Fixed yet another issue with smmap's latest changes

Now we deal with memory views as well ...

Author: Byron

gitdb/pack.py

@@ -62,7 +62,12 @@
 from binascii import crc32
 
 from gitdb.const import NULL_BYTE
-from gitdb.utils.compat import izip, buffer, xrange
+from gitdb.utils.compat import (
+    izip, 
+    buffer, 
+    xrange,
+    to_bytes
+)
 
 import tempfile
 import array
@@ -864,7 +869,7 @@ def collect_streams_at_offset(self, offset):
             stream = streams[-1]
             while stream.type_id in delta_types:
                 if stream.type_id == REF_DELTA:
-                    sindex = self._index.sha_to_index(stream.delta_info)
+                    sindex = self._index.sha_to_index(to_bytes(stream.delta_info))
                     if sindex is None:
                         break
                     stream = self._pack.stream(self._index.offset(sindex))

gitdb/utils/compat.py

@@ -15,6 +15,9 @@
     # Python 2
     buffer = buffer
     memoryview = buffer
+    # Assume no memory view ...
+    def to_bytes(i):
+        return i
 except NameError:
     # Python 3 has no `buffer`; only `memoryview`
     # However, it's faster to just slice the object directly, maybe it keeps a view internally
@@ -26,6 +29,11 @@ def buffer(obj, offset, size=None):
             # return memoryview(obj)[offset:offset+size]
             return obj[offset:offset + size]
     # end buffer reimplementation
+    # smmap can return memory view objects, which can't be compared as buffers/bytes can ... 
+    def to_bytes(i):
+        if isinstance(i, memoryview):
+            return i.tobytes()
+        return i
 
     memoryview = memoryview