302s and losing cookies with urllib2

问题:

I am using liburl2 with CookieJar / HTTPCookieProcessor in an attempt to simulate a login to a page to automate an upload.

I've seen some questions and answers on this, but nothing which solves my problem. I am losing my cookie when I simulate the login which ends up at a 302 redirect. The 302 response is where the cookie gets set by the server, but urllib2 HTTPCookieProcessor does not seem to save the cookie during a redirect. I tried creating a HTTPRedirectHandler class to ignore the redirect, but that didn't seem to do the trick. I tried referencing the CookieJar globally to handle the cookies from the HTTPRedirectHandler, but 1. This didn't work (because I was handling the header from the redirector, and the CookieJar function that I was using, extract_cookies, needed a full request) and 2. It's an ugly way to handle it.

I probably need some guidance on this as I'm fairly green with Python. I think I'm mostly barking up the right tree here, but maybe focusing on the wrong branch.

cj = cookielib.CookieJar()
cookieprocessor = urllib2.HTTPCookieProcessor(cj)


class MyHTTPRedirectHandler(urllib2.HTTPRedirectHandler):
  def http_error_302(self, req, fp, code, msg, headers):
    global cj
    cookie = headers.get("set-cookie")
    if cookie:
      # Doesn't work, but you get the idea
      cj.extract_cookies(headers, req)

    return urllib2.HTTPRedirectHandler.http_error_302(self, req, fp, code, msg, headers)

  http_error_301 = http_error_303 = http_error_307 = http_error_302

cookieprocessor = urllib2.HTTPCookieProcessor(cj)

# Oh yeah.  I'm using a proxy too, to follow traffic.
proxy = urllib2.ProxyHandler({'http': '127.0.0.1:8888'})
opener = urllib2.build_opener(MyHTTPRedirectHandler, cookieprocessor, proxy)

Addition: I had tried using mechanize as well, without success. This is probably a new question, but I'll pose it here since it is the same ultimate goal:

This simple code using mechanize, when used with a 302 emitting url (http://fxfeeds.mozilla.com/firefox/headlines.xml) -- note that the same behavior occurs when not using set_handle_robots(False). I just wanted to ensure that wasn't it:

import urllib2, mechanize

browser = mechanize.Browser()
browser.set_handle_robots(False)
opener = mechanize.build_opener(*(browser.handlers))
r = opener.open("http://fxfeeds.mozilla.com/firefox/headlines.xml")

Output:

Traceback (most recent call last):
  File "redirecttester.py", line 6, in <module>
    r = opener.open("http://fxfeeds.mozilla.com/firefox/headlines.xml")
  File "build/bdist.macosx-10.6-universal/egg/mechanize/_opener.py", line 204, in open
  File "build/bdist.macosx-10.6-universal/egg/mechanize/_urllib2_fork.py", line 457, in http_response
  File "build/bdist.macosx-10.6-universal/egg/mechanize/_opener.py", line 221, in error
  File "build/bdist.macosx-10.6-universal/egg/mechanize/_urllib2_fork.py", line 332, in _call_chain
  File "build/bdist.macosx-10.6-universal/egg/mechanize/_urllib2_fork.py", line 571, in http_error_302
  File "build/bdist.macosx-10.6-universal/egg/mechanize/_opener.py", line 188, in open
  File "build/bdist.macosx-10.6-universal/egg/mechanize/_mechanize.py", line 71, in http_request
AttributeError: OpenerDirector instance has no attribute '_add_referer_header'

Any ideas?

回答1:

I have been having the exact same problem recently but in the interest of time scrapped it and decided to go with mechanize. It can be used as a total replacement for urllib2 that behaves exactly as you would expect a browser to behave with regards to Referer headers, redirects, and cookies.

import mechanize
cj = mechanize.CookieJar()
browser = mechanize.Browser()
browser.set_cookiejar(cj)
browser.set_proxies({'http': '127.0.0.1:8888'})

# Use browser's handlers to create a new opener
opener = mechanize.build_opener(*browser.handlers)

The Browser object can be used as an opener itself (using the .open() method). It maintains state internally but also returns a response object on every call. So you get a lot of flexibility.

Also, if you don't have a need to inspect the cookiejar manually or pass it along to something else, you can omit the explicit creation and assignment of that object as well.

I am fully aware this doesn't address what is really going on and why urllib2 can't provide this solution out of the box or at least without a lot of tweaking, but if you're short on time and just want it to work, just use mechanize.

回答2:

Depends on how the redirect is done. If it's done via a HTTP Refresh, then mechanize has a HTTPRefreshProcessor you can use. Try to create an opener like this:

cj = mechanize.CookieJar()
opener = mechanize.build_opener(
    mechanize.HTTPCookieProcessor(cj),
    mechanize.HTTPRefererProcessor,
    mechanize.HTTPEquivProcessor,
    mechanize.HTTPRefreshProcessor)

回答3:

I've just got a variation of the below working for me, at least when trying to read Atom from http://www.fudzilla.com/home?format=feed&type=atom

I can't verify that the below snippet will run as-is, but might give you a start:

import cookielib
cookie_jar = cookielib.LWPCookieJar()
cookie_handler = urllib2.HTTPCookieProcessor(cookie_jar)
handlers = [cookie_handler] #+others, we have proxy + progress handlers
opener = apply(urllib2.build_opener, tuple(handlers + [_FeedURLHandler()])) #see http://code.google.com/p/feedparser/source/browse/trunk/feedparser/feedparser.py#2848 for implementation of _FeedURLHandler
opener.addheaders = [] #may not be needed but see the comments around the link referred to below
try:
    return opener.open(request) #see http://code.google.com/p/feedparser/source/browse/trunk/feedparser/feedparser.py#2954 for implementation of request
finally:
    opener.close()

回答4:

I was also having the same problem where the server would respond to the login POST request with a 302 and the session token in the Set-Cookie header. Using Wireshark it was clearly visible that urllib was following the redirect but not including the session token in the Cookie.

I literally just ripped out urllib and did a direct replacement with requests and it worked perfectly first time without having to change a thing. Big props to those guys.