Fixed #18269 -- Applied unicode_literals for Python 3 compatibility.

Thanks Vinay Sajip for the support of his django3 branch and
Jannis Leidel for the review.

tests/regressiontests/templates/unicode.py

@@ -1,4 +1,6 @@
 # -*- coding: utf-8 -*-
+from __future__ import unicode_literals
+
 from django.template import Template, TemplateEncodingError, Context
 from django.utils.safestring import SafeData
 from django.utils.unittest import TestCase
@@ -7,7 +9,7 @@ from django.utils.unittest import TestCase
 class UnicodeTests(TestCase):
     def test_template(self):
         # Templates can be created from unicode strings.
-        t1 = Template(u'ŠĐĆŽćžšđ {{ var }}')
+        t1 = Template('ŠĐĆŽćžšđ {{ var }}')
         # Templates can also be created from bytestrings. These are assumed to
         # be encoded using UTF-8.
         s = b'\xc5\xa0\xc4\x90\xc4\x86\xc5\xbd\xc4\x87\xc5\xbe\xc5\xa1\xc4\x91 {{ var }}'
@@ -17,9 +19,9 @@ class UnicodeTests(TestCase):
 
         # Contexts can be constructed from unicode or UTF-8 bytestrings.
         c1 = Context({b"var": b"foo"})
-        c2 = Context({u"var": b"foo"})
-        c3 = Context({b"var": u"Đđ"})
-        c4 = Context({u"var": b"\xc4\x90\xc4\x91"})
+        c2 = Context({"var": b"foo"})
+        c3 = Context({b"var": "Đđ"})
+        c4 = Context({"var": b"\xc4\x90\xc4\x91"})
 
         # Since both templates and all four contexts represent the same thing,
         # they all render the same (and are returned as unicode objects and