2022-08-28 14:37:31 +00:00
|
|
|
|
import asyncio
|
|
|
|
|
import re
|
|
|
|
|
from abc import abstractmethod
|
|
|
|
|
from asyncio import Queue
|
|
|
|
|
from multiprocessing import Value
|
2022-09-16 04:22:22 +00:00
|
|
|
|
from ssl import SSLZeroReturnError
|
2022-08-28 14:37:31 +00:00
|
|
|
|
from typing import AsyncIterator, ClassVar, List, Optional, Tuple, Union
|
|
|
|
|
|
|
|
|
|
import anyio
|
|
|
|
|
import ujson as json
|
|
|
|
|
from bs4 import BeautifulSoup
|
|
|
|
|
from httpx import AsyncClient, HTTPError, Response, URL
|
|
|
|
|
from pydantic import (
|
|
|
|
|
BaseConfig as PydanticBaseConfig,
|
|
|
|
|
BaseModel as PydanticBaseModel,
|
|
|
|
|
)
|
|
|
|
|
from typing_extensions import Self
|
|
|
|
|
|
|
|
|
|
__all__ = ['Model', 'WikiModel', 'SCRAPE_HOST']
|
|
|
|
|
|
|
|
|
|
SCRAPE_HOST = URL("https://genshin.honeyhunterworld.com/")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
class Model(PydanticBaseModel):
|
|
|
|
|
"""基类"""
|
|
|
|
|
|
|
|
|
|
def __new__(cls, *args, **kwargs):
|
|
|
|
|
# 让每次new的时候都解析
|
|
|
|
|
cls.update_forward_refs()
|
|
|
|
|
return super(Model, cls).__new__(cls)
|
|
|
|
|
|
|
|
|
|
class Config(PydanticBaseConfig):
|
|
|
|
|
# 使用 ujson 作为解析库
|
|
|
|
|
json_dumps = json.dumps
|
|
|
|
|
json_loads = json.loads
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
class WikiModel(Model):
|
2022-09-08 01:08:37 +00:00
|
|
|
|
# noinspection PyUnresolvedReferences
|
2022-08-28 14:37:31 +00:00
|
|
|
|
"""wiki所用到的基类
|
|
|
|
|
|
2022-09-08 01:08:37 +00:00
|
|
|
|
Attributes:
|
|
|
|
|
id (:obj:`int`): ID
|
|
|
|
|
name (:obj:`str`): 名称
|
|
|
|
|
rarity (:obj:`int`): 星级
|
2022-08-28 14:37:31 +00:00
|
|
|
|
|
2022-09-08 01:08:37 +00:00
|
|
|
|
_client (:class:`httpx.AsyncClient`): 发起 http 请求的 client
|
|
|
|
|
"""
|
2022-08-28 14:37:31 +00:00
|
|
|
|
_client: ClassVar[AsyncClient] = AsyncClient()
|
|
|
|
|
|
|
|
|
|
id: str
|
|
|
|
|
name: str
|
|
|
|
|
rarity: int
|
|
|
|
|
|
|
|
|
|
@staticmethod
|
|
|
|
|
@abstractmethod
|
|
|
|
|
def scrape_urls() -> List[URL]:
|
|
|
|
|
"""爬取的目标网页集合
|
|
|
|
|
|
|
|
|
|
例如有关武器的页面有:
|
|
|
|
|
[单手剑](https://genshin.honeyhunterworld.com/fam_sword/?lang=CHS)
|
|
|
|
|
[双手剑](https://genshin.honeyhunterworld.com/fam_claymore/?lang=CHS)
|
|
|
|
|
[长柄武器](https://genshin.honeyhunterworld.com/fam_polearm/?lang=CHS)
|
|
|
|
|
。。。
|
|
|
|
|
这个函数就是返回这些页面的网址所组成的 List
|
|
|
|
|
|
|
|
|
|
"""
|
|
|
|
|
|
|
|
|
|
@classmethod
|
|
|
|
|
async def _client_get(cls, url: Union[URL, str], retry_times: int = 5, sleep: float = 1) -> Response:
|
|
|
|
|
"""用自己的 client 发起 get 请求的快捷函数
|
|
|
|
|
|
|
|
|
|
Args:
|
|
|
|
|
url: 发起请求的 url
|
|
|
|
|
retry_times: 发生错误时的重复次数。不能小于 0 .
|
|
|
|
|
sleep: 发生错误后等待重试的时间,单位为秒。
|
|
|
|
|
Returns:
|
|
|
|
|
返回对应的请求
|
|
|
|
|
Raises:
|
|
|
|
|
请求所需要的异常
|
|
|
|
|
"""
|
|
|
|
|
for _ in range(retry_times):
|
|
|
|
|
try:
|
|
|
|
|
return await cls._client.get(url, follow_redirects=True)
|
2022-09-16 04:22:22 +00:00
|
|
|
|
except (HTTPError, SSLZeroReturnError):
|
2022-08-28 14:37:31 +00:00
|
|
|
|
await anyio.sleep(sleep)
|
|
|
|
|
return await cls._client.get(url, follow_redirects=True) # 防止 retry_times 等于 0 的时候无法发生请求
|
|
|
|
|
|
|
|
|
|
@classmethod
|
|
|
|
|
@abstractmethod
|
|
|
|
|
async def _parse_soup(cls, soup: BeautifulSoup) -> Self:
|
|
|
|
|
"""解析 soup 生成对应 WikiModel
|
|
|
|
|
|
|
|
|
|
Args:
|
|
|
|
|
soup: 需要解析的 soup
|
|
|
|
|
Returns:
|
|
|
|
|
返回对应的 WikiModel
|
|
|
|
|
"""
|
|
|
|
|
|
|
|
|
|
@classmethod
|
|
|
|
|
async def _scrape(cls, url: Union[URL, str]) -> Self:
|
|
|
|
|
"""从 url 中爬取数据,并返回对应的 Model
|
|
|
|
|
|
|
|
|
|
Args:
|
|
|
|
|
url: 目标 url. 可以为字符串 str , 也可以为 httpx.URL
|
|
|
|
|
Returns:
|
|
|
|
|
返回对应的 WikiModel
|
|
|
|
|
"""
|
|
|
|
|
response = await cls._client_get(url)
|
|
|
|
|
return await cls._parse_soup(BeautifulSoup(response.text, 'lxml'))
|
|
|
|
|
|
|
|
|
|
@classmethod
|
|
|
|
|
async def get_by_id(cls, id_: str) -> Self:
|
|
|
|
|
"""通过ID获取Model
|
|
|
|
|
|
|
|
|
|
Args:
|
|
|
|
|
id_: 目标 ID
|
|
|
|
|
Returns:
|
|
|
|
|
返回对应的 WikiModel
|
|
|
|
|
"""
|
|
|
|
|
return await cls._scrape(await cls.get_url_by_id(id_))
|
|
|
|
|
|
|
|
|
|
@classmethod
|
|
|
|
|
async def get_by_name(cls, name: str) -> Optional[Self]:
|
|
|
|
|
"""通过名称获取Model
|
|
|
|
|
|
|
|
|
|
Args:
|
|
|
|
|
name: 目标名
|
|
|
|
|
Returns:
|
|
|
|
|
返回对应的 WikiModel
|
|
|
|
|
"""
|
|
|
|
|
url = await cls.get_url_by_name(name)
|
|
|
|
|
if url is None:
|
|
|
|
|
return None
|
|
|
|
|
else:
|
|
|
|
|
return await cls._scrape(url)
|
|
|
|
|
|
|
|
|
|
@classmethod
|
|
|
|
|
async def get_full_data(cls) -> List[Self]:
|
|
|
|
|
"""获取全部数据的 Model
|
|
|
|
|
|
|
|
|
|
Returns:
|
|
|
|
|
返回能爬到的所有的 Model 所组成的 List
|
|
|
|
|
"""
|
|
|
|
|
return [i async for i in cls.full_data_generator()]
|
|
|
|
|
|
|
|
|
|
@classmethod
|
|
|
|
|
async def full_data_generator(cls) -> AsyncIterator[Self]:
|
|
|
|
|
"""Model 生成器
|
|
|
|
|
|
|
|
|
|
这是一个异步生成器,该函数在使用时会爬取所有数据,并将其转为对应的 Model,然后存至一个队列中
|
|
|
|
|
当有需要时,再一个一个地迭代取出
|
|
|
|
|
|
|
|
|
|
Returns:
|
|
|
|
|
返回能爬到的所有的 WikiModel 所组成的 List
|
|
|
|
|
"""
|
|
|
|
|
queue: Queue[Self] = Queue() # 存放 Model 的队列
|
|
|
|
|
signal = Value('i', 0) # 一个用于异步任务同步的信号
|
|
|
|
|
|
|
|
|
|
async def task(u):
|
|
|
|
|
# 包装的爬虫任务
|
|
|
|
|
await queue.put(await cls._scrape(u)) # 爬取一条数据,并将其放入队列中
|
|
|
|
|
signal.value -= 1 # 信号量减少 1 ,说明该爬虫任务已经完成
|
|
|
|
|
|
|
|
|
|
for _, url in await cls.get_name_list(with_url=True): # 遍历爬取所有需要爬取的页面
|
|
|
|
|
signal.value += 1 # 信号量增加 1 ,说明有一个爬虫任务被添加
|
|
|
|
|
asyncio.create_task(task(url)) # 创建一个爬虫任务
|
|
|
|
|
|
|
|
|
|
while signal.value > 0 or not queue.empty(): # 当还有未完成的爬虫任务或存放数据的队列不为空时
|
|
|
|
|
yield await queue.get() # 取出并返回一个存放的 Model
|
|
|
|
|
|
|
|
|
|
def __str__(self) -> str:
|
|
|
|
|
return f"<{self.__class__.__name__} {super(WikiModel, self).__str__()}>"
|
|
|
|
|
|
|
|
|
|
def __repr__(self) -> str:
|
|
|
|
|
return self.__str__()
|
|
|
|
|
|
|
|
|
|
@staticmethod
|
|
|
|
|
async def get_url_by_id(id_: str) -> URL:
|
|
|
|
|
"""根据 id 获取对应的 url
|
|
|
|
|
|
|
|
|
|
例如神里绫华的ID为 ayaka_002,对应的数据页url为 https://genshin.honeyhunterworld.com/ayaka_002/?lang=CHS
|
|
|
|
|
|
|
|
|
|
Args:
|
|
|
|
|
id_ : 实列ID
|
|
|
|
|
Returns:
|
|
|
|
|
返回对应的 url
|
|
|
|
|
"""
|
|
|
|
|
return SCRAPE_HOST.join(f"{id_}/?lang=CHS")
|
|
|
|
|
|
|
|
|
|
@classmethod
|
|
|
|
|
async def _name_list_generator(cls, *, with_url: bool = False) -> AsyncIterator[Union[str, Tuple[str, URL]]]:
|
|
|
|
|
"""一个 Model 的名称 和 其对应 url 的异步生成器
|
|
|
|
|
|
|
|
|
|
Args:
|
|
|
|
|
with_url: 是否返回相应的 url
|
|
|
|
|
Returns:
|
|
|
|
|
返回对应的名称列表 或者 名称与url 的列表
|
|
|
|
|
"""
|
|
|
|
|
urls = cls.scrape_urls()
|
|
|
|
|
queue: Queue[Union[str, Tuple[str, URL]]] = Queue() # 存放 Model 的队列
|
|
|
|
|
signal = Value('i', len(urls)) # 一个用于异步任务同步的信号,初始值为存放所需要爬取的页面数
|
|
|
|
|
|
2022-09-08 01:08:37 +00:00
|
|
|
|
async def task(page: URL):
|
2022-08-28 14:37:31 +00:00
|
|
|
|
"""包装的爬虫任务"""
|
|
|
|
|
response = await cls._client_get(page)
|
|
|
|
|
# 从页面中获取对应的 chaos data (未处理的json格式字符串)
|
|
|
|
|
chaos_data = re.findall(r'sortable_data\.push\((.*)\);\s*sortable_cur_page', response.text)[0]
|
|
|
|
|
json_data = json.loads(chaos_data) # 转为 json
|
|
|
|
|
for data in json_data: # 遍历 json
|
|
|
|
|
data_name = re.findall(r'>(.*)<', data[1])[0] # 获取 Model 的名称
|
|
|
|
|
if with_url: # 如果需要返回对应的 url
|
|
|
|
|
data_url = SCRAPE_HOST.join(re.findall(r'\"(.*?)\"', data[0])[0])
|
|
|
|
|
await queue.put((data_name, data_url))
|
|
|
|
|
else:
|
|
|
|
|
await queue.put(data_name)
|
|
|
|
|
signal.value = signal.value - 1 # 信号量减少 1 ,说明该爬虫任务已经完成
|
|
|
|
|
|
|
|
|
|
for url in urls: # 遍历需要爬出的页面
|
2022-09-08 01:08:37 +00:00
|
|
|
|
asyncio.create_task(task(url)) # 添加爬虫任务
|
2022-08-28 14:37:31 +00:00
|
|
|
|
while signal.value > 0 or not queue.empty(): # 当还有未完成的爬虫任务或存放数据的队列不为空时
|
|
|
|
|
yield await queue.get() # 取出并返回一个存放的 Model
|
|
|
|
|
|
|
|
|
|
@classmethod
|
|
|
|
|
async def get_name_list(cls, *, with_url: bool = False) -> List[Union[str, Tuple[str, URL]]]:
|
|
|
|
|
"""获取全部 Model 的 名称
|
|
|
|
|
|
|
|
|
|
Returns:
|
|
|
|
|
返回能爬到的所有的 Model 的名称所组成的 List
|
|
|
|
|
"""
|
|
|
|
|
return [i async for i in cls._name_list_generator(with_url=with_url)]
|
|
|
|
|
|
|
|
|
|
@classmethod
|
|
|
|
|
async def get_url_by_name(cls, name: str) -> Optional[URL]:
|
|
|
|
|
"""通过 Model 的名称获取对应的 url
|
|
|
|
|
|
|
|
|
|
Args:
|
|
|
|
|
name: 实列名
|
|
|
|
|
Returns:
|
|
|
|
|
若有对应的实列,则返回对应的 url; 若没有, 则返回 None
|
|
|
|
|
"""
|
|
|
|
|
async for n, url in cls._name_list_generator(with_url=True):
|
|
|
|
|
if name == n:
|
|
|
|
|
return url
|
|
|
|
|
|
|
|
|
|
@property
|
|
|
|
|
@abstractmethod
|
|
|
|
|
def icon(self):
|
|
|
|
|
"""返回此 Model 的图标链接"""
|