Avoid converting to unicode if not needed

In python3, str is already represents unicode data. The current code
called oslo.utils safe_encode irrespective of incoming message encoding
and encoded it to 'utf-8', however that resolves to a byte string.

Since the syslog module expects a message of type unicode[1]. We don't
need to encode if python3 is being used, since the string representation
is sequence of unicode codepoints by default.

[1] https://hg.python.org/cpython/file/tip/Modules/syslogmodule.c#l162

Change-Id: Id40854d6b762b2d639afd7168ddb3a1e728d13d7

oslo_log/handlers.py

@@ -1,3 +1,4 @@
+# -*- coding: utf-8 -*-
 #    Licensed under the Apache License, Version 2.0 (the "License"); you may
 #    not use this file except in compliance with the License. You may obtain
 #    a copy of the License at
@@ -15,6 +16,7 @@ import logging
 import logging.config
 import logging.handlers
 import os
+import six
 try:
     import syslog
 except ImportError:
@@ -55,7 +57,29 @@ if syslog is not None:
         def emit(self, record):
             priority = self.severity_map.get(record.levelname,
                                              syslog.LOG_DEBUG)
-            message = encodeutils.safe_encode(self.format(record))
+            message = self.format(record)
+
+            # NOTE(gangila): In python2, the syslog function takes in 's' as
+            # the format argument, which means it either accepts python string
+            # (str = 'a') or unicode strings (str = u'a'), the PyArg_ParseTuple
+            # then if needed converts the unicode objects to C strings using
+            # the *default encoding*. This default encoding is 'ascii' in case
+            # of python2 while it has been changed to 'utf-8' in case of
+            # python3. What this leads to is when we supply a syslog message
+            # like:
+            # >>> syslog.syslog(syslog.LOG_DEBUG, u"François Deppierraz")
+            # In case of python2 the above fails with TypeError: [priority,]
+            # message string. Because python2 doesn't explicitly encode as
+            # 'utf-8' and use the system default encoding ascii, which raises
+            # a UnicodeEncodeError (UnicodeEncodeError: 'ascii' codec can't
+            # encode character u'\xe7' in position 4: ordinal not in
+            # range(128)), and hence the error message that's set in the code
+            # (TypeError: [priority,] message string) gets shown to the user.
+            # However, this in the case of Python3, where the system default
+            # encoding is 'utf-8' works without any issues. Therefore, we need
+            # to safe_encode in case of python2 and not in case of Python3.
+            if six.PY2:
+                message = encodeutils.safe_encode(self.format(record))
 
             syslog.syslog(priority, message)
 

oslo_log/tests/unit/test_log.py

@@ -291,7 +291,10 @@ class OSSysLogHandlerTestCase(BaseTestCase):
         handler.emit(
             logging.LogRecord("name", logging.INFO, "path", 123,
                               msg_unicode, None, None))
-        syslog.syslog.assert_called_once_with(syslog.LOG_INFO, msg_utf8)
+        if six.PY2:
+            syslog.syslog.assert_called_once_with(syslog.LOG_INFO, msg_utf8)
+        else:
+            syslog.syslog.assert_called_once_with(syslog.LOG_INFO, msg_unicode)
 
 
 class LogLevelTestCase(BaseTestCase):