Fix AttributeError on login

[qwertyuu]

Facebook decided they needed an "Accept" header to return data on request and that broke fbchat

[qwertyuu]

Seems to fix:
#584
#567
#555
#551
#565

Maybe fixes
#556
Callen93 confirmed that it was not fixed by this PR (yet) 🤷

[Callen93]

Hello,
I don't know if it's the right place to discuss this but as it seems to exists multiple login/logout problem right now I want to expose some conclusion and try to find solution. Thank you by the way for this fix.

I think the update of facebook with new UI did broke a lot of things. The thing is activating/desactivating the UI change (or at least changed) the behaviour of facebook against fbchat. For instance on friday (can't reproduce today), I could login with fbchat only when I desactivated new UI. So one fix can work for people with old UI and not people with new UI.

Lets first focus on branch carpedm20:v1 :

• Connecting with session cookies work with your fix.
• Trying to connect with login password fail and block account redirecting causing this error fbchat._exception.FBchatUserError: Login failed. Check email/password. (Failed on url: https://m.facebook.com/checkpoint/?_rdr) . This is even the case when using the same user agent as when, connecting with webUI. Login failed #477 Cannot login, Facebook detects the API then redirects to checkpoint #532
• I can't reproduce the error 'NoneType' object has no attribute 'group' Cannot login  #567 Login failed for 2FA enabled account. #555 Unable to login after last facebook update #551 Login issue AttributeError: 'NoneType' object has no attribute 'group' #565 Problem Logging In #584
• Error with logout function #556 is not fixed by this commit

On master version, it's a bit worse. I used your change on master version and get

• When trying to connect with session cookies fbchat.NotLoggedIn: Found empty fb_dtsg, the session was probably invalid.
• If trying to apply proposed changes in Change method search fb_dtsg #552 same problem but message error is fbchat.NotLoggedIn: Could not find fb_dtsg
• When trying to connect via login/password, It still fail and block temporary my account.

I'm still trying to investigate on this and come back at you if I find something interesting.

[qwertyuu]

@Callen93 Things I can tell you about my setup and why it fixed the problems I had:

1. I am not using the new Facebook UI
2. I do not have 2FA (just stating this makes me want to enable it, I feel like someone will scan through the github comments and try to hack me)

I went through the real messenger.com and saw that the headers differed from what the "Sessions" client headers tried to send to facebook. By replicating my browser's headers, I was able to get a response in brotli form. When I commented out all the unnecessary headers, I ended up with only the one I uploaded through the fix. My initial problem was 'NoneType' object has no attribute 'group' and it happened because r.text at that place was an empty string (facebook just didn't respond any content but it as a 200 OK response anyways). You are right, I think all these login/logout problems are from the new UI update and may have impacted the old one too.

[madsmtm]

Hmm, we're using requests, which should set the Accept: */* pr. default, which is the same thing browsers do! So I don't really understand this, are you sure this actually fixes anything, and isn't just by coincidence that it suddenly worked?

Regarding the new UI, that is partly the reason why I changed the login logic on the master branch to use messenger.com urls, see 079d409. Haven't had the time to backport a lot of the fixes (and bugs) in there to v1, so that's the primary reason v1 can't handle the new UI

[madsmtm]

Theory: Using Accept: text/html triggers legacy code, so that the new UI isn't activated?

[qwertyuu]

are you sure this actually fixes anything, and isn't just by coincidence that it suddenly worked?

This is a hard question because I don't have time to test it elsewhere and it's hard to be sure unless many others try it and report back as positive. In my case it was a causal effect: adding that line made fbchat login, while removing it had a devastating effect (empty response body so all the regexes fired on an empty string)

The only way I can be pretty sure it works is that @Callen93 tried it and reported back as positive, and this is the closest I can be to being sure it at least fixes the login procedure.

[madsmtm]

Well, I think that's good enough for me, I'll take it, thanks!