MySQL 5.5 (refs #28552) was the last database to use it.

Reverted "Fixed #28248 -- Fixed password reset tokens being valid for 1 day longer than PASSWORD_RESET_TIMEOUT_DAYS."

This reverts commit 95993a89.

Regression introduced in 1f3dfd78.

Forward port of 60010417 from stable/2.0.x.

Per deprecation timeline.

Per deprecation timeline.

Refs #25187 -- Required the authenticate() method of authentication backends to have request as the first positional argument.

Per deprecation timeline.

Per deprecation timeline.

Per deprecation timeline.

Refs #20892 -- Removed support for passing pylibmc behavior settings as top-level attributes of CACHES['OPTIONS'].

Per deprecation timeline.

Per deprecation timeline.

Per deprecation timeline.

Thanks Anssi Kääriäinen for contributing to the patch.

Unused since its introduction in d725cc97.

_populate() sets the populating attribute to prevent infinite recursion in
case a urlconf includes itself. The flag is a threadlocal to avoid a race
condition [1] where one thread sets the flag and another checks it, then
proceeds to access data that's supposed to be populated (e.g. _reverse_dict)
but isn't yet.

The potential still exists for a thread to set the threadlocal, then be
interrupted by a signal such as SIGALRM and raise before resetting the
threadlocal flag. In this scenario, subsequent calls to _populate() in the
same thread will short-circuit erroneously.

The bulk of the method was already wrapped in a try/finally in df41b5a0, but
since a signal interrupt can occur at any line executed by the interpreter,
this moves up the try to ensure threadlocal gets reset.

[1]: https://groups.google.com/d/msg/django-developers/D_bIeinKHjE/4NmVQUJqAgAJ

Thanks Adam Johnson, Carl Meyer, Anssi Kääriäinen, Mariusz Felisiak,
Michael Manfre, and Tim Graham for discussion and review.