PamGram/modules/wiki/weapon.py
2022-10-10 19:07:28 +08:00

148 lines
4.7 KiB
Python

import itertools
import re
from typing import List, Optional, Tuple, Union
from bs4 import BeautifulSoup
from httpx import URL
from modules.wiki.base import Model, HONEY_HOST, WikiModel
from modules.wiki.other import AttributeType, WeaponType
__all__ = ["Weapon", "WeaponAffix", "WeaponAttribute"]
class WeaponAttribute(Model):
"""武器词条"""
type: AttributeType
value: str
class WeaponAffix(Model):
"""武器技能
Attributes:
name: 技能名
description: 技能描述
"""
name: str
description: List[str]
class WeaponState(Model):
level: str
ATK: float
bonus: Optional[str]
class WeaponIcon(Model):
icon: str
awakened: str
gacha: str
class Weapon(WikiModel):
"""武器
Attributes:
weapon_type: 武器类型
attack: 基础攻击力
attribute:
affix: 武器技能
description: 描述
ascension: 突破材料
story: 武器故事
"""
weapon_type: WeaponType
attack: float
attribute: Optional[WeaponAttribute]
affix: Optional[WeaponAffix]
description: str
ascension: List[str]
story: Optional[str]
stats: List[WeaponState]
@staticmethod
def scrape_urls() -> List[URL]:
return [HONEY_HOST.join(f"fam_{i.lower()}/?lang=CHS") for i in WeaponType.__members__]
@classmethod
async def _parse_soup(cls, soup: BeautifulSoup) -> "Weapon":
"""解析武器页"""
soup = soup.select(".wp-block-post-content")[0]
tables = soup.find_all("table")
table_rows = tables[0].find_all("tr")
def get_table_text(row_num: int) -> str:
"""一个快捷函数,用于返回表格对应行的最后一个单元格中的文本"""
return table_rows[row_num].find_all("td")[-1].text.replace("\xa0", "")
def find_table(select: str):
"""一个快捷函数,用于寻找对应表格头的表格"""
return list(filter(lambda x: select in " ".join(x.attrs["class"]), tables))
id_ = re.findall(r"/img/(.*?)_gacha", str(table_rows[0]))[0]
weapon_type = WeaponType[get_table_text(1).split(",")[-1].strip()]
name = get_table_text(0)
rarity = len(table_rows[2].find_all("img"))
attack = float(get_table_text(4))
ascension = [re.findall(r"/(.*)/", tag.attrs["href"])[0] for tag in table_rows[-1].find_all("a")]
if rarity > 2: # 如果是 3 星及其以上的武器
attribute = WeaponAttribute(
type=AttributeType.convert(tables[2].find("thead").find("tr").find_all("td")[2].text.split(" ")[1]),
value=get_table_text(6),
)
affix = WeaponAffix(
name=get_table_text(7), description=[i.find_all("td")[1].text for i in tables[3].find_all("tr")[1:]]
)
description = get_table_text(-1) if len(tables) < 11 else get_table_text(9)
if story_table := find_table("quotes"):
story = story_table[0].text.strip()
else:
story = None
else: # 如果是 2 星及其以下的武器
attribute = affix = None
description = get_table_text(5)
story = tables[-1].text.strip()
stats = []
for row in tables[2].find_all("tr")[1:]:
cells = row.find_all("td")
if rarity > 2:
stats.append(WeaponState(level=cells[0].text, ATK=cells[1].text, bonus=cells[2].text))
else:
stats.append(WeaponState(level=cells[0].text, ATK=cells[1].text))
return Weapon(
id=id_,
name=name,
rarity=rarity,
attack=attack,
attribute=attribute,
affix=affix,
weapon_type=weapon_type,
story=story,
stats=stats,
description=description,
ascension=ascension,
)
@classmethod
async def get_name_list(cls, *, with_url: bool = False) -> List[Union[str, Tuple[str, URL]]]:
# 重写此函数的目的是名字去重,例如单手剑页面中有三个 “「一心传」名刀”
name_list = [i async for i in cls._name_list_generator(with_url=with_url)]
if with_url:
return [(i[0], list(i[1])[0][1]) for i in itertools.groupby(name_list, lambda x: x[0])]
else:
return [i[0] for i in itertools.groupby(name_list, lambda x: x)]
@property
def icon(self) -> WeaponIcon:
return WeaponIcon(
icon=str(HONEY_HOST.join(f"/img/{self.id}.webp")),
awakened=str(HONEY_HOST.join(f"/img/{self.id}_awaken_icon.webp")),
gacha=str(HONEY_HOST.join(f"/img/{self.id}_gacha_icon.webp")),
)