django.utils.encoding.force_text() should call unicode() instead of __unicode__()

[lie.1296@…]

Unlike the unicode() builtin function, the method __unicode__() does not validate its return type.

If people wrote a buggy __unicode__ which does not return str or unicode, this can lead to a situation where force_text() will return non-text, and so the bug may only appear much, much later when other parts of the application assumes they were getting text and often it will become very difficult to find how the non-string object came from.

Expected result:

force_text() should be switched to use unicode() so returning non-text from __unicode__() will immediately raise an error at the point where the error would have been obvious.