-def recv_chunk(sock):
- buf = sock.recv(10) # Scan for an up to a 4GiB chunk size (0xffffffff).
- if buf:
- crlf = buf.find(b'\r\n') # Find the HTTP chunk size.
- if crlf > 0:
- remaining = int(buf[:crlf], 16) # Decode the chunk size.
- chunk = bytearray(remaining) # Create the chunk buffer.
-
- start = crlf + 2 # Add in the length of the header's CRLF pair.
- end = len(buf) - start
-
- chunk[:end] = buf[start:]
- chunk[end:] = sock.recv(remaining - end)
-
+python27_3 = sys.version_info >= (2, 7)
+def recv_chunk(sock): # -> bytearray:
+
+ header = sock.recv(8) # Scan for an up to 16MiB chunk size (0xffffff).
+ crlf = header.find(b'\r\n') # Find the HTTP chunk size.
+
+ if crlf > 0: # If there is a length, then process it
+
+ size = int(header[:crlf], 16) # Decode the chunk size. Rarely exceeds 8KiB.
+ chunk = bytearray(size)
+ start = crlf + 2 # Add in the length of the header's CRLF pair.
+
+ if size <= 3: # E.g. an HTTP chunk with just a keep-alive delimiter or end of stream (0).
+ chunk[:size] = header[start:start + size]
+ # There are several edge cases (size == [4-6]) as the chunk size exceeds the length
+ # of the initial read of 8 bytes. With Twitter, these do not, in practice, occur. The
+ # shortest JSON message starts with '{"limit":{'. Hence, it exceeds in size the edge cases
+ # and eliminates the need to address them.
+ else: # There is more to read in the chunk.
+ end = len(header) - start
+ chunk[:end] = header[start:]
+ if python27_3: # When possible, use less memory by reading directly into the buffer.
+ buffer = memoryview(chunk)[end:] # Create a view into the bytearray to hold the rest of the chunk.
+ sock.recv_into(buffer)
+ else: # less efficient for python2.6 compatibility
+ chunk[end:] = sock.recv(max(0, size - end))