Add Python 2.3 support

This commit also adds the 'six' dependency

vcr/serializers/compat.py

@@ -0,0 +1,66 @@
+import six
+
+def convert_to_bytes(resp):
+    resp = convert_headers_to_bytes(resp)
+    resp = convert_body_to_bytes(resp)
+    return resp
+
+def convert_to_unicode(resp):
+    resp = convert_headers_to_unicode(resp)
+    resp = convert_body_to_unicode(resp)
+    return resp
+
+def convert_headers_to_bytes(resp):
+    try:
+        resp['headers'] = [h.encode('utf-8') for h in resp['headers']]
+    except (KeyError, TypeError):
+        pass
+    return resp
+
+def convert_headers_to_unicode(resp):
+    try:
+        resp['headers'] = [h.decode('utf-8') for h in resp['headers']]
+    except (KeyError, TypeError):
+        pass
+    return resp
+
+def convert_body_to_bytes(resp):
+    """
+    If the request body is a string, encode it to bytes (for python3 support)
+
+    By default yaml serializes to utf-8 encoded bytestrings.
+    When this cassette is loaded by python3, it's automatically decoded
+    into unicode strings.  This makes sure that it stays a bytestring, since
+    that's what all the internal httplib machinery is expecting.
+
+    For more info on py3 yaml: http://pyyaml.org/wiki/PyYAMLDocumentation#Python3support
+    """
+    try:
+        if not isinstance(resp['body']['string'], six.binary_type):
+            resp['body']['string'] = resp['body']['string'].encode('utf-8')
+    except (KeyError, TypeError, UnicodeEncodeError):
+        # The thing we were converting either wasn't a dictionary or didn't have
+	# the keys we were expecting.  Some of the tests just serialize and
+	# deserialize a string.
+
+	# Also, sometimes the thing actually is binary, so if you can't encode
+	# it, just give up.
+        pass
+    return resp
+
+def convert_body_to_unicode(resp):
+    """
+    If the request body is bytes, decode it to a string (for python3 support)
+    """
+    try:
+        if not isinstance(resp['body']['string'], six.text_type):
+            resp['body']['string'] = resp['body']['string'].decode('utf-8')
+    except (KeyError, TypeError, UnicodeDecodeError):
+        # The thing we were converting either wasn't a dictionary or didn't have
+	# the keys we were expecting.  Some of the tests just serialize and
+	# deserialize a string.
+
+	# Also, sometimes the thing actually is binary, so if you can't decode
+	# it, just give up.
+        pass
+    return resp

vcr/serializers/jsonserializer.py

@@ -1,4 +1,5 @@
 from vcr.request import Request
+from . import compat
 try:
     import simplejson as json
 except ImportError:
@@ -11,22 +12,17 @@ def _json_default(obj):
     return obj
 
 
-def _fix_response_unicode(d):
-    d['body']['string'] = d['body']['string'].encode('utf-8')
-    return d
-
-
 def deserialize(cassette_string):
     data = json.loads(cassette_string)
     requests = [Request._from_dict(r['request']) for r in data]
-    responses = [_fix_response_unicode(r['response']) for r in data]
+    responses = [compat.convert_to_bytes(r['response']) for r in data]
     return requests, responses
 
 
 def serialize(cassette_dict):
     data = ([{
         'request': request._to_dict(),
-        'response': response,
+        'response': compat.convert_to_unicode(response),
     } for request, response in zip(
         cassette_dict['requests'],
         cassette_dict['responses']

vcr/serializers/yamlserializer.py

@@ -1,4 +1,6 @@
+import sys
 import yaml
+from . import compat
 
 # Use the libYAML versions if possible
 try:
@@ -6,11 +8,38 @@ try:
 except ImportError:
     from yaml import Loader, Dumper
 
+"""
+Just a general note on the serialization philosophy here:
+I prefer cassettes to be human-readable if possible.  Yaml serializes
+bytestrings to !!binary, which isn't readable, so I would like to serialize to
+strings and from strings, which yaml will encode as utf-8 automatically.
+All the internal HTTP stuff expects bytestrings, so this whole serialization
+process feels backwards.
+
+Serializing: bytestring -> string (yaml persists to utf-8)
+Deserializing: string (yaml converts from utf-8) -> bytestring
+"""
+
+def _restore_frozenset():
+    """
+    Restore __builtin__.frozenset for cassettes serialized in python2 but 
+    deserialized in python3 and builtins.frozenset for cassettes serialized
+    in python3 and deserialized in python2
+    """
+
+    if '__builtin__' not in sys.modules:
+        import builtins
+        sys.modules['__builtin__'] = builtins
+
+    if 'builtins' not in sys.modules:
+        sys.modules['builtins'] = sys.modules['__builtin__']
 
 def deserialize(cassette_string):
+    _restore_frozenset()
     data = yaml.load(cassette_string, Loader=Loader)
     requests = [r['request'] for r in data]
     responses = [r['response'] for r in data]
+    responses = [compat.convert_to_bytes(r['response']) for r in data]
     return requests, responses
 
 
@@ -20,6 +49,6 @@ def serialize(cassette_dict):
         'response': response,
     } for request, response in zip(
         cassette_dict['requests'],
-        cassette_dict['responses']
+        [compat.convert_to_unicode(r) for r in cassette_dict['responses']],
     )])
     return yaml.dump(data, Dumper=Dumper)