支付宝十年账单上的数字有点吓人,但它统计的项目太多,只是想看看到底单纯在淘宝上支出了多少,于是写了段脚本,统计任意时间段淘宝订单的消费情况,看那结果其实在淘宝上我还是相当节约的说。
脚本的主要工作是模拟了浏览器登录,解析“已买到的宝贝”页面以获得指定的订单及宝贝信息。

Python模拟登陆淘宝并统计淘宝消费情况的代码实例分享-LMLPHP

使用方法见代码或执行命令加参数-h,另外需要BeautifulSoup4支持,BeautifulSoup的官方项目列表页:https://www.crummy.com/software/BeautifulSoup/bs4/download/

首先来说一下代码使用方法:

python taobao.py -u USERNAME -p PASSWORD -s START-DATE -e END-DATE --verbose
登录后复制

所有参数均可选,如:

python taobao.py -u jinnlynn 
登录后复制

统计用户jinnlynn所有订单的情况

python taobao.py -s 2014-12-12 -e 2014-12-12
登录后复制

统计用户(用户名在命令执行时会要求输入)在2014-12-12当天的订单情况

python taobao.py --verbose
登录后复制

这样就可以统计并输出订单明细。

好了,说了这么多我们就来看代码吧:

from __future__ import unicode_literals, print_function, absolute_import, division
import urllib
import urllib2
import urlparse
import cookielib
import re
import sys
import os
import json
import subprocess
import argparse
import platform
from getpass import getpass
from datetime import datetime
from pprint import pprint

try:
  from bs4 import BeautifulSoup
except ImportError:
  sys.exit('BeautifulSoup4 missing.')

__version__ = '1.0.0'
__author__ = 'JinnLynn'
__copyright__ = 'Copyright (c) 2014 JinnLynn'
__license__ = 'The MIT License'

HEADERS = {
  'x-requestted-with' : 'XMLHttpRequest',
  'Accept-Language' : 'zh-cn',
  'Accept-Encoding' : 'gzip, deflate',
  'ContentType' : 'application/x-www-form-urlencoded; chartset=UTF-8',
  'Cache-Control' : 'no-cache',
  'User-Agent' :'Mozilla/5.0 (Macintosh; Intel Mac OS X 10_9_5) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/40.0.2214.38 Safari/537.36',
  'Connection' : 'Keep-Alive'
}

DEFAULT_POST_DATA = {
  'TPL_username' : '', #用户名
  'TPL_password' : '', #密码
  'TPL_checkcode' : '',
  'need_check_code' : 'false',
  'callback' : '0', # 有值返回JSON
}

# 无效订单状态
INVALID_ORDER_STATES = [
  'CREATE_CLOSED_OF_TAOBAO', # 取消
  'TRADE_CLOSED', # 订单关闭
]

LOGIN_URL = 'https://login.taobao.com/member/login.jhtml'

RAW_IMPUT_ENCODING = 'gbk' if platform.system() == 'Windows' else 'utf-8'

def _request(url, data, method='POST'):
  if data:
    data = urllib.urlencode(data)
  if method == 'GET':
    if data:
      url = '{}?{}'.format(url, data)
    data = None
  # print(url)
  # print(data)
  req = urllib2.Request(url, data, HEADERS)
  return urllib2.urlopen(req)

def stdout_cr(msg=''):
  sys.stdout.write('\r{:10}'.format(' '))
  sys.stdout.write('\r{}'.format(msg))
  sys.stdout.flush()

def get(url, data=None):
  return _request(url, data, method='GET')

def post(url, data=None):
  return _request(url, data, method='POST')

def login_post(data):
  login_data = DEFAULT_POST_DATA
  login_data.update(data)
  res = post(LOGIN_URL, login_data)
  return json.load(res, encoding='gbk')

def login(usr, pwd):
  data = {
    'TPL_username' : usr.encode('utf-8' if platform.system() == 'Windows' else 'GB18030'),
    'TPL_password' : pwd
  }

  # 1. 尝试登录
  ret = login_post(data)
  while not ret.get('state', False):
    code = ret.get('data', {}).get('code', 0)
    if code == 3425 or code == 1000:
      print('INFO: {}'.format(ret.get('message')))
      check_code = checkcode(ret.get('data', {}).get('ccurl'))
      data.update({'TPL_checkcode' : check_code, 'need_check_code' : 'true'})
      ret = login_post(data)
    else:
      sys.exit('ERROR. code: {}, message:{}'.format(code, ret.get('message', '')))

  token = ret.get('data', {}).get('token')
  print('LOGIN SUCCESS. token: {}'.format(token))

  # 2. 重定向
  # 2.1 st值
  res = get('https://passport.alipay.com/mini_apply_st.js', {
    'site' : '0',
    'token' : token,
    'callback' : 'stCallback4'})
  content = res.read()
  st = re.search(r'"st":"(\S*)"( |})', content).group(1)
  # 2.1 重定向
  get('http://login.taobao.com/member/vst.htm',
    {'st' : st, 'TPL_uesrname' : usr.encode('GB18030')})

def checkcode(url):
  filename, _ = urllib.urlretrieve(url)
  if not filename.endswith('.jpg'):
    old_fn = filename
    filename = '{}.jpg'.format(filename)
    os.rename(old_fn, filename)
  if platform.system() == 'Darwin':
    # mac 下直接preview打开
    subprocess.call(['open', filename])
  elif platform.system() == 'Windows':
    # windows 执行文件用默认程序打开
    subprocess.call(filename, shell=True)
  else:
    # 其它系统 输出文件名
    print('打开该文件获取验证码: {}'.format(filename))
  return raw_input('输入验证码: '.encode(RAW_IMPUT_ENCODING))

def parse_bought_list(start_date=None, end_date=None):
  url = 'http://buyer.trade.taobao.com/trade/itemlist/list_bought_items.htm'
  #         运费险      增值服务     分段支付(定金,尾款)
  extra_service = ['freight-info', 'service-info', 'stage-item']

  stdout_cr('working... {:.0%}'.format(0))
  # 1. 解析第一页
  res = urllib2.urlopen(url)
  soup = BeautifulSoup(res.read().decode('gbk'))
  # 2. 获取页数相关
  page_jump = soup.find('span', id='J_JumpTo')
  jump_url = page_jump.attrs['data-url']
  url_parts = urlparse.urlparse(jump_url)
  query_data = dict(urlparse.parse_qsl(url_parts.query))
  total_pages = int(query_data['tPage'])

  # 解析
  orders = []
  cur_page = 1
  out_date = False
  errors = []
  while True:
    bought_items = soup.find_all('tbody', attrs={'data-orderid' : True})
    # pprint(len(bought_items))
    count = 0
    for item in bought_items:
      count += 1
      # pprint('{}.{}'.format(cur_page, count))
      try:
        info = {}
        # 订单在页面上的位置 页数.排序号
        info['pos'] = '{}.{}'.format(cur_page, count)
        info['orderid'] = item.attrs['data-orderid']
        info['status'] = item.attrs['data-status']
        # 店铺
        node = item.select('tr.order-hd a.shopname')
        if not node:
          # 店铺不存在,可能是赠送彩票订单,忽略
          # print('ignore')
          continue
        info['shop_name'] = node[0].attrs['title'].strip()
        info['shop_url'] = node[0].attrs['href']
        # 日期
        node = item.select('tr.order-hd span.dealtime')[0]
        info['date'] = datetime.strptime(node.attrs['title'], '%Y-%m-%d %H:%M')

        if end_date and info['date'].toordinal() > end_date.toordinal():
          continue

        if start_date and info['date'].toordinal() < start_date.toordinal():
          out_date = True
          break

        # 宝贝
        baobei = []
        node = item.find_all('tr', class_='order-bd')
        # pprint(len(node))
        for n in node:
          try:
            bb = {}
            if [True for ex in extra_service if ex in n.attrs['class']]:
              # 额外服务处理
              # print('额外服务处理')
              name_node = n.find('td', class_='baobei')
              # 宝贝地址
              bb['name'] = name_node.text.strip()
              bb['url'] = ''
              bb['spec'] = ''
              # 宝贝快照
              bb['snapshot'] = ''
              # 宝贝价格
              bb['price'] = 0.0
              # 宝贝数量
              bb['quantity'] = 1
              bb['is_goods'] = False
              try:
                bb['url'] = name_node.find('a').attrs['href']
                bb['price'] = float(n.find('td', class_='price').text)
              except:
                pass
            else:
              name_node = n.select('p.baobei-name a')
              # 宝贝地址
              bb['name'] = name_node[0].text.strip()
              bb['url'] = name_node[0].attrs['href']
              # 宝贝快照
              bb['snapshot'] = ''
              if len(name_node) > 1:
                bb['snapshot'] = name_node[1].attrs['href']
              # 宝贝规格
              bb['spec'] = n.select('.spec')[0].text.strip()
              # 宝贝价格
              bb['price'] = float(n.find('td', class_='price').attrs['title'])
              # 宝贝数量
              bb['quantity'] = int(n.find('td', class_='quantity').attrs['title'])
              bb['is_goods'] = True
            baobei.append(bb)
            # 尝试获取实付款
            # 实付款所在的节点可能跨越多个tr的td
            amount_node = n.select('td.amount em.real-price')
            if amount_node:
              info['amount'] = float(amount_node[0].text)
          except Exception as e:
            errors.append({
              'type' : 'baobei',
              'id' : '{}.{}'.format(cur_page, count),
              'node' : '{}'.format(n),
              'error' : '{}'.format(e)
            })
      except Exception as e:
        errors.append({
          'type' : 'order',
          'id' : '{}.{}'.format(cur_page, count),
          'node' : '{}'.format(item),
          'error' : '{}'.format(e)
        })

      info['baobei'] = baobei
      orders.append(info)

    stdout_cr('working... {:.0%}'.format(cur_page / total_pages))

    # 下一页
    cur_page += 1
    if cur_page > total_pages or out_date:
      break
    query_data.update({'pageNum' : cur_page})
    page_url = '{}?{}'.format(url, urllib.urlencode(query_data))
    res = urllib2.urlopen(page_url)
    soup = BeautifulSoup(res.read().decode('gbk'))

  stdout_cr()
  if errors:
    print('INFO. 有错误发生,统计结果可能不准确。')
    # pprint(errors)
  return orders

def output(orders, start_date, end_date):
  amount = 0.0
  org_amount = 0
  baobei_count = 0
  order_count = 0
  invaild_order_count = 0
  for order in orders:
    if order['status'] in INVALID_ORDER_STATES:
      invaild_order_count += 1
      continue
    amount += order['amount']
    order_count += 1
    for baobei in order.get('baobei', []):
      if not baobei['is_goods']:
        continue
      org_amount += baobei['price'] * baobei['quantity']
      baobei_count += baobei['quantity']

  print('{:<9} {}'.format('累计消费:', amount))
  print('{:<9} {}/{}'.format('订单/宝贝:', order_count, baobei_count))
  if invaild_order_count:
    print('{:<9} {} (退货或取消等, 不在上述订单之内)'.format('无效订单:', invaild_order_count))
  print('{:<7} {}'.format('宝贝原始总价:', org_amount))
  print('{:<7} {:.2f}'.format('宝贝平均单价:', 0 if baobei_count == 0 else org_amount / baobei_count))
  print('{:<9} {} ({:.2%})'.format('节约了(?):',
                   org_amount - amount,
                   0 if org_amount == 0 else (org_amount - amount) / org_amount))
  from_date = start_date if start_date else orders[-1]['date']
  to_date = end_date if end_date else datetime.now()
  print('{:<9} {:%Y-%m-%d} - {:%Y-%m-%d}'.format('统计区间:', from_date, to_date))
  if not start_date:
    print('{:<9} {:%Y-%m-%d %H:%M}'.format('败家始于:', orders[-1]['date']))

def ouput_orders(orders):
  print('所有订单:')
  if not orders:
    print(' --')
    return
  for order in orders:
    print(' {:-^20}'.format('-'))
    print(' * 订单号: {orderid} 实付款: {amount} 店铺: {shop_name} 时间: {date:%Y-%m-%d %H:%M}'.format(**order))
    for bb in order['baobei']:
      if not bb['is_goods']:
        continue
      print('  - {name}'.format(**bb))
      if bb['spec']:
        print('   {spec}'.format(**bb))
      print('   {price} X {quantity}'.format(**bb))

def main():
  parser = argparse.ArgumentParser(
    prog='python {}'.format(__file__)
  )
  parser.add_argument('-u', '--username', help='淘宝用户名')
  parser.add_argument('-p', '--password', help='淘宝密码')
  parser.add_argument('-s', '--start', help='起始时间,可选, 格式如: 2014-11-11')
  parser.add_argument('-e', '--end', help='结束时间,可选, 格式如: 2014-11-11')
  parser.add_argument('--verbose', action='store_true', default=False, help='订单详细输出')
  parser.add_argument('-v', '--version', action='version',
            version='v{}'.format(__version__), help='版本号')
  args = parser.parse_args()

  usr = args.username
  if not usr:
    usr = raw_input('输入淘宝用户名: '.encode(RAW_IMPUT_ENCODING))
  usr = usr.decode('utf-8') # 中文输入问题
  pwd = args.password
  if not pwd:
    if platform.system() == 'Windows':
      # Windows下中文输出有问题
      pwd = getpass()
    else:
      pwd = getpass('输入淘宝密码: '.encode('utf-8'))

  pwd = pwd.decode('utf-8')

  verbose = args.verbose

  start_date = None
  if args.start:
    try:
      start_date = datetime.strptime(args.start, '%Y-%m-%d')
    except Exception as e:
      sys.exit('ERROR. {}'.format(e))

  end_date = None
  if args.end:
    try:
      end_date = datetime.strptime(args.end, '%Y-%m-%d')
    except Exception as e:
      sys.exit('ERROR. {}'.format(e))

  if start_date and end_date and start_date > end_date:
    sys.exit('ERROR, 结束日期必须晚于或等于开始日期')

  cj_file = './{}.tmp'.format(usr)
  cj = cookielib.LWPCookieJar()
  try:
    cj.load(cj_file)
  except:
    pass
  opener = urllib2.build_opener(urllib2.HTTPCookieProcessor(cj), urllib2.HTTPHandler)
  urllib2.install_opener(opener)

  login(usr, pwd)

  try:
    cj.save(cj_file)
  except:
    pass

  orders = parse_bought_list(start_date, end_date)
  output(orders, start_date, end_date)

  # 输出订单明细
  if verbose:
    ouput_orders(orders)

if __name__ == '__main__':
  main()
登录后复制

09-14 07:51