from asyncio.windows_events import NULL
from doctest import Example
import importlib
from pathlib import Path
from lxml import etree
import re
from urllib import request, response
import requests
from urllib import request
import re #进行数据清洗要导入此模块
from lxml import etree
import xlsxwriter
from asyncio import sleep
import xlwt
import xlsxwriter
from datetime import date,datetime
from openpyxl import load_workbook
import time
import os
import sys
import qrcode
# 获取所有的推文链接
def GetSiteList(start, end,path):
siteLists = []
print("\n程序正在运行...")
for i in range(start, end):
parser = etree.HTMLParser(encoding='utf-8')
try:
tree = etree.parse(path+str(i)+".html", parser=parser)
except Exception as result:
print("错误:你保存的html文件名称错误,正确文件名称为:1.html、2.html、3.html,请重新运行程序。")
pass
html = etree.tostring(tree,encoding="utf-8").decode()
result = tree.xpath("//*[@class='weui-desktop-mass-appmsg__title']/@href")
siteLists.extend(result)
return siteLists
# 获取所有推文中的责任姓名
def GetName(siteList,path):
headers = {"User-Agent":"Mozilla/5.0 (Linux; Android 6.0; Nexus 5 Build/MRA58N) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/105.0.0.0 Mobile Safari/537.36 Edg/105.0.1343.33"}
# 创建表格
workbook = xlsxwriter.Workbook(path+'官微每月作品表.xlsx')
worksheet = workbook.add_worksheet()
worksheet.merge_range('A1:I1',"记者团官方微信公众号 每月作品表汇总")
bold = workbook.add_format({'bold': True,"align":"center"})
worksheet.write('A1', '记者团官方微信公众号 每月作品表汇总', workbook.add_format({'bold': True,"size":15,"align":"center"}))
worksheet.write('A2', '日期', bold)
worksheet.write('A2', '日期', bold)
worksheet.write('B2', '选题名称', bold)
worksheet.write('C2', '编辑', bold)
worksheet.write('D2', '校对', bold)
worksheet.write('E2', '文字', bold)
worksheet.write('F2', '图片', bold)
worksheet.write('G2', '视频', bold)
worksheet.write('H2', '音频', bold)
worksheet.write('I2', '学生主编', bold)
worksheet.write('J2', '推文链接', bold)
time.sleep(0.5)
parser = etree.HTMLParser(encoding='utf-8')
line = 2
alignCenter = workbook.add_format({"align":"center"})
for i in range(len(siteList)-1,0,-1):
line = line + 1
# dict = {"time":NULL, "activity":NULL, "edit":NULL, "text":NULL, "proofreading":NULL, "picture":0, "video":NULL}
reponse = request.urlopen(siteList[i]).read().decode()
try:
pat1 = r"var ct = \"(\d+)\""
date1 = re.search(pat1, reponse).group(1)
date1 = int(date1)
#转换为其他日期格式,如:"%Y-%m-%d %H:%M:%S"
timeArray = time.localtime(date1)
otherStyleTime = time.strftime("%m月%d日", timeArray)
worksheet.write('A'+str(line), otherStyleTime,alignCenter)
print(otherStyleTime,end=" ")
except Exception as result:
try:
pat2 = r"window.ct = \'(\d+)\'"
date2 = re.search(pat2, reponse).group(1)
date2 = int(date2)
timeArray2 = time.localtime(date2)
otherStyleTime2 = time.strftime("%m月%d日", timeArray2)
worksheet.write('A'+str(line), otherStyleTime2,alignCenter)
print(otherStyleTime2,end=" ")
except Exception as result:
pass
html = etree.HTML(reponse)
activity_name = html.xpath("//h1")[0].text.strip() #标题
worksheet.write('B'+str(line), activity_name,alignCenter)
print(activity_name)
try:
bianJi = r"编辑:([\u4e00-\u9fa5].*?)<"
bianJi = re.search(bianJi, reponse).group(1).replace(" "," ")
worksheet.write('C'+str(line),bianJi,alignCenter)
# print(bianJi)
except Exception as result:
pass
try:
jiaoDui = r"校对:([\u4e00-\u9fa5].*?)<"
jiaoDui = re.search(jiaoDui, reponse).group(1).replace(" "," ")
worksheet.write('D'+str(line),jiaoDui,alignCenter)
#print(jiaoDui)
except Exception as result:
pass
try:
text = r"文字:([\u4e00-\u9fa5].*?)<"
text = re.search(text, reponse).group(1).replace(" "," ")
worksheet.write('E'+str(line),text,alignCenter)
#print(text)
except Exception as result:
pass
try:
picture = r"图片:([\u4e00-\u9fa5].*?)<"
picture = re.search(picture, reponse).group(1).replace(" "," ")
worksheet.write('F'+str(line),picture,alignCenter)
#print(picture)
except Exception as result:
pass
try:
video = r"视频:([\u4e00-\u9fa5].*?)<"
video = re.search(video, reponse).group(1).replace(" "," ")
worksheet.write('G'+str(line),video,alignCenter)
#print(video)
except Exception as result:
pass
try:
audio = r"音频:([\u4e00-\u9fa5].*?)<"
audio = re.search(audio, reponse).group(1).replace(" "," ")
worksheet.write('H'+str(line),audio,alignCenter)
#print(audio)
except Exception as result:
pass
try:
zhuBian = r"学生主编:([\u4e00-\u9fa5].*?)<"
zhuBian = re.search(zhuBian, reponse).group(1).replace(" "," ")
worksheet.write('I'+str(line),zhuBian,alignCenter)
#print(zhuBian)
except Exception as result:
pass
try:
worksheet.write('J'+str(line), siteList[i])
# print(siteList[i])
except Exception as result:
#print("pass")
pass
# 关闭工作薄
workbook.close()
if __name__ == '__main__':
# 获取文件的路径
if getattr(sys, 'frozen', False):
application_path = os.path.dirname(sys.executable)
elif __file__:
application_path = os.path.dirname(__file__)
path = application_path.replace("\\","/")+"/"
# 生成二维码
qr = qrcode.QRCode(
version=2,
error_correction=qrcode.constants.ERROR_CORRECT_L,
box_size=10,
border=1
)#设置二维码的大小
qr.add_data("https://www.cnblogs.com/MrFlySand/p/16844153.html")
qr.make(fit=True)
img = qr.make_image()
img.save(path+"0程序使用说明二维码.jpg")
print("程序使用说明二维码已只在到了当前文件夹\n")
print("文件个数从1开始读取")
maxNum = input("请输入最大的html文件个数:")
getSiteList = GetSiteList(1,int(maxNum)+1, path)
GetName(getSiteList,path)
print("\nExcel文件位置:"+path+"官微每月作品表.xlsx")
input("\n程序运行完毕")
标签:官微,bold,worksheet,etree,write,html,作品,import,记者团
From: https://www.cnblogs.com/MrFlySand/p/16944296.html