Incorrect QUERY_STRING handling on Python 3

[Armin Ronacher]

Certain browsers (IE cough) will not fully encode the path in the request in all situations. As such you will encounter non ascii letters in the request line. Currently the QueryDict does not handle that properly. In addition to that it also means that the WSGI QUERY_STRING variable needs to be handled the same way as PATH_INFO and SCRIPT_NAME.

Here is what is necessary to handle the case properly:

1. the environ['QUERY_STRING'] attribute needs to go through the PEP 3333 dance on Python 3 that creates a bytes object
2. unquoting happens on the bytes
3. finally everything is done to the intended encoding (UTF-8)

The logic currently employed by QueryDict in combination with the WSGIRequest object is double wrong:

1. the WSGIRequest object is not properly doing the dance and passes a (potentially mangled) unicode string to query dict
2. the query dict decodes that incorrectly formatted unicode string (WSGI on 3.x intentionally incorrectly encodes information) causing invalid data to show up in request.args

Independently of that if bytes are passed to the QueryDict it does not do proper decoding unless the bytes are a subset of ASCII.

[Aymeric Augustin]

Thanks for the report. I'll take care of that.

[Aymeric Augustin]

Item 1 above was fixed by ​https://github.com/django/django/commit/8aaca651cf5732bbf395d24a7d9f2edfab00250c#L0L136

[Aymeric Augustin <aymeric.augustin@…>]

In 7bb627936034c1b9500a8d250cce75b30f980b23:

Fixed an encoding issue in the test client.

Fixed
comment_tests.tests.test_comment_view.CommentViewTests.testCommentPostRedirectWithInvalidIntegerPK.

Refs #20530.

[Aymeric Augustin <aymeric.augustin@…>]

In 65b6eff322a4a3331601e111934dee95c090961c:

Fixed #20530 -- Properly decoded non-ASCII query strings on Python 3.

Thanks mitsuhiko for the report.

This commit just adds a test since the problem was fixed in 8aaca651.

[Aymeric Augustin <aymeric.augustin@…>]

In 9244447cc4a91c22f8f2668f9667e92a1b2de958:

[1.6.x] Fixed an encoding issue in the test client.

Refs #20530.

Backport of 7bb62793 and 476b0764 from master.

Conflicts:

django/test/client.py

[Aymeric Augustin <aymeric.augustin@…>]

In 7fcd6aa6695b39370154d6993cdbb3ba4363de91:

[1.6.x] Fixed #20530 -- Properly decoded non-ASCII query strings on Python 3.

Thanks mitsuhiko for the report.

Backport of 65b6eff3 and adaptation of 8aaca65 from master.

[Aymeric Augustin <aymeric.augustin@…>]

In 63b95ca452ea7ef1103e599f8dd733b67278c8dc:

[1.6.x] Fixed 9244447c -- incomplete backport.

The test client had been refactored in the mean time. This commit
de-factors the fix. Refs #20530.

[Tim Graham <timograham@…>]

In 4206765:

Refs #20530 -- Added a test for test client query string encoding.

7bb627936034c1b9500a8d250cce75b30f980b23 is only tested in
django-contrib-comments.