, str]:
|
||
if not os.path.exists(TextMap.DATA_FOLDER):
|
||
logger.critical('`TextMap.DATA_FOLDER` does not exist, please set it to your path to StarRailData')
|
||
exit(1)
|
||
file = os.path.join(TextMap.DATA_FOLDER, 'TextMap', f'TextMap{self.lang.upper()}.json')
|
||
data = {}
|
||
for id_, text in read_file(file).items():
|
||
text = text.replace('\u00A0', '')
|
||
text = text.replace(r'{NICKNAME}', 'Trailblazer')
|
||
data[int(id_)] = text
|
||
return data
|
||
|
||
def find(self, name: t.Union[int, str]) -> tuple[int, str]:
|
||
"""
|
||
Args:
|
||
name:
|
||
|
||
Returns:
|
||
text id (hash in TextMap)
|
||
text
|
||
"""
|
||
if isinstance(name, int) or (isinstance(name, str) and name.isdigit()):
|
||
name = int(name)
|
||
try:
|
||
return name, self.data[name]
|
||
except KeyError:
|
||
pass
|
||
|
||
name = str(name)
|
||
for row_id, row_name in self.data.items():
|
||
if row_id >= 0 and row_name == name:
|
||
return row_id, row_name
|
||
for row_id, row_name in self.data.items():
|
||
if row_name == name:
|
||
return row_id, row_name
|
||
logger.error(f'Cannot find name: "{name}" in language {self.lang}')
|
||
return 0, ''
|
||
|
||
|
||
def replace_templates(text: str) -> str:
|
||
"""
|
||
Replace templates in data to make sure it equals to what is shown in game
|
||
|
||
Examples:
|
||
replace_templates("Complete Echo of War #4 time(s)")
|
||
== "Complete Echo of War 1 time(s)"
|
||
"""
|
||
text = re.sub(r'#4', '1', text)
|
||
text = re.sub(r'</?\w+>', '', text)
|
||
return text
|
||
|
||
|
||
class KeywordExtract:
|
||
def __init__(self):
|
||
self.text_map: dict[str, TextMap] = {lang: TextMap(lang) for lang in UI_LANGUAGES}
|
||
self.text_map['cn'] = TextMap('chs')
|
||
self.keywords_id: list[int] = []
|
||
|
||
def iter_guide(self) -> t.Iterable[int]:
|
||
file = os.path.join(TextMap.DATA_FOLDER, './ExcelOutput/GameplayGuideData.json')
|
||
# visited = set()
|
||
temp_save = ""
|
||
for data in read_file(file).values():
|
||
hash_ = deep_get(data, keys='Name.Hash')
|
||
_, name = self.find_keyword(hash_, lang='cn')
|
||
if '永屹之城遗秘' in name: # load after all forgotten hall to make sure the same order in Game UI
|
||
temp_save = hash_
|
||
continue
|
||
if '忘却之庭' in name:
|
||
continue
|
||
# if name in visited:
|
||
# continue
|
||
# visited.add(name)
|
||
yield hash_
|
||
yield temp_save
|
||
# 'Memory of Chaos' is not a real dungeon, but represents a group
|
||
yield '混沌回忆'
|
||
|
||
def find_keyword(self, keyword, lang) -> tuple[int, str]:
|
||
"""
|
||
Args:
|
||
keyword: text string or text id
|
||
lang: Language to find
|
||
|
||
Returns:
|
||
text id (hash in TextMap)
|
||
text
|
||
"""
|
||
text_map = self.text_map[lang]
|
||
return text_map.find(keyword)
|
||
|
||
def load_keywords(self, keywords: list[str | int], lang='cn'):
|
||
text_map = self.text_map[lang]
|
||
keywords_id = [text_map.find(keyword)[0] for keyword in keywords]
|
||
self.keywords_id = [keyword for keyword in keywords_id if keyword != 0]
|
||
|
||
def clear_keywords(self):
|
||
self.keywords_id = []
|
||
|
||
def write_keywords(
|
||
self,
|
||
keyword_class,
|
||
output_file: str = '',
|
||
text_convert=text_to_variable,
|
||
generator: CodeGenerator = None,
|
||
extra_attrs: dict[str, dict] = None
|
||
):
|
||
"""
|
||
Args:
|
||
keyword_class:
|
||
output_file:
|
||
text_convert:
|
||
generator: Reuse an existing code generator
|
||
extra_attrs: Extra attributes write in keywords
|
||
"""
|
||
if generator is None:
|
||
gen = CodeGenerator()
|
||
gen.Import(f"""
|
||
from .classes import {keyword_class}
|
||
""")
|
||
gen.CommentAutoGenerage('dev_tools.keyword_extract')
|
||
else:
|
||
gen = generator
|
||
|
||
last_id = getattr(gen, 'last_id', 0)
|
||
if extra_attrs:
|
||
keyword_num = len(self.keywords_id)
|
||
for attr_key, attr_value in extra_attrs.items():
|
||
if len(attr_value) != keyword_num:
|
||
print(f"Extra attribute {attr_key} does not match the size of keywords")
|
||
return
|
||
for index, keyword in enumerate(self.keywords_id):
|
||
_, name = self.find_keyword(keyword, lang='en')
|
||
name = text_convert(replace_templates(name))
|
||
with gen.Object(key=name, object_class=keyword_class):
|
||
gen.ObjectAttr(key='id', value=index + last_id + 1)
|
||
gen.ObjectAttr(key='name', value=name)
|
||
for lang in UI_LANGUAGES:
|
||
gen.ObjectAttr(key=lang, value=replace_templates(self.find_keyword(keyword, lang=lang)[1]))
|
||
if extra_attrs:
|
||
for attr_key, attr_value in extra_attrs.items():
|
||
gen.ObjectAttr(key=attr_key, value=attr_value[keyword])
|
||
gen.last_id = index + last_id + 1
|
||
|
||
if output_file:
|
||
print(f'Write {output_file}')
|
||
gen.write(output_file)
|
||
self.clear_keywords()
|
||
return gen
|
||
|
||
def load_quests(self, quests, lang='cn'):
|
||
"""
|
||
Load a set of quest keywords
|
||
|
||
Args:
|
||
quests: iterable quest id collection
|
||
lang:
|
||
|
||
"""
|
||
quest_data = read_file(os.path.join(TextMap.DATA_FOLDER, 'ExcelOutput', 'QuestData.json'))
|
||
quests_hash = [quest_data[str(quest_id)]["QuestTitle"]["Hash"] for quest_id in quests]
|
||
quest_keywords = list(dict.fromkeys([self.text_map[lang].find(quest_hash)[1] for quest_hash in quests_hash]))
|
||
self.load_keywords(quest_keywords, lang)
|
||
|
||
def generate_daily_quests(self):
|
||
daily_quest = read_file(os.path.join(TextMap.DATA_FOLDER, 'ExcelOutput', 'DailyQuest.json'))
|
||
self.load_quests(daily_quest.keys())
|
||
self.write_keywords(keyword_class='DailyQuest', output_file='./tasks/daily/keywords/daily_quest.py')
|
||
|
||
def load_character_name_keywords(self, lang='en'):
|
||
file_name = 'ItemConfigAvatarPlayerIcon.json'
|
||
path = os.path.join(TextMap.DATA_FOLDER, 'ExcelOutput', file_name)
|
||
character_data = read_file(path)
|
||
characters_hash = [character_data[key]["ItemName"]["Hash"] for key in character_data]
|
||
|
||
text_map = self.text_map[lang]
|
||
keywords_id = sorted(
|
||
{text_map.find(keyword)[1] for keyword in characters_hash}
|
||
)
|
||
self.load_keywords(keywords_id, lang)
|
||
|
||
def generate_shadow_with_characters(self):
|
||
# Damage type -> damage hash
|
||
damage_info = dict()
|
||
for type_name, data in read_file(os.path.join(
|
||
TextMap.DATA_FOLDER, 'ExcelOutput',
|
||
'DamageType.json'
|
||
)).items():
|
||
damage_info[type_name] = deep_get(data, 'DamageTypeName.Hash')
|
||
# Character id -> character hash & damage type
|
||
character_info = dict()
|
||
for data in read_file(os.path.join(
|
||
TextMap.DATA_FOLDER, 'ExcelOutput',
|
||
'AvatarConfig.json'
|
||
)).values():
|
||
name_hash = deep_get(data, 'AvatarName.Hash')
|
||
damage_type = deep_get(data, 'DamageType')
|
||
character_info[data['AvatarID']] = (
|
||
name_hash, damage_info[damage_type])
|
||
# Item id -> character id
|
||
promotion_info = defaultdict(list)
|
||
for data in read_file(os.path.join(
|
||
TextMap.DATA_FOLDER, 'ExcelOutput',
|
||
'AvatarPromotionConfig.json'
|
||
)).values():
|
||
character_id = deep_get(data, '0.AvatarID')
|
||
item_id = deep_get(data, '2.PromotionCostList')[-1]['ItemID']
|
||
promotion_info[item_id].append(character_info[character_id])
|
||
# Shadow hash -> item id
|
||
shadow_info = dict()
|
||
for data in read_file(os.path.join(
|
||
TextMap.DATA_FOLDER, 'ExcelOutput',
|
||
'MappingInfo.json'
|
||
)).values():
|
||
farm_type = deep_get(data, '0.FarmType')
|
||
if farm_type != 'ELEMENT':
|
||
continue
|
||
shadow_hash = deep_get(data, '0.Name.Hash')
|
||
item_id = deep_get(data, '5.DisplayItemList')[-1]['ItemID']
|
||
shadow_info[shadow_hash] = promotion_info[item_id]
|
||
prefix_dict = {
|
||
'cn': '角色晋阶材料:',
|
||
'cht': '角色晉階材料:',
|
||
'jp': 'キャラクター昇格素材:',
|
||
'en': 'Ascension: ',
|
||
'es': 'Ascension: '
|
||
}
|
||
keyword_class = 'DungeonDetailed'
|
||
output_file = './tasks/dungeon/keywords/dungeon_detailed.py'
|
||
gen = CodeGenerator()
|
||
gen.Import(f"""
|
||
from .classes import {keyword_class}
|
||
""")
|
||
gen.CommentAutoGenerage('dev_tools.keyword_extract')
|
||
for index, (keyword, characters) in enumerate(shadow_info.items()):
|
||
_, name = self.find_keyword(keyword, lang='en')
|
||
name = text_to_variable(name).replace('Shape_of_', '')
|
||
with gen.Object(key=name, object_class=keyword_class):
|
||
gen.ObjectAttr(key='id', value=index + 1)
|
||
gen.ObjectAttr(key='name', value=name)
|
||
for lang in UI_LANGUAGES:
|
||
character_names = ' / '.join([
|
||
self.find_keyword(c[0], lang)[1]
|
||
for c in characters
|
||
])
|
||
damage_type = self.find_keyword(characters[0][1], lang)[1]
|
||
if lang in {'en', 'es'}:
|
||
value = f'{prefix_dict[lang]}{damage_type} ({character_names})'
|
||
else:
|
||
value = f'{prefix_dict[lang]}{damage_type}({character_names})'
|
||
gen.ObjectAttr(key=lang, value=value)
|
||
print(f'Write {output_file}')
|
||
gen.write(output_file)
|
||
self.clear_keywords()
|
||
|
||
def generate_forgotten_hall_stages(self):
|
||
keyword_class = "ForgottenHallStage"
|
||
output_file = './tasks/forgotten_hall/keywords/stage.py'
|
||
gen = CodeGenerator()
|
||
gen.Import(f"""
|
||
from .classes import {keyword_class}
|
||
""")
|
||
gen.CommentAutoGenerage('dev_tools.keyword_extract')
|
||
for stage_id in range(1, 16):
|
||
id_str = str(stage_id).rjust(2, '0')
|
||
with gen.Object(key=f"Stage_{stage_id}", object_class=keyword_class):
|
||
gen.ObjectAttr(key='id', value=stage_id)
|
||
gen.ObjectAttr(key='name', value=id_str)
|
||
for lang in UI_LANGUAGES:
|
||
gen.ObjectAttr(key=lang, value=id_str)
|
||
|
||
print(f'Write {output_file}')
|
||
gen.write(output_file)
|
||
self.clear_keywords()
|
||
|
||
def generate_assignments(self):
|
||
self.load_keywords(['空间站特派'])
|
||
self.write_keywords(
|
||
keyword_class='AssignmentEventGroup',
|
||
output_file='./tasks/assignment/keywords/event_group.py'
|
||
)
|
||
for file_name, class_name, output_file in (
|
||
('ExpeditionGroup.json', 'AssignmentGroup', './tasks/assignment/keywords/group.py'),
|
||
('ExpeditionData.json', 'AssignmentEntry', './tasks/assignment/keywords/entry.py'),
|
||
('ActivityExpedition.json', 'AssignmentEventEntry', './tasks/assignment/keywords/event_entry.py'),
|
||
):
|
||
file = os.path.join(TextMap.DATA_FOLDER, 'ExcelOutput', file_name)
|
||
self.load_keywords(deep_get(data, 'Name.Hash') for data in read_file(file).values())
|
||
self.write_keywords(keyword_class=class_name, output_file=output_file)
|
||
|
||
def generate_map_planes(self):
|
||
planes = {
|
||
'Special': ['黑塔的办公室', '锋芒崭露'],
|
||
'Rogue': [ '区域-战斗', '区域-事件', '区域-遭遇', '区域-休整', '区域-精英', '区域-首领', '区域-交易'],
|
||
'Herta': ['观景车厢', '主控舱段', '基座舱段', '收容舱段', '支援舱段'],
|
||
'Jarilo': ['行政区', '城郊雪原', '边缘通路', '铁卫禁区', '残响回廊', '永冬岭',
|
||
'造物之柱', '旧武器试验场', '磐岩镇', '大矿区', '铆钉镇', '机械聚落'],
|
||
'Luofu': ['星槎海中枢', '流云渡', '迴星港', '长乐天', '金人巷', '太卜司',
|
||
'工造司', '绥园', '丹鼎司', '鳞渊境'],
|
||
}
|
||
|
||
def text_convert(world_):
|
||
def text_convert_wrapper(name):
|
||
name = text_to_variable(name).replace('_', '')
|
||
name = f'{world_}_{name}'
|
||
return name
|
||
|
||
return text_convert_wrapper
|
||
|
||
gen = None
|
||
for world, plane in planes.items():
|
||
self.load_keywords(plane)
|
||
gen = self.write_keywords(keyword_class='MapPlane', output_file='',
|
||
text_convert=text_convert(world), generator=gen)
|
||
gen.write('./tasks/map/keywords/plane.py')
|
||
self.load_keywords(['Herta Space Station', 'Jarilo-VI', 'The Xianzhou Luofu'], lang='en')
|
||
self.write_keywords(keyword_class='MapWorld', output_file='./tasks/map/keywords/world.py')
|
||
|
||
def generate_character_keywords(self):
|
||
self.load_character_name_keywords()
|
||
self.write_keywords(keyword_class='CharacterList', output_file='./tasks/character/keywords/character_list.py',
|
||
text_convert=character_name)
|
||
|
||
def generate_battle_pass_quests(self):
|
||
battle_pass_quests = read_file(os.path.join(TextMap.DATA_FOLDER, 'ExcelOutput', 'BattlePassConfig.json'))
|
||
latest_quests = list(battle_pass_quests.values())[-1]
|
||
quests = deep_get(latest_quests, "DailyQuestList") + deep_get(latest_quests, "WeekQuestList") + deep_get(
|
||
latest_quests, "WeekOrder1")
|
||
self.load_quests(quests)
|
||
self.write_keywords(keyword_class='BattlePassQuest', output_file='./tasks/battle_pass/keywords/quest.py')
|
||
|
||
def generate_rogue_buff(self):
|
||
# paths
|
||
aeons = read_file(os.path.join(TextMap.DATA_FOLDER, 'ExcelOutput', 'RogueAeonDisplay.json'))
|
||
aeons_hash = [deep_get(aeon, 'RogueAeonPathName2.Hash') for aeon in aeons.values()]
|
||
self.keywords_id = aeons_hash
|
||
self.write_keywords(keyword_class='RoguePath', output_file='./tasks/rogue/keywords/path.py')
|
||
|
||
# blessings
|
||
blessings_info = read_file(os.path.join(TextMap.DATA_FOLDER, 'ExcelOutput', 'RogueBuff.json'))
|
||
blessings_name_map = read_file(os.path.join(TextMap.DATA_FOLDER, 'ExcelOutput', 'RogueMazeBuff.json'))
|
||
blessings_id = [deep_get(blessing, '1.MazeBuffID') for blessing in blessings_info.values()
|
||
if not deep_get(blessing, '1.AeonID')][1:]
|
||
resonances_id = [deep_get(blessing, '1.MazeBuffID') for blessing in blessings_info.values()
|
||
if deep_get(blessing, '1.AeonID')]
|
||
|
||
def get_blessing_infos(id_list, with_enhancement: bool):
|
||
blessings_hash = [deep_get(blessings_name_map, f"{blessing_id}.1.BuffName.Hash")
|
||
for blessing_id in id_list]
|
||
blessings_path_id = {blessing_hash: int(deep_get(blessings_info, f'{blessing_id}.1.RogueBuffType')) - 119
|
||
# 119 is the magic number make type match with path in keyword above
|
||
for blessing_hash, blessing_id in zip(blessings_hash, id_list)}
|
||
blessings_rarity = {blessing_hash: deep_get(blessings_info, f'{blessing_id}.1.RogueBuffRarity')
|
||
for blessing_hash, blessing_id in zip(blessings_hash, id_list)}
|
||
enhancement = {blessing_hash: "" for blessing_hash in blessings_hash}
|
||
if with_enhancement:
|
||
return blessings_hash, {'path_id': blessings_path_id, 'rarity': blessings_rarity,
|
||
'enhancement': enhancement}
|
||
else:
|
||
return blessings_hash, {'path_id': blessings_path_id, 'rarity': blessings_rarity}
|
||
|
||
hash_list, extra_attrs = get_blessing_infos(blessings_id, with_enhancement=True)
|
||
self.keywords_id = hash_list
|
||
self.write_keywords(keyword_class='RogueBlessing', output_file='./tasks/rogue/keywords/blessing.py',
|
||
text_convert=blessing_name, extra_attrs=extra_attrs)
|
||
|
||
hash_list, extra_attrs = get_blessing_infos(resonances_id, with_enhancement=False)
|
||
self.keywords_id = hash_list
|
||
self.write_keywords(keyword_class='RogueResonance', output_file='./tasks/rogue/keywords/resonance.py',
|
||
text_convert=blessing_name, extra_attrs=extra_attrs)
|
||
|
||
def generate_rogue_events(self):
|
||
# An event contains several options
|
||
event_title_file = os.path.join(
|
||
TextMap.DATA_FOLDER, 'ExcelOutput',
|
||
'RogueTalkNameConfig.json'
|
||
)
|
||
event_title_ids = {
|
||
id_: deep_get(data, 'Name.Hash')
|
||
for id_, data in read_file(event_title_file).items()
|
||
}
|
||
event_title_texts = defaultdict(list)
|
||
for title_id, title_hash in event_title_ids.items():
|
||
if title_hash not in self.text_map['en']:
|
||
continue
|
||
_, title_text = self.find_keyword(title_hash, lang='en')
|
||
event_title_texts[text_to_variable(title_text)].append(title_id)
|
||
option_file = os.path.join(
|
||
TextMap.DATA_FOLDER, 'ExcelOutput',
|
||
'DialogueEventDisplay.json'
|
||
)
|
||
option_ids = {
|
||
id_: deep_get(data, 'EventTitle.Hash')
|
||
for id_, data in read_file(option_file).items()
|
||
}
|
||
# Key: event name hash, value: list of option id/hash
|
||
options_grouped = dict()
|
||
# Key: option md5, value: option text hash in StarRailData
|
||
option_md5s = dict()
|
||
|
||
@cache
|
||
def get_option_md5(option_hash):
|
||
m = md5()
|
||
for lang in UI_LANGUAGES:
|
||
option_text = self.find_keyword(option_hash, lang=lang)[1]
|
||
m.update(option_text.encode())
|
||
return m.hexdigest()
|
||
|
||
# Drop invalid or duplicate options
|
||
def clean_options(options):
|
||
visited = set()
|
||
for i in options:
|
||
option_hash = option_ids[str(i)]
|
||
if option_hash not in self.text_map['en']:
|
||
continue
|
||
option_md5 = get_option_md5(option_hash)
|
||
if option_md5 in visited:
|
||
continue
|
||
if option_md5 not in option_md5s:
|
||
option_md5s[option_md5] = option_hash
|
||
visited.add(option_md5)
|
||
yield option_md5s[option_md5]
|
||
|
||
for group_title_ids in event_title_texts.values():
|
||
group_option_ids = []
|
||
for title_id in group_title_ids:
|
||
# Special case for Nildis (尼尔迪斯牌)
|
||
# Missing option: Give up
|
||
if title_id == '13501':
|
||
group_option_ids.append(13506)
|
||
option_id = title_id
|
||
# title ids in Swarm Disaster (寰宇蝗灾) have a "1" prefix
|
||
if option_id not in option_ids:
|
||
option_id = title_id[1:]
|
||
# Some title may not has corresponding options
|
||
if option_id not in option_ids:
|
||
continue
|
||
group_option_ids += list(itertools.takewhile(
|
||
lambda x: str(x) in option_ids,
|
||
itertools.count(int(option_id))
|
||
))
|
||
if group_option_ids:
|
||
title_hash = event_title_ids[group_title_ids[0]]
|
||
options_grouped[title_hash] = group_option_ids
|
||
|
||
for title_hash, options in options_grouped.items():
|
||
options_grouped[title_hash] = list(clean_options(options))
|
||
for title_hash in list(options_grouped.keys()):
|
||
if len(options_grouped[title_hash]) == 0:
|
||
options_grouped.pop(title_hash)
|
||
option_dup_count = defaultdict(int)
|
||
for option_hash in option_md5s.values():
|
||
if option_hash not in self.text_map['en']:
|
||
continue
|
||
_, option_text = self.find_keyword(option_hash, lang='en')
|
||
option_dup_count[text_to_variable(option_text)] += 1
|
||
|
||
def option_text_convert(option_md5, md5_prefix_len=4):
|
||
def wrapper(option_text):
|
||
option_var = text_to_variable(option_text)
|
||
if option_dup_count[option_var] > 1:
|
||
option_var = f'{option_var}_{option_md5[:md5_prefix_len]}'
|
||
return option_var
|
||
return wrapper
|
||
|
||
option_gen = None
|
||
option_hash_to_keyword_id = dict() # option hash -> option keyword id
|
||
for i, (option_md5, option_hash) in enumerate(option_md5s.items(), start=1):
|
||
self.load_keywords([option_hash])
|
||
option_gen = self.write_keywords(
|
||
keyword_class='RogueEventOption',
|
||
text_convert=option_text_convert(option_md5),
|
||
generator=option_gen
|
||
)
|
||
option_hash_to_keyword_id[option_hash] = i
|
||
output_file = './tasks/rogue/keywords/event_option.py'
|
||
print(f'Write {output_file}')
|
||
option_gen.write(output_file)
|
||
|
||
# title hash -> option keyword id
|
||
title_to_option_keyword_id = {
|
||
title_hash: sorted(
|
||
option_hash_to_keyword_id[x] for x in option_hashes
|
||
) for title_hash, option_hashes in options_grouped.items()
|
||
}
|
||
self.load_keywords(options_grouped.keys())
|
||
self.write_keywords(
|
||
keyword_class='RogueEventTitle',
|
||
output_file='./tasks/rogue/keywords/event_title.py',
|
||
extra_attrs={'option_ids': title_to_option_keyword_id}
|
||
)
|
||
try:
|
||
from tasks.rogue.event.event import OcrRogueEventOption
|
||
except AttributeError:
|
||
logger.critical(
|
||
f'Importing OcrRogueEventOption fails, probably due to changes in {output_file}')
|
||
try:
|
||
from tasks.rogue.event.preset import STRATEGIES
|
||
except AttributeError:
|
||
logger.critical(
|
||
f'Importing preset strategies fails, probably due to changes in {output_file}')
|
||
|
||
def iter_without_duplication(self, file: dict, keys):
|
||
visited = set()
|
||
for data in file.values():
|
||
hash_ = deep_get(data, keys=keys)
|
||
_, name = self.find_keyword(hash_, lang='cn')
|
||
if name in visited:
|
||
continue
|
||
visited.add(name)
|
||
yield hash_
|
||
|
||
def generate(self):
|
||
self.load_keywords(['模拟宇宙', '拟造花萼(金)', '拟造花萼(赤)', '凝滞虚影', '侵蚀隧洞', '历战余响', '忘却之庭'])
|
||
self.write_keywords(keyword_class='DungeonNav', output_file='./tasks/dungeon/keywords/nav.py')
|
||
self.load_keywords(['行动摘要', '生存索引', '每日实训'])
|
||
self.write_keywords(keyword_class='DungeonTab', output_file='./tasks/dungeon/keywords/tab.py')
|
||
self.load_keywords(['前往', '领取', '进行中', '已领取', '本日活跃度已满'])
|
||
self.write_keywords(keyword_class='DailyQuestState', output_file='./tasks/daily/keywords/daily_quest_state.py')
|
||
self.load_keywords(['领取', '追踪'])
|
||
self.write_keywords(keyword_class='BattlePassQuestState',
|
||
output_file='./tasks/battle_pass/keywords/quest_state.py')
|
||
self.load_keywords(list(self.iter_guide()))
|
||
self.write_keywords(keyword_class='DungeonList', output_file='./tasks/dungeon/keywords/dungeon.py',
|
||
text_convert=dungeon_name)
|
||
self.load_keywords(['传送', '追踪'])
|
||
self.write_keywords(keyword_class='DungeonEntrance', output_file='./tasks/dungeon/keywords/dungeon_entrance.py')
|
||
self.generate_shadow_with_characters()
|
||
self.load_keywords(['奖励', '任务', ])
|
||
self.write_keywords(keyword_class='BattlePassTab', output_file='./tasks/battle_pass/keywords/tab.py')
|
||
self.load_keywords(['本日任务', '本周任务', '本期任务'])
|
||
self.write_keywords(keyword_class='BattlePassMissionTab',
|
||
output_file='./tasks/battle_pass/keywords/mission_tab.py')
|
||
self.generate_assignments()
|
||
self.generate_forgotten_hall_stages()
|
||
self.generate_map_planes()
|
||
self.generate_character_keywords()
|
||
self.generate_daily_quests()
|
||
self.generate_battle_pass_quests()
|
||
self.load_keywords(['养成材料', '光锥', '遗器', '其他材料', '消耗品', '任务', '贵重物'])
|
||
self.write_keywords(keyword_class='ItemTab', text_convert=lambda name: name.replace(' ', ''),
|
||
output_file='./tasks/item/keywords/tab.py')
|
||
self.generate_rogue_buff()
|
||
self.load_keywords(['已强化'])
|
||
self.write_keywords(keyword_class='RogueEnhancement', output_file='./tasks/rogue/keywords/enhancement.py')
|
||
self.load_keywords(list(self.iter_without_duplication(
|
||
read_file(os.path.join(TextMap.DATA_FOLDER, 'ExcelOutput', 'RogueMiracleDisplay.json')),
|
||
'MiracleName.Hash')))
|
||
self.write_keywords(keyword_class='RogueCurio', output_file='./tasks/rogue/keywords/curio.py')
|
||
self.load_keywords(list(self.iter_without_duplication(
|
||
read_file(os.path.join(TextMap.DATA_FOLDER, 'ExcelOutput', 'RogueBonus.json')), 'BonusTitle.Hash')))
|
||
self.write_keywords(keyword_class='RogueBonus', output_file='./tasks/rogue/keywords/bonus.py')
|
||
self.generate_rogue_events()
|
||
|
||
|
||
if __name__ == '__main__':
|
||
TextMap.DATA_FOLDER = '../StarRailData'
|
||
KeywordExtract().generate()
|