mirror of
https://github.com/PaiGramTeam/genshin-wiki.git
synced 2024-11-25 01:21:57 +00:00
473 lines
16 KiB
Python
473 lines
16 KiB
Python
from functools import lru_cache
|
|
from itertools import chain
|
|
from typing import TypeVar
|
|
|
|
import ujson as json
|
|
from aiofiles import open as async_open
|
|
from humps import pascalize
|
|
from logging import getLogger
|
|
|
|
from model.avatar import (
|
|
AddProp,
|
|
AlternateSprint,
|
|
Avatar,
|
|
AvatarAttribute,
|
|
AvatarBirth,
|
|
AvatarConstellation,
|
|
AvatarInfo,
|
|
AvatarPromote,
|
|
AvatarStories,
|
|
AvatarTalents,
|
|
CombatTalent,
|
|
ElementalBurst,
|
|
ElementalSkill,
|
|
FirstAscensionPassive,
|
|
FourthAscensionPassive,
|
|
MiscellaneousPassive,
|
|
NormalAttack,
|
|
PassiveTalent,
|
|
Seuyu,
|
|
Story,
|
|
Talent,
|
|
TalentAttribute,
|
|
UtilityPassive,
|
|
)
|
|
from model.enums import Association, AvatarQuality, Element, PropType, WeaponType
|
|
from model.other import ItemCount
|
|
from utils.const import PROJECT_ROOT
|
|
from utils.funcs import remove_rich_tag
|
|
from utils.manager import ResourceManager
|
|
from utils.typedefs import Lang
|
|
|
|
try:
|
|
import regex as re
|
|
except ImportError:
|
|
import re
|
|
|
|
logger = getLogger("scripts.avatar")
|
|
|
|
TalentType = TypeVar("TalentType", bound=Talent)
|
|
|
|
OUT_DIR = PROJECT_ROOT.joinpath("out")
|
|
OUT_DIR.mkdir(parents=True, exist_ok=True)
|
|
|
|
prop_type_map = {
|
|
"Hp": PropType.HP,
|
|
"RockAddHurt": PropType.Geo,
|
|
"ElecAddHurt": PropType.Electro,
|
|
"FireAddHurt": PropType.Pyro,
|
|
"WaterAddHurt": PropType.Hydro,
|
|
"IceAddHurt": PropType.Cryo,
|
|
"WindAddHurt": PropType.Anemo,
|
|
"GrassAddHurt": PropType.Dendro,
|
|
}
|
|
|
|
|
|
def get_skill_attributes(
|
|
proud_skill_group_id: int,
|
|
) -> tuple[list[TalentAttribute], list[str]]:
|
|
proud_skill_datas = sorted(
|
|
filter(
|
|
lambda x: x["proudSkillGroupId"] == proud_skill_group_id,
|
|
proud_skill_json_data,
|
|
),
|
|
key=lambda x: x["level"],
|
|
)
|
|
result: list[TalentAttribute] = []
|
|
param_descriptions: list[str] = []
|
|
for proud_skill_data in proud_skill_datas:
|
|
param_descriptions = list(
|
|
filter(
|
|
lambda x: x is not None,
|
|
map(
|
|
lambda x: manager.get_text(x),
|
|
proud_skill_data["paramDescList"],
|
|
),
|
|
)
|
|
)
|
|
param_num = len(param_descriptions)
|
|
for param_description in param_descriptions:
|
|
param_num = max(
|
|
param_num,
|
|
*map(int, re.findall(r"param(\d*)\:", param_description)),
|
|
)
|
|
result.append(
|
|
TalentAttribute(
|
|
level=proud_skill_data["level"],
|
|
params=proud_skill_data["paramList"][:param_num],
|
|
break_level=proud_skill_data.get("breakLevel", 0),
|
|
coin=proud_skill_data.get("coinCost", 0),
|
|
cost_items=list(
|
|
map(
|
|
lambda x: ItemCount(item_id=x["id"], count=x["count"]),
|
|
filter(lambda x: x, proud_skill_data["costItems"]),
|
|
)
|
|
),
|
|
)
|
|
)
|
|
return result, param_descriptions
|
|
|
|
|
|
def parse_skill(skill_id: int, skill_cls: type[CombatTalent]) -> CombatTalent:
|
|
skill_data = next(filter(lambda x: x["id"] == skill_id, skill_json_data))
|
|
_name = manager.get_text(skill_data["nameTextMapHash"])
|
|
_description = manager.get_text(skill_data["descTextMapHash"])
|
|
icon = skill_data["skillIcon"]
|
|
cooldown = (
|
|
skill_data.get("cdTime", 0) if "cooldown" in skill_cls.__fields__ else None
|
|
)
|
|
attributes, param_descriptions = get_skill_attributes(
|
|
skill_data["proudSkillGroupId"]
|
|
)
|
|
return skill_cls(
|
|
**{
|
|
i[0]: i[1]
|
|
for i in zip(
|
|
[
|
|
"name",
|
|
"description",
|
|
"icon",
|
|
"cooldown",
|
|
"attributes",
|
|
"param_descriptions",
|
|
],
|
|
[_name, _description, icon, cooldown, attributes, param_descriptions],
|
|
)
|
|
if i is not None
|
|
}
|
|
)
|
|
|
|
|
|
def parse_passive_talent(
|
|
talent_data: dict, talent_cls: type[PassiveTalent]
|
|
) -> PassiveTalent:
|
|
group_id = talent_data["proudSkillGroupId"]
|
|
_promote_level = talent_data.get("needAvatarPromoteLevel", 0)
|
|
skill_data = next(
|
|
filter(lambda x: x["proudSkillGroupId"] == group_id, proud_skill_json_data)
|
|
)
|
|
param_descriptions = list(
|
|
filter(
|
|
lambda x: x is not None,
|
|
map(
|
|
lambda x: manager.get_text(x),
|
|
skill_data["paramDescList"],
|
|
),
|
|
)
|
|
)
|
|
_description = manager.get_text(skill_data["descTextMapHash"])
|
|
_param_list = skill_data["paramList"][
|
|
: len(re.findall(r"(\d+(?:\.)?\d+)", remove_rich_tag(_description) or ""))
|
|
]
|
|
return talent_cls(
|
|
name=manager.get_text(skill_data["nameTextMapHash"]) or "",
|
|
description=_description or "",
|
|
icon=skill_data["icon"],
|
|
promote_level=_promote_level,
|
|
param_descriptions=param_descriptions,
|
|
attribute=TalentAttribute(
|
|
params=_param_list,
|
|
break_level=skill_data.get("breakLevel", 0),
|
|
),
|
|
)
|
|
|
|
|
|
@lru_cache
|
|
def get_element_data() -> dict[Element, set[int]]:
|
|
_manager = ResourceManager("chs")
|
|
_avatar_json_data = _manager.fetch("AvatarExcelConfigData")
|
|
_fetter_info_json_data = manager.fetch("FetterInfoExcelConfigData")
|
|
text_map = {
|
|
"火": Element.Pyro,
|
|
"水": Element.Hydro,
|
|
"风": Element.Anemo,
|
|
"雷": Element.Electro,
|
|
"草": Element.Dendro,
|
|
"冰": Element.Cryo,
|
|
"岩": Element.Geo,
|
|
"无": Element.Null,
|
|
}
|
|
result = {k: set() for k in text_map.values()}
|
|
for data in _avatar_json_data:
|
|
_id = data["id"]
|
|
if (
|
|
info_data := next(
|
|
chain(
|
|
filter(lambda x: x["avatarId"] == _id, _fetter_info_json_data),
|
|
[None],
|
|
)
|
|
)
|
|
) is None:
|
|
continue
|
|
if (
|
|
vision := _manager.get_text(
|
|
text_id := info_data["avatarVisionBeforTextMapHash"]
|
|
)
|
|
) is not None:
|
|
result[text_map[vision]] = set(list(result[text_map[vision]]) + [text_id])
|
|
return result
|
|
|
|
|
|
# noinspection PyShadowingBuiltins,SpellCheckingInspection,PyGlobalUndefined
|
|
async def parse_avatar_data(lang: Lang):
|
|
global out_path, manager
|
|
global avatar_json_data, fetter_info_json_data, story_json_data, promote_json_data
|
|
global skill_depot_json_data, skill_json_data, proud_skill_json_data, talent_json_data
|
|
out_path = OUT_DIR.joinpath(f"{lang}")
|
|
out_path.mkdir(exist_ok=True, parents=True)
|
|
|
|
manager = ResourceManager(lang=lang)
|
|
avatar_json_data = manager.fetch("AvatarExcelConfigData")
|
|
fetter_info_json_data = manager.fetch("FetterInfoExcelConfigData")
|
|
story_json_data = manager.fetch("FetterStoryExcelConfigData")
|
|
promote_json_data = manager.fetch("AvatarPromoteExcelConfigData")
|
|
|
|
skill_depot_json_data = manager.fetch("AvatarSkillDepotExcelConfigData")
|
|
skill_json_data = manager.fetch("AvatarSkillExcelConfigData")
|
|
proud_skill_json_data = manager.fetch("ProudSkillExcelConfigData")
|
|
talent_json_data = manager.fetch("AvatarTalentExcelConfigData")
|
|
|
|
element_data = get_element_data()
|
|
|
|
avatar_list = []
|
|
for data in avatar_json_data:
|
|
id = data["id"]
|
|
if (
|
|
info_data := next(
|
|
chain(
|
|
filter(lambda x: x["avatarId"] == id, fetter_info_json_data), [None]
|
|
)
|
|
)
|
|
) is None:
|
|
continue
|
|
name = manager.get_text(data["nameTextMapHash"])
|
|
element = next(
|
|
filter(
|
|
lambda x: info_data["avatarVisionBeforTextMapHash"] in x[1],
|
|
element_data.items(),
|
|
)
|
|
)[0]
|
|
quality = AvatarQuality(
|
|
data["qualityType"].removeprefix("QUALITY_").replace("ORANGE_SP", "SPECIAL")
|
|
)
|
|
weapon_type = next(
|
|
filter(
|
|
lambda x: x in data["weaponType"].replace("POLE", "POLEARM"),
|
|
WeaponType.__members__.values(),
|
|
)
|
|
)
|
|
|
|
# 角色信息
|
|
title = manager.get_text(info_data["avatarTitleTextMapHash"])
|
|
birth = (
|
|
AvatarBirth(
|
|
month=info_data["infoBirthMonth"], day=info_data["infoBirthDay"]
|
|
)
|
|
if id not in [10000005, 10000007]
|
|
else None
|
|
)
|
|
occupation = manager.get_text(info_data["avatarNativeTextMapHash"])
|
|
vision = manager.get_text(info_data["avatarVisionBeforTextMapHash"])
|
|
constellation = manager.get_text(
|
|
info_data["avatarConstellationBeforTextMapHash"]
|
|
)
|
|
description = manager.get_text(info_data["avatarDetailTextMapHash"])
|
|
association = Association(
|
|
info_data["avatarAssocType"].removeprefix("ASSOC_TYPE_")
|
|
)
|
|
seuyu = Seuyu(
|
|
cn=manager.get_text(info_data["cvChineseTextMapHash"]),
|
|
jp=manager.get_text(info_data["cvJapaneseTextMapHash"]),
|
|
en=manager.get_text(info_data["cvEnglishTextMapHash"]),
|
|
kr=manager.get_text(info_data["cvKoreanTextMapHash"]),
|
|
)
|
|
story_datas = sorted(
|
|
filter(lambda x: x["avatarId"] == id, story_json_data),
|
|
key=lambda x: x["fetterId"],
|
|
)
|
|
stories = []
|
|
for story_data in sorted(
|
|
filter(lambda x: x["avatarId"] == id, story_datas),
|
|
key=lambda x: x["fetterId"],
|
|
):
|
|
tips = list(
|
|
filter(
|
|
lambda x: x is not None,
|
|
map(lambda x: manager.get_text(x), story_data["tips"]),
|
|
)
|
|
)
|
|
story = Story(
|
|
title=manager.get_text(story_data["storyTitleTextMapHash"]),
|
|
content=manager.get_text(story_data["storyContextTextMapHash"]),
|
|
tips=tips,
|
|
)
|
|
stories.append(story)
|
|
|
|
avatar_stories = AvatarStories(
|
|
**{i[0]: i[1] for i in zip(AvatarStories.__fields__, stories)}
|
|
)
|
|
information = AvatarInfo(
|
|
title=title,
|
|
birth=birth,
|
|
occupation=occupation,
|
|
vision=vision,
|
|
constellation=constellation,
|
|
description=description,
|
|
association=association,
|
|
seuyu=seuyu,
|
|
stories=avatar_stories,
|
|
)
|
|
|
|
# 角色基础属性
|
|
attributes = AvatarAttribute(
|
|
HP=data["hpBase"],
|
|
Attack=data["attackBase"],
|
|
Defense=data["defenseBase"],
|
|
Critical=data["critical"],
|
|
CriticalHurt=data["criticalHurt"],
|
|
ChargeEfficiency=data["chargeEfficiency"],
|
|
)
|
|
|
|
# 天赋
|
|
skill_depot_data = next(
|
|
filter(lambda x: x["id"] == data["skillDepotId"], skill_depot_json_data)
|
|
)
|
|
skill_ids = list(filter(lambda x: x != 0, skill_depot_data["skills"]))
|
|
# 普通攻击
|
|
normal_attack = parse_skill(skill_ids[0], NormalAttack)
|
|
# 冲刺技能
|
|
alternate_sprint = (
|
|
parse_skill(skill_ids[2], AlternateSprint) if len(skill_ids) == 3 else None
|
|
)
|
|
if id not in [10000005, 10000007]:
|
|
# 元素战技
|
|
elemental_skill = parse_skill(skill_ids[1], ElementalSkill)
|
|
# 元素爆发
|
|
burst_skill = parse_skill(skill_depot_data["energySkill"], ElementalBurst)
|
|
# 第一次突破被动天赋
|
|
first_passive = parse_passive_talent(
|
|
skill_depot_data["inherentProudSkillOpens"][0], FirstAscensionPassive
|
|
)
|
|
# 第四次突破被动天赋
|
|
fourth_passive = parse_passive_talent(
|
|
skill_depot_data["inherentProudSkillOpens"][1], FourthAscensionPassive
|
|
)
|
|
# 实用固有天赋
|
|
utility_passive = parse_passive_talent(
|
|
skill_depot_data["inherentProudSkillOpens"][2], UtilityPassive
|
|
)
|
|
else:
|
|
elemental_skill = (
|
|
burst_skill
|
|
) = first_passive = fourth_passive = utility_passive = None
|
|
# 杂项固有天赋
|
|
if skill_depot_data["inherentProudSkillOpens"][3]:
|
|
miscellaneous_passive = parse_passive_talent(
|
|
skill_depot_data["inherentProudSkillOpens"][3], MiscellaneousPassive
|
|
)
|
|
else:
|
|
miscellaneous_passive = None
|
|
# noinspection PyTypeChecker
|
|
avatar_talents = AvatarTalents(
|
|
normal_attack=normal_attack,
|
|
elemental_skill=elemental_skill,
|
|
elemental_burst=burst_skill,
|
|
alternate_sprint=alternate_sprint,
|
|
first_ascension_passive=first_passive,
|
|
fourth_ascension_passive=fourth_passive,
|
|
utility_passive=utility_passive,
|
|
miscellaneous_passive=miscellaneous_passive,
|
|
)
|
|
|
|
# 角色突破数据
|
|
promote_id = data["avatarPromoteId"]
|
|
promote_datas = sorted(
|
|
filter(lambda x: x["avatarPromoteId"] == promote_id, promote_json_data),
|
|
key=lambda x: x.get("promoteLevel", 0),
|
|
)
|
|
promotes = []
|
|
for promote_data in promote_datas:
|
|
items = []
|
|
for item_data in promote_data["costItems"]:
|
|
if item_data and id not in [10000005, 10000007]:
|
|
items.append(
|
|
ItemCount(item_id=item_data["id"], count=item_data["count"])
|
|
)
|
|
add_props = []
|
|
for add_prop_data in promote_data["addProps"]:
|
|
_string = pascalize(
|
|
add_prop_data["propType"]
|
|
.removeprefix("FIGHT_PROP_")
|
|
.removeprefix("BASE_")
|
|
.lower()
|
|
).replace("Hp", "HP")
|
|
prop_type_string = {
|
|
**prop_type_map,
|
|
**PropType.__members__,
|
|
**{v: v for k, v in PropType.__members__.items()},
|
|
}[_string]
|
|
prop_type = PropType(prop_type_string)
|
|
add_props.append(
|
|
AddProp(
|
|
type=prop_type,
|
|
value=add_prop_data.get(
|
|
"value", attributes.dict().get(prop_type, 0)
|
|
),
|
|
)
|
|
)
|
|
promotes.append(
|
|
AvatarPromote(
|
|
promote_level=promote_data.get("promoteLevel", 0),
|
|
max_level=promote_data["unlockMaxLevel"],
|
|
add_props=add_props,
|
|
coin=promote_data.get("scoinCost", 0),
|
|
cost_items=items,
|
|
)
|
|
)
|
|
|
|
# 角色命座信息
|
|
constellations: list[AvatarConstellation] = []
|
|
for constellation_id in filter(lambda x: x != 0, skill_depot_data["talents"]):
|
|
constellation_data = next(
|
|
filter(lambda x: x["talentId"] == constellation_id, talent_json_data)
|
|
)
|
|
constellation_description = manager.get_text(
|
|
constellation_data["descTextMapHash"]
|
|
)
|
|
# noinspection PyTypeChecker
|
|
constellations.append(
|
|
AvatarConstellation(
|
|
name=manager.get_text(constellation_data["nameTextMapHash"]),
|
|
description=constellation_description,
|
|
icon=constellation_data["icon"],
|
|
param_list=list(
|
|
filter(lambda x: x != 0.0, constellation_data["paramList"])
|
|
),
|
|
)
|
|
)
|
|
avatar = Avatar(
|
|
id=id,
|
|
name=name,
|
|
element=element,
|
|
quality=quality,
|
|
weapon_type=weapon_type,
|
|
information=information,
|
|
attributes=attributes,
|
|
talents=avatar_talents,
|
|
promotes=promotes,
|
|
constellations=constellations,
|
|
)
|
|
avatar_list.append(avatar)
|
|
logger.info(f"成功爬取了 {name} 的数据")
|
|
|
|
async with async_open(out_path / "avatar.json", encoding="utf-8", mode="w") as file:
|
|
await file.write(
|
|
json.dumps(
|
|
[i.dict(exclude_none=True) for i in avatar_list],
|
|
ensure_ascii=False,
|
|
encode_html_chars=False,
|
|
indent=4,
|
|
),
|
|
)
|
|
return out_path, avatar_list
|