Python3 的urllib实例

在Python3中合并了 urllib 和 urllib2, 统一命名为 urllib 了,我觉得这样更加合理了。urllib真的是太方便了,让我们可以像读取本地文件一样读取WEB上的数据。这周学习了一番,并封装了一个类,供以后方便使用吧!并附带有许多的应用实例。

一、自己封装的类

#!/usr/bin/env python3
# -*- coding: utf-8 -*-
 
import time
import sys
import gzip
import socket
import urllib.request, urllib.parse, urllib.error
import http.cookiejar
 
class HttpTester:
    def __init__(self, timeout=10, addHeaders=True):
        socket.setdefaulttimeout(timeout)   # 设置超时时间
 
        self.__opener = urllib.request.build_opener()
        urllib.request.install_opener(self.__opener)
 
        if addHeaders: self.__addHeaders()
 
    def __error(self, e):
        '''错误处理'''
        print(e)
 
    def __addHeaders(self):
        '''添加默认的 headers.'''
        self.__opener.addheaders = [('User-Agent', 'Mozilla/5.0 (X11; Ubuntu; Linux x86_64; rv:22.0) Gecko/20100101 Firefox/22.0'),
                                    ('Connection', 'keep-alive'),
                                    ('Cache-Control', 'no-cache'),
                                    ('Accept-Language:', 'zh-cn,zh;q=0.8,en-us;q=0.5,en;q=0.3'),
                                    ('Accept-Encoding', 'gzip, deflate'),
                                    ('Accept', 'text/html,application/xhtml+xml,application/xml;q=0.9,*/*;q=0.8')]
 
    def __decode(self, webPage, charset):
        '''gzip解压,并根据指定的编码解码网页'''
        if webPage.startswith(b'\x1f\x8b'):
            return gzip.decompress(webPage).decode(charset)
        else:
            return webPage.decode(charset)
 
    def addCookiejar(self):
        '''为 self.__opener 添加 cookiejar handler。'''
        cj = http.cookiejar.CookieJar()
        self.__opener.add_handler(urllib.request.HTTPCookieProcessor(cj))
 
    def addProxy(self, host, type='http'):
        '''设置代理'''
        proxy = urllib.request.ProxyHandler({type: host})
        self.__opener.add_handler(proxy)
 
    def addAuth(self, url, user, pwd):
        '''添加认证'''
        pwdMsg = urllib.request.HTTPPasswordMgrWithDefaultRealm()
        pwdMsg.add_password(None, url, user, pwd)
        auth = urllib.request.HTTPBasicAuthHandler(pwdMsg)
        self.__opener.add_handler(auth)
 
    def get(self, url, params={}, headers={}, charset='UTF-8'):
        '''HTTP GET 方法'''
        if params: url += '?' + urllib.parse.urlencode(params)
        request = urllib.request.Request(url)
        for k,v in headers.items(): request.add_header(k, v)    # 为特定的 request 添加指定的 headers
 
        try:
            response = urllib.request.urlopen(request)
        except urllib.error.HTTPError as e:
            self.__error(e)
        else:
            return self.__decode(response.read(), charset)
 
    def post(self, url, params={}, headers={}, charset='UTF-8'):
        '''HTTP POST 方法'''
        params = urllib.parse.urlencode(params)
        request = urllib.request.Request(url, data=params.encode(charset))  # 带 data 参数的 request 被认为是 POST 方法。
        for k,v in headers.items(): request.add_header(k, v)
 
        try:
            response = urllib.request.urlopen(request)
        except urllib.error.HTTPError as e:
            self.__error(e)
        else:
            return self.__decode(response.read(), charset)
 
    def download(self, url, savefile):
        '''下载文件或网页'''
        header_gzip = None
 
        for header in self.__opener.addheaders:     # 移除支持 gzip 压缩的 header
            if 'Accept-Encoding' in header:
                header_gzip = header
                self.__opener.addheaders.remove(header)
 
        __perLen = 0
        def reporthook(a, b, c):    # a:已经下载的数据大小; b:数据大小; c:远程文件大小;
            if c > 1000000:
                nonlocal __perLen
                per = (100.0 * a * b) / c
                if per>100: per=100
                per = '{:.2f}%'.format(per)
                print('\b'*__perLen, per, end='')     # 打印下载进度百分比
                sys.stdout.flush()
                __perLen = len(per)+1
 
        print('--> {}\t'.format(url), end='')
        try:
            urllib.request.urlretrieve(url, savefile, reporthook)   # reporthook 为回调钩子函数,用于显示下载进度
        except urllib.error.HTTPError as e:
            self.__error(e)
        finally:
            self.__opener.addheaders.append(header_gzip)
            print()

二、应用实例

  1. 在OSC上动弹一下
ht = HttpTester()
ht.addCookiejar()
 
# 为了隐私,把有些关键字隐藏了哦!
ht.get('https://www.oschina.net/home/login?goto_page=http%3A%2F%2Fwww.oschina.net%2F')
ht.post(url = 'https://www.oschina.net/action/user/hash_login',
        params = {'email': '****@foxmail.com',
                  'pwd': 'e4a1425583d37fcd33b9*************',   #密码哈希,Firefox开发工具抓取的
                  'save_login': '1'}
)
ht.get('http://www.oschina.net/')
ht.post(url = 'http://www.oschina.net/action/tweet/pub',
        params = {'user_code': '8VZTqhkJOqhnuugHvzBtME4***********',
                  'user': '102*****',
                  'msg': '大家在动弹什么? via:(python3, urllib) ->{t}'.format(t = time.ctime())}
)

2.金山快盘签到送空间

ht = HttpTester()
ht.addCookiejar()
 
# 为了隐私,把有些关键字隐藏了哦!
ht.get('https://www.kuaipan.cn/account_login.htm')
ht.post(url='https://www.kuaipan.cn/index.php?ac=account&op=login',
        params={'username': '****@qq.com',
                'userpwd': 'lyb********',
                'isajax': 'yes'}
)
ht.get('http://www.kuaipan.cn/index.php?ac=zone&op=taskdetail')
ht.get('http://www.kuaipan.cn/index.php?ac=common&op=usersign')

转自:
http://www.itwhy.org/%E8%BD%AF%E4%BB%B6%E5%B7%A5%E7%A8%8B/python/python3-urllib-%E5%AE%9E%E4%BE%8B.html

    原文作者:星明月稀
    原文地址: https://blog.csdn.net/hxsstar/article/details/20382281
    本文转自网络文章,转载此文章仅为分享知识,如有侵权,请联系博主进行删除。
点赞