Redirects in the Werkzeug browser retain method, headers and params

[jh-sportcaster]

capybara.werkzeug.browser.Browser._process_and_follow_redirects handles redirects in what I would consider a wrong way. I'm talking specifically about

capybara.py/capybara/werkzeug/browser.py

Line 61 in 566b136

self._process(method, path, params, headers)

What it does at the moment is to retain basically all data from the original request, except the path, which it overrides. This means that if you for instance make a POST request to a URL, and that URL redirects to another page, the second request will be the same. This is a violation of HTTP/1.1 for the 303 status code (which demands a GET request for the new location), and – worse – inconsistent with how all browsers I know of handle 302 and possibly 301.

Thus, my proposal would be a diff of the following nature:

-                self._process(method, path, params, headers)
+                self._process('GET', path, None, headers)

Whatever params would be needed for the subsequent request would be indicated in the response's Location header.

I am unsure whether headers should be retained in its entirety, but I don't know what would be consistent with browser behaviour.

[elliterate]

Thanks for the bug report!

You're spot-on about the need to use GET and exclude the params. The headers should be fine—indeed, necessary—as they're only used internally to specify the Referer.