mirror of
https://gitee.com/g1879/DrissionPage.git
synced 2024-12-10 04:00:23 +08:00
弃用requests_html,未完成
This commit is contained in:
parent
bbea8871a0
commit
41d700a3d9
@ -10,18 +10,20 @@ from re import split as re_SPLIT
|
||||
from shutil import rmtree
|
||||
from typing import Union
|
||||
|
||||
from requests_html import Element
|
||||
from lxml.etree import _Element
|
||||
# from lxml.html import HtmlElement
|
||||
# from requests_html import Element
|
||||
from selenium.webdriver.remote.webelement import WebElement
|
||||
|
||||
|
||||
class DrissionElement(object):
|
||||
"""SessionElement和DriverElement的基类"""
|
||||
|
||||
def __init__(self, ele: Union[Element, WebElement]):
|
||||
def __init__(self, ele: Union[WebElement, _Element]):
|
||||
self._inner_ele = ele
|
||||
|
||||
@property
|
||||
def inner_ele(self) -> Union[WebElement, Element]:
|
||||
def inner_ele(self) -> Union[WebElement, _Element]:
|
||||
return self._inner_ele
|
||||
|
||||
@property
|
||||
|
@ -8,7 +8,7 @@ from typing import Union
|
||||
from urllib.parse import urlparse
|
||||
|
||||
from requests import Session
|
||||
from requests_html import HTMLSession
|
||||
# from requests_html import HTMLSession
|
||||
from selenium import webdriver
|
||||
from selenium.common.exceptions import WebDriverException
|
||||
from selenium.webdriver.chrome.options import Options
|
||||
@ -23,12 +23,12 @@ class Drission(object):
|
||||
|
||||
def __init__(self,
|
||||
driver_or_options: Union[WebDriver, dict, Options] = None,
|
||||
session_or_options: Union[Session, HTMLSession, dict] = None,
|
||||
session_or_options: Union[Session, dict] = None,
|
||||
ini_path: str = None,
|
||||
proxy: dict = None):
|
||||
"""初始化,可接收现成的WebDriver和Session对象,或接收它们的配置信息 \n
|
||||
:param driver_or_options: driver对象或chrome设置,Options类或设置字典
|
||||
:param session_or_options: session、HTMLSession对象或session设置
|
||||
:param session_or_options: Session对象设置
|
||||
:param ini_path: ini文件路径
|
||||
:param proxy: 代理设置
|
||||
"""
|
||||
@ -36,15 +36,15 @@ class Drission(object):
|
||||
self._driver = None
|
||||
self._driver_path = 'chromedriver'
|
||||
self._proxy = proxy
|
||||
if isinstance(session_or_options, HTMLSession):
|
||||
if isinstance(session_or_options, Session):
|
||||
self._session = session_or_options
|
||||
elif isinstance(session_or_options, Session):
|
||||
self._session = HTMLSession()
|
||||
for key in session_or_options.__dict__: # session对象强制升级为子类HTMLSession对象
|
||||
if key != 'hooks':
|
||||
self._session.__dict__[key] = session_or_options.__dict__[key]
|
||||
else:
|
||||
self._session.hooks['response'].extend(session_or_options.hooks['response'])
|
||||
# elif isinstance(session_or_options, Session):
|
||||
# self._session = HTMLSession()
|
||||
# for key in session_or_options.__dict__: # session对象强制升级为子类HTMLSession对象
|
||||
# if key != 'hooks':
|
||||
# self._session.__dict__[key] = session_or_options.__dict__[key]
|
||||
# else:
|
||||
# self._session.hooks['response'].extend(session_or_options.hooks['response'])
|
||||
else:
|
||||
if session_or_options is None:
|
||||
self._session_options = OptionsManager(ini_path).get_option('session_options')
|
||||
@ -64,10 +64,10 @@ class Drission(object):
|
||||
self._driver_path = self._driver_options['driver_path']
|
||||
|
||||
@property
|
||||
def session(self) -> HTMLSession:
|
||||
def session(self) -> Session:
|
||||
"""返回HTMLSession对象,如为None则按配置信息创建"""
|
||||
if self._session is None:
|
||||
self._session = HTMLSession()
|
||||
self._session = Session()
|
||||
attrs = ['headers', 'cookies', 'auth', 'proxies', 'hooks', 'params', 'verify',
|
||||
'cert', 'adapters', 'stream', 'trust_env', 'max_redirects']
|
||||
for i in attrs:
|
||||
|
@ -6,8 +6,8 @@
|
||||
"""
|
||||
from typing import Union, List, Tuple
|
||||
|
||||
from requests import Response
|
||||
from requests_html import HTMLSession, Element
|
||||
from requests import Response, Session
|
||||
# from requests_html import HTMLSession, Element
|
||||
from selenium.webdriver.chrome.webdriver import WebDriver
|
||||
from selenium.webdriver.remote.webelement import WebElement
|
||||
|
||||
@ -127,7 +127,7 @@ class MixPage(Null, SessionPage, DriverPage):
|
||||
return self._drission.driver
|
||||
|
||||
@property
|
||||
def session(self) -> HTMLSession:
|
||||
def session(self) -> Session:
|
||||
"""返回session对象,如没有则创建 \n
|
||||
:return: HTMLSession对象
|
||||
"""
|
||||
@ -279,7 +279,7 @@ class MixPage(Null, SessionPage, DriverPage):
|
||||
return super().get(url, go_anyway, show_errmsg, retry, interval, **kwargs)
|
||||
|
||||
def ele(self,
|
||||
loc_or_ele: Union[Tuple[str, str], str, DriverElement, SessionElement, Element, WebElement],
|
||||
loc_or_ele: Union[Tuple[str, str], str, DriverElement, SessionElement, WebElement],
|
||||
mode: str = None,
|
||||
timeout: float = None,
|
||||
show_errmsg: bool = False) -> Union[DriverElement, SessionElement, str]:
|
||||
|
@ -8,28 +8,33 @@ import re
|
||||
from html import unescape
|
||||
from typing import Union, List, Tuple
|
||||
|
||||
from requests_html import Element, BaseParser
|
||||
from lxml import etree
|
||||
from lxml.etree import _Element
|
||||
|
||||
from .common import DrissionElement, get_loc_from_str, translate_loc_to_xpath
|
||||
|
||||
|
||||
# from lxml.html import HtmlElement
|
||||
# from requests_html import Element, BaseParser
|
||||
|
||||
|
||||
class SessionElement(DrissionElement):
|
||||
"""session模式的元素对象,包装了一个Element对象,并封装了常用功能"""
|
||||
|
||||
def __init__(self, ele: Element):
|
||||
def __init__(self, ele: _Element):
|
||||
super().__init__(ele)
|
||||
|
||||
def __repr__(self):
|
||||
attrs = [f"{attr}='{self.attrs[attr]}'" for attr in self.attrs]
|
||||
return f'<SessionElement {self.tag} {" ".join(attrs)}>'
|
||||
# def __repr__(self):
|
||||
# attrs = [f"{attr}='{self.attrs[attr]}'" for attr in self.attrs]
|
||||
# return f'<SessionElement {self.tag} {" ".join(attrs)}>'
|
||||
|
||||
@property
|
||||
def attrs(self) -> dict:
|
||||
"""返回元素所有属性及值"""
|
||||
attrs = dict()
|
||||
for attr in self.inner_ele.attrs:
|
||||
attrs[attr] = self.attr(attr)
|
||||
return attrs
|
||||
# @property
|
||||
# def attrs(self) -> dict:
|
||||
# """返回元素所有属性及值"""
|
||||
# attrs = dict()
|
||||
# for attr in self.inner_ele.attrs:
|
||||
# attrs[attr] = self.attr(attr)
|
||||
# return attrs
|
||||
|
||||
@property
|
||||
def text(self) -> str:
|
||||
@ -45,42 +50,45 @@ class SessionElement(DrissionElement):
|
||||
@property
|
||||
def html(self) -> str:
|
||||
"""返回元素innerHTML文本"""
|
||||
html = unescape(self._inner_ele.html).replace('\xa0', ' ')
|
||||
# ee=self.ele('xpath:./*')
|
||||
html = unescape(etree.tostring(self._inner_ele).decode()).replace('\xa0', ' ')
|
||||
# html = unescape(self._inner_ele.html).replace('\xa0', ' ')
|
||||
r = re.match(r'<.*?>(.*)</.*?>', html, flags=re.DOTALL)
|
||||
return None if not r else r.group(1)
|
||||
# return html
|
||||
|
||||
@property
|
||||
def tag(self) -> str:
|
||||
"""返回元素类型"""
|
||||
return self._inner_ele.tag
|
||||
|
||||
@property
|
||||
def css_path(self) -> str:
|
||||
"""返回css path路径"""
|
||||
return self._get_ele_path('css')
|
||||
# @property
|
||||
# def css_path(self) -> str:
|
||||
# """返回css path路径"""
|
||||
# return self._get_ele_path('css')
|
||||
|
||||
@property
|
||||
def xpath(self) -> str:
|
||||
"""返回xpath路径"""
|
||||
return self._get_ele_path('xpath')
|
||||
# @property
|
||||
# def xpath(self) -> str:
|
||||
# """返回xpath路径"""
|
||||
# return self._get_ele_path('xpath')
|
||||
|
||||
def _get_ele_path(self, mode):
|
||||
"""获取css路径或xpath路径"""
|
||||
path_str = ''
|
||||
ele = self
|
||||
while ele:
|
||||
ele_id = ele.attr('id')
|
||||
if ele_id:
|
||||
return f'#{ele_id}{path_str}' if mode == 'css' else f'//{ele.tag}[@id="{ele_id}"]{path_str}'
|
||||
else:
|
||||
if mode == 'css':
|
||||
brothers = len(ele.eles(f'xpath:./preceding-sibling::*'))
|
||||
path_str = f'>:nth-child({brothers + 1}){path_str}'
|
||||
else:
|
||||
brothers = len(ele.eles(f'xpath:./preceding-sibling::{ele.tag}'))
|
||||
path_str = f'/{ele.tag}[{brothers + 1}]{path_str}' if brothers > 0 else f'/{ele.tag}{path_str}'
|
||||
ele = ele.parent
|
||||
return path_str[1:] if mode == 'css' else path_str
|
||||
# def _get_ele_path(self, mode):
|
||||
# """获取css路径或xpath路径"""
|
||||
# path_str = ''
|
||||
# ele = self
|
||||
# while ele:
|
||||
# ele_id = ele.attr('id')
|
||||
# if ele_id:
|
||||
# return f'#{ele_id}{path_str}' if mode == 'css' else f'//{ele.tag}[@id="{ele_id}"]{path_str}'
|
||||
# else:
|
||||
# if mode == 'css':
|
||||
# brothers = len(ele.eles(f'xpath:./preceding-sibling::*'))
|
||||
# path_str = f'>:nth-child({brothers + 1}){path_str}'
|
||||
# else:
|
||||
# brothers = len(ele.eles(f'xpath:./preceding-sibling::{ele.tag}'))
|
||||
# path_str = f'/{ele.tag}[{brothers + 1}]{path_str}' if brothers > 0 else f'/{ele.tag}{path_str}'
|
||||
# ele = ele.parent
|
||||
# return path_str[1:] if mode == 'css' else path_str
|
||||
|
||||
@property
|
||||
def parent(self):
|
||||
@ -199,47 +207,47 @@ class SessionElement(DrissionElement):
|
||||
"""
|
||||
return self.ele(loc_or_str, mode='all', show_errmsg=show_errmsg)
|
||||
|
||||
def attr(self, attr: str) -> Union[str, None]:
|
||||
"""返回属性值 \n
|
||||
:param attr: 属性名
|
||||
:return: 属性值文本,没有该属性返回None
|
||||
"""
|
||||
try:
|
||||
if attr == 'href':
|
||||
# 如直接获取attr只能获取相对地址
|
||||
link = self._inner_ele.attrs['href']
|
||||
if link.lower().startswith(('javascript:', 'mailto:')):
|
||||
return link
|
||||
elif link.startswith('#'):
|
||||
if '#' in self.inner_ele.url:
|
||||
return re.sub(r'#.*', link, self.inner_ele.url)
|
||||
else:
|
||||
return f'{self.inner_ele.url}{link}'
|
||||
elif link.startswith('?'): # 避免当相对URL以?开头时requests-html丢失参数的bug
|
||||
if '?' in self.inner_ele.url:
|
||||
return re.sub(r'\?.*', link, self.inner_ele.url)
|
||||
else:
|
||||
return f'{self.inner_ele.url}{link}'
|
||||
else:
|
||||
for link in self._inner_ele.absolute_links:
|
||||
return link
|
||||
elif attr == 'src':
|
||||
return self._inner_ele._make_absolute(self._inner_ele.attrs['src'])
|
||||
elif attr == 'class':
|
||||
return ' '.join(self._inner_ele.attrs['class'])
|
||||
elif attr == 'text':
|
||||
return self.text
|
||||
elif attr == 'outerHTML':
|
||||
return self.inner_ele.html
|
||||
elif attr == 'innerHTML':
|
||||
return self.html
|
||||
else:
|
||||
return self._inner_ele.attrs[attr]
|
||||
except:
|
||||
return None
|
||||
# def attr(self, attr: str) -> Union[str, None]:
|
||||
# """返回属性值 \n
|
||||
# :param attr: 属性名
|
||||
# :return: 属性值文本,没有该属性返回None
|
||||
# """
|
||||
# try:
|
||||
# if attr == 'href':
|
||||
# # 如直接获取attr只能获取相对地址
|
||||
# link = self._inner_ele.attrs['href']
|
||||
# if link.lower().startswith(('javascript:', 'mailto:')):
|
||||
# return link
|
||||
# elif link.startswith('#'):
|
||||
# if '#' in self.inner_ele.url:
|
||||
# return re.sub(r'#.*', link, self.inner_ele.url)
|
||||
# else:
|
||||
# return f'{self.inner_ele.url}{link}'
|
||||
# elif link.startswith('?'): # 避免当相对URL以?开头时requests-html丢失参数的bug
|
||||
# if '?' in self.inner_ele.url:
|
||||
# return re.sub(r'\?.*', link, self.inner_ele.url)
|
||||
# else:
|
||||
# return f'{self.inner_ele.url}{link}'
|
||||
# else:
|
||||
# for link in self._inner_ele.absolute_links:
|
||||
# return link
|
||||
# elif attr == 'src':
|
||||
# return self._inner_ele._make_absolute(self._inner_ele.attrs['src'])
|
||||
# elif attr == 'class':
|
||||
# return ' '.join(self._inner_ele.attrs['class'])
|
||||
# elif attr == 'text':
|
||||
# return self.text
|
||||
# elif attr == 'outerHTML':
|
||||
# return self.inner_ele.html
|
||||
# elif attr == 'innerHTML':
|
||||
# return self.html
|
||||
# else:
|
||||
# return self._inner_ele.attrs[attr]
|
||||
# except:
|
||||
# return None
|
||||
|
||||
|
||||
def execute_session_find(page_or_ele: BaseParser,
|
||||
def execute_session_find(page_or_ele: _Element,
|
||||
loc: Tuple[str, str],
|
||||
mode: str = 'single',
|
||||
show_errmsg: bool = False) -> Union[SessionElement, List[SessionElement or str]]:
|
||||
@ -254,32 +262,47 @@ def execute_session_find(page_or_ele: BaseParser,
|
||||
mode = mode or 'single'
|
||||
if mode not in ['single', 'all']:
|
||||
raise ValueError("Argument mode can only be 'single' or 'all'.")
|
||||
|
||||
loc_by, loc_str = loc
|
||||
# print(loc)
|
||||
# ele = page_or_ele.xpath(loc_str)
|
||||
# print(ele)
|
||||
try:
|
||||
ele = None
|
||||
# ele = None
|
||||
if loc_by == 'xpath':
|
||||
if 'PyQuery' in str(type(page_or_ele.element)):
|
||||
# 从页面查找。
|
||||
ele = page_or_ele.xpath(loc_str)
|
||||
elif 'HtmlElement' in str(type(page_or_ele.element)):
|
||||
# 从元素查找。这样区分是为了能找到上级元素
|
||||
try:
|
||||
elements = page_or_ele.element.xpath(loc_str)
|
||||
ele = [Element(element=e, url=page_or_ele.url) for e in elements]
|
||||
except AttributeError:
|
||||
ele = page_or_ele.xpath(loc_str)
|
||||
ele = page_or_ele.xpath(loc_str)
|
||||
# if 'PyQuery' in str(type(page_or_ele.element)):
|
||||
# # 从页面查找。
|
||||
# ele = page_or_ele.xpath(loc_str)
|
||||
# elif 'HtmlElement' in str(type(page_or_ele.element)):
|
||||
# # 从元素查找。这样区分是为了能找到上级元素
|
||||
# try:
|
||||
# elements = page_or_ele.element.xpath(loc_str)
|
||||
# ele = [Element(element=e, url=page_or_ele.url) for e in elements]
|
||||
# except AttributeError:
|
||||
# ele = page_or_ele.xpath(loc_str)
|
||||
else: # 用css selector获取
|
||||
ele = page_or_ele.find(loc_str)
|
||||
ele = page_or_ele.cssselect(loc_str)
|
||||
|
||||
if mode == 'single':
|
||||
ele = ele[0] if ele else None
|
||||
return SessionElement(ele) if isinstance(ele, Element) else unescape(ele).replace('\xa0', ' ')
|
||||
return SessionElement(ele) if isinstance(ele, _Element) else unescape(ele).replace('\xa0', ' ')
|
||||
elif mode == 'all':
|
||||
ele = filter(lambda x: x != '\n', ele) # 去除元素间换行符
|
||||
ele = map(lambda x: unescape(x).replace('\xa0', ' ') if isinstance(x, str) else x, ele) # 替换空格
|
||||
return [SessionElement(e) if isinstance(e, Element) else e for e in ele]
|
||||
return [SessionElement(e) if isinstance(e, _Element) else e for e in ele]
|
||||
except:
|
||||
if show_errmsg:
|
||||
print('Element(s) not found.', loc)
|
||||
raise
|
||||
return [] if mode == 'all' else None
|
||||
|
||||
|
||||
def get_HtmlElement(html: str) -> _Element:
|
||||
# html = f'<drission_root>{html}</drission_root>'
|
||||
ele_or_page = etree.HTML(html)
|
||||
# html = etree.tostring(ele_or_page).decode()
|
||||
# if str(html).startswith('<html><body>') and str(html).endswith('</body></html>'):
|
||||
# html = etree.tostring(ele_or_page)[12:-14].decode()
|
||||
# ele_or_page = etree.fromstring(html)
|
||||
return ele_or_page
|
||||
|
@ -15,17 +15,18 @@ from typing import Union, List, Tuple
|
||||
from urllib import parse
|
||||
from urllib.parse import urlparse, quote
|
||||
|
||||
from requests_html import HTMLSession, HTMLResponse, Element
|
||||
# from requests_html import HTMLSession, HTMLResponse, Element
|
||||
from requests import Session, Response
|
||||
|
||||
from .common import get_loc_from_str, translate_loc_to_xpath, get_available_file_name
|
||||
from .config import OptionsManager
|
||||
from .session_element import SessionElement, execute_session_find
|
||||
from .session_element import SessionElement, execute_session_find, get_HtmlElement
|
||||
|
||||
|
||||
class SessionPage(object):
|
||||
"""SessionPage封装了页面操作的常用功能,使用requests_html来获取、解析网页。"""
|
||||
|
||||
def __init__(self, session: HTMLSession, timeout: float = 10):
|
||||
def __init__(self, session: Session, timeout: float = 10):
|
||||
"""初始化函数"""
|
||||
self._session = session
|
||||
self.timeout = timeout
|
||||
@ -34,12 +35,12 @@ class SessionPage(object):
|
||||
self._response = None
|
||||
|
||||
@property
|
||||
def session(self) -> HTMLSession:
|
||||
def session(self) -> Session:
|
||||
"""返回session对象"""
|
||||
return self._session
|
||||
|
||||
@property
|
||||
def response(self) -> HTMLResponse:
|
||||
def response(self) -> Response:
|
||||
"""返回访问url得到的response对象"""
|
||||
return self._response
|
||||
|
||||
@ -66,10 +67,10 @@ class SessionPage(object):
|
||||
@property
|
||||
def html(self) -> str:
|
||||
"""返回页面html文本"""
|
||||
return self.response.html.html
|
||||
return self.response.text
|
||||
|
||||
def ele(self,
|
||||
loc_or_ele: Union[Tuple[str, str], str, SessionElement, Element],
|
||||
loc_or_ele: Union[Tuple[str, str], str, SessionElement], # , Element
|
||||
mode: str = None,
|
||||
show_errmsg: bool = False) -> Union[SessionElement, List[SessionElement or str], str, None]:
|
||||
"""返回页面中符合条件的元素,默认返回第一个 \n
|
||||
@ -112,13 +113,13 @@ class SessionPage(object):
|
||||
elif isinstance(loc_or_ele, SessionElement):
|
||||
return loc_or_ele
|
||||
|
||||
elif isinstance(loc_or_ele, Element):
|
||||
return SessionElement(loc_or_ele)
|
||||
# elif isinstance(loc_or_ele, Element):
|
||||
# return SessionElement(loc_or_ele)
|
||||
|
||||
else:
|
||||
raise ValueError('Argument loc_or_str can only be tuple, str, SessionElement, Element.')
|
||||
|
||||
return execute_session_find(self.response.html, loc_or_ele, mode, show_errmsg)
|
||||
return execute_session_find(get_HtmlElement(self.response.text), loc_or_ele, mode, show_errmsg)
|
||||
|
||||
def eles(self,
|
||||
loc_or_str: Union[Tuple[str, str], str],
|
||||
@ -156,7 +157,7 @@ class SessionPage(object):
|
||||
times: int = 0,
|
||||
interval: float = 1,
|
||||
show_errmsg: bool = False,
|
||||
**kwargs) -> HTMLResponse:
|
||||
**kwargs) -> Response:
|
||||
"""尝试连接,重试若干次 \n
|
||||
:param to_url: 要访问的url
|
||||
:param times: 重试次数
|
||||
@ -434,7 +435,7 @@ class SessionPage(object):
|
||||
|
||||
if not_stream: # 加载网页时修复编码
|
||||
r._content = r.content.replace(b'\x08', b'\\b') # 避免存在退格符导致乱码或解析出错
|
||||
r.html.encoding = r.encoding # 修复requests_html丢失编码方式的bug
|
||||
# r.html.encoding = r.encoding # 修复requests_html丢失编码方式的bug
|
||||
if charset:
|
||||
r.encoding = charset
|
||||
return r, 'Success'
|
||||
|
@ -1,4 +1,5 @@
|
||||
selenium~=3.141.0
|
||||
requests-html~=0.10.0
|
||||
requests~=2.23.0
|
||||
tldextract~=2.2.2
|
||||
selenium
|
||||
requests
|
||||
tldextract
|
||||
DrissionPage
|
||||
lxml
|
Loading…
x
Reference in New Issue
Block a user