fix #2247

[t-8ch]

We have to pass urllib3 the url without the authentication information,
else it will be parsed by httplib as a netloc and included in the request line
and Host header

[Lukasa]

This looks right, but it's hard for me to know because URLs are a crazy mess. I'll let @sigmavirus24 review this, he knows URLs better than I do.

[sigmavirus24]

So if we used a better (read: less forgiving) URL parser library we wouldn't need to split on @ (like we do here). In principle this works fine. In reality, we should get something like rfc3986 into acceptable shape for @shazow or just make the Url object in urllib3 more RFC compliant (whichever works better) and use that for parsing the URL and reconstructing it. In short, you have 5 major components of the URL:

{scheme}://{authority}{/path}{?query}{#fragment}

And authority which we're dealing with right now has 3 sub-components so a URL would look like:

{scheme}://{userinfo@}{hostname}{:port}{/path}{?query}{#fragment}

rfc3986 would split this up and allow you to replace userinfo with None and then reconstruct the URL. For a quick fix, this is great. I'd rather not have so much URL/URI parsing logic in requests though, this is an HTTP library not an HTTP + URL + ... library.

[sigmavirus24]

Please remove the import of urldefrag here.

[t-8ch]

@sigmavirus24 done

[t-8ch]

+∞. I think it would be best to have URL objects that are immutable and an API like URL.replace(userinfo=None, fragment=None)

[sigmavirus24]

@t-8ch that's roughly rfc3986's API, except that I don't think I call it replace but maybe I do.... I'll double check later.

[shazow]

@sigmavirus24 Please open the specific issue with urllib3's url parser, no need to be backhanded. :)

[sigmavirus24]

Sorry @shazow, it wasn't meant to be back handed. I'll pull together the list of things the object is missing and make an issue with it tonight.

[shazow]

Thanks. :)

[kennethreitz42]

What's the status of this?

[sigmavirus24]

@kennethreitz 👍