Fix a regression where reindent filter failed when comments contained non-ascii chars.

2 files changed, 10 insertions, 2 deletions

diff --git a/sqlparse/filters.py b/sqlparse/filters.py
index 2f1c825..e238f69 100644
--- a/sqlparse/filters.py
+++ b/sqlparse/filters.py
@@ -288,8 +288,8 @@ class ReindentFilter:
                 offset += 1
             if (prev
                 and isinstance(prev, sql.Comment)
-                and (str(prev).endswith('\n')
-                     or str(prev).endswith('\r'))):
+                and (unicode(prev).endswith('\n')
+                     or unicode(prev).endswith('\r'))):
                 nl = tlist.token_next(token)
             else:
                 nl = self.nl()
diff --git a/tests/test_regressions.py b/tests/test_regressions.py
index ba156f0..d6108d5 100644
--- a/tests/test_regressions.py
+++ b/tests/test_regressions.py
@@ -157,3 +157,11 @@ $_$;
 ALTER TABLE..... ;"""
     t = sqlparse.split(sql)
     assert len(t) == 3
+
+
+def test_comment_encoding_when_reindent():
+    # There was an UnicodeEncodeError in the reindent filter that
+    # casted every comment followed by a keyword to str.
+    sql = u'select foo -- Comment containing Ümläuts\nfrom bar'
+    formatted = sqlparse.format(sql, reindent=True)
+    assert formatted == sql