X-Git-Url: https://jfr.im/git/z_archive/twitter.git/blobdiff_plain/28a8ef602a9a4f91a400f4d23917c4d35aaa8aa3..99407dab6c2455664a46e0c171972cfd7b61f43b:/twitter/stream.py diff --git a/twitter/stream.py b/twitter/stream.py index b9035f0..101d31b 100644 --- a/twitter/stream.py +++ b/twitter/stream.py @@ -14,26 +14,31 @@ from .api import TwitterCall, wrap_response, TwitterHTTPError def recv_chunk(sock): # -> bytearray: - timeout = sock.gettimeout(); sock.setblocking(True) # Read the whole HTTP chunk. - 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. + buf = sock.recv(8) # Scan for an up to 16MiB chunk size (0xffffff). + crlf = buf.find(b'\r\n') # Find the HTTP chunk size. - start = crlf + 2 # Add in the length of the header's CRLF pair. - end = len(buf) - start + if crlf > 0: # If there is a length, then process it + remaining = int(buf[:crlf], 16) # Decode the chunk size. + + start = crlf + 2 # Add in the length of the header's CRLF pair. + end = len(buf) - start + + chunk = bytearray(remaining) + + if remaining <= 2: # E.g. an HTTP chunk with just a keep-alive delimiter or end of stream (0). + chunk[:remaining] = buf[start:start + remaining] + # There are several edge cases (remaining == [3-6]) 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. chunk[:end] = buf[start:] chunk[end:] = sock.recv(remaining - end) - sock.recv(2) # Read the trailing CRLF pair. Throw it away. - sock.settimeout(timeout) - return chunk + return chunk - sock.settimeout(timeout) return bytearray() ## recv_chunk() @@ -52,8 +57,8 @@ class TwitterJSONIter(object): def __iter__(self): sock = self.handle.fp.raw._sock if sys.version_info >= (3, 0) else self.handle.fp._sock.fp._sock sock.setsockopt(socket.SOL_SOCKET, socket.SO_KEEPALIVE, 1) - sock.setblocking(self.block and not self.timeout) # not (not self.block or self.timeout) - buf = u'' + sock.setblocking(self.block and not self.timeout) + buf = '' json_decoder = json.JSONDecoder() timer = time.time() while True: @@ -65,35 +70,32 @@ class TwitterJSONIter(object): timer = time.time() continue except ValueError as e: - if self.block: - pass - else: - yield None + if self.block: pass + else: yield None try: - if self.timeout: # this is a non-blocking read (ie, it will return if any data is available) - + buf = buf.lstrip() # Remove any keep-alive delimiters to detect hangups. + if self.timeout: ready_to_read = select.select([sock], [], [], self.timeout) if ready_to_read[0]: - buf += recv_chunk(sock).decode('utf-8') + buf += recv_chunk(sock).decode('utf-8') # This is a non-blocking read. if time.time() - timer > self.timeout: yield {'timeout': True} - else: - yield {'timeout': True} + else: yield {'timeout': True} else: buf += recv_chunk(sock).decode('utf-8') if not buf and self.block: yield {'hangup': True} + break except SSLError as e: - if (not self.block or self.timeout) and (e.errno == 2): - # Apparently this means there was nothing in the socket buf - pass - else: - raise - except urllib_error.HTTPError as e: - raise TwitterHTTPError(e, self.uri, 'json', self.arg_data) + # Error from a non-blocking read of an empty buffer. + if (not self.block or self.timeout) and (e.errno == 2): pass + else: raise def handle_stream_response(req, uri, arg_data, block, timeout=None): - handle = urllib_request.urlopen(req,) + try: + handle = urllib_request.urlopen(req,) + except urllib_error.HTTPError as e: + raise TwitterHTTPError(e, uri, 'json', arg_data) return iter(TwitterJSONIter(handle, uri, arg_data, block, timeout=timeout)) class TwitterStreamCallWithTimeout(TwitterCall):