try:
import urllib.request as urllib_request
import urllib.error as urllib_error
- import io
except ImportError:
import urllib2 as urllib_request
import urllib2 as urllib_error
+import io
import json
from ssl import SSLError
import socket
from .api import TwitterCall, wrap_response, TwitterHTTPError
-def recv_chunk(sock): # -> bytearray:
+PY_3_OR_HIGHER = sys.version_info >= (3, 0)
- buf = sock.recv(8) # Scan for an up to 16MiB chunk size (0xffffff).
- crlf = buf.find(b'\r\n') # Find the HTTP chunk size.
+CRLF = b'\r\n'
- if crlf > 0: # Check because non-blocking calls can return empty data.
+Timeout = {'timeout': True}
+Hangup = {'hangup': True}
+HeartbeatTimeout = {'heartbeat_timeout': True, 'hangup': True}
- remaining = int(buf[:crlf], 16) # Decode the chunk size.
+class ChunkDecodeError(Exception):
+ pass
- start = crlf + 2 # Add in the length of the header's CRLF pair.
- end = len(buf) - start
+class EndOfStream(Exception):
+ pass
- chunk = bytearray(remaining)
- chunk[:end] = buf[start:]
- chunk[end:] = sock.recv(remaining - end)
+range = range if PY_3_OR_HIGHER else xrange
- sock.recv(2) # Read the trailing CRLF pair. Throw it away.
-
- return chunk
+class SocketShim(io.IOBase):
+ """
+ Adapts a raw socket to fit the IO protocol.
+ """
+ def __init__(self, sock):
+ self.sock = sock
+ def readable(self):
+ return True
+ def read(self, size):
+ return self.sock.read(size)
+ def readinto(self, buf):
+ return self.sock.recv_into(buf)
+
+def recv_chunk(reader): # -> bytearray:
+ for headerlen in range(12):
+ header = reader.peek(headerlen)[:headerlen]
+ if header.endswith(CRLF):
+ break
+ else:
+ raise ChunkDecodeError()
+
+ size = int(header, 16) # Decode the chunk size
+ reader.read(headerlen) # Ditch the header
+
+ if size == 0:
+ raise EndOfStream()
+
+ chunk = bytearray()
+ while len(chunk) < size:
+ remainder = size - len(chunk)
+ chunk.extend(reader.read(remainder))
+
+ reader.read(2) # Ditch remaining CRLF
+
+ return chunk
+
+
+class Timer(object):
+ def __init__(self, timeout):
+ # If timeout is None, we never expire.
+ self.timeout = timeout
+ self.reset()
- return bytearray()
+ def reset(self):
+ self.time = time.time()
-## recv_chunk()
+ def expired(self):
+ """
+ If expired, reset the timer and return True.
+ """
+ if self.timeout is None:
+ return False
+ elif time.time() - self.time > self.timeout:
+ self.reset()
+ return True
+ return False
class TwitterJSONIter(object):
- def __init__(self, handle, uri, arg_data, block=True, timeout=None):
+ def __init__(self, handle, uri, arg_data, block, timeout, heartbeat_timeout):
self.handle = handle
self.uri = uri
self.arg_data = arg_data
self.block = block
- self.timeout = timeout
+ self.timeout = float(timeout) if timeout else None
+ self.heartbeat_timeout = float(heartbeat_timeout) if heartbeat_timeout else None
def __iter__(self):
- sock = self.handle.fp.raw._sock if sys.version_info >= (3, 0) else self.handle.fp._sock.fp._sock
+ actually_block = self.block and not self.timeout
+ sock_timeout = min(self.timeout or 1000000, self.heartbeat_timeout) if actually_block else None
+ sock = self.handle.fp.raw._sock if PY_3_OR_HIGHER 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''
- json_decoder = json.JSONDecoder()
- timer = time.time()
+ sock.setblocking(actually_block)
+ reader = io.BufferedReader(SocketShim(sock))
+ buf = ''
+ raw_decode = json.JSONDecoder().raw_decode
+ timer = Timer(self.timeout)
+ heartbeat_timer = Timer(self.heartbeat_timeout)
while True:
+ buf = buf.lstrip() # Remove any keep-alive delimiters
try:
- buf = buf.lstrip()
- res, ptr = json_decoder.raw_decode(buf)
+ res, ptr = raw_decode(buf)
buf = buf[ptr:]
+ except ValueError:
+ if not self.block and not self.timeout:
+ yield None
+ else:
yield wrap_response(res, self.handle.headers)
- timer = time.time()
+ timer.reset()
+ heartbeat_timer.reset()
continue
- except ValueError as e:
- if self.block:
- pass
- else:
- yield None
+
+ if heartbeat_timer.expired():
+ yield HeartbeatTimeout
+ break
+ if timer.expired():
+ yield Timeout
+
try:
- 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') # This is a non-blocking read.
- if time.time() - timer > self.timeout:
- yield {'timeout': True}
- else:
- yield {'timeout': True}
- else:
- buf += recv_chunk(sock).decode('utf-8')
- if not buf and self.block:
- yield {'hangup': True}
+ if sock_timeout:
+ ready_to_read = select.select([sock], [], [], sock_timeout)[0]
+ if not ready_to_read:
+ continue
+ received = recv_chunk(reader)
+ buf += received.decode('utf-8')
+ if received:
+ heartbeat_timer.reset()
+ except (ChunkDecodeError, EndOfStream):
+ yield Hangup
+ break
except SSLError as e:
- if (not self.block or self.timeout) and (e.errno == 2): pass # Empty buffer during polling.
- else: raise
+ # Code 2 is error from a non-blocking read of an empty buffer.
+ if e.errno != 2:
+ raise
-def handle_stream_response(req, uri, arg_data, block, timeout=None):
+def handle_stream_response(req, uri, arg_data, block, timeout, heartbeat_timeout):
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):
- def _handle_response(self, req, uri, arg_data, _timeout=None):
- return handle_stream_response(req, uri, arg_data, block=True, timeout=self.timeout)
-
-class TwitterStreamCall(TwitterCall):
- def _handle_response(self, req, uri, arg_data, _timeout=None):
- return handle_stream_response(req, uri, arg_data, block=True)
+ return iter(TwitterJSONIter(handle, uri, arg_data, block, timeout, heartbeat_timeout))
-class TwitterStreamCallNonBlocking(TwitterCall):
- def _handle_response(self, req, uri, arg_data, _timeout=None):
- return handle_stream_response(req, uri, arg_data, block=False)
-
-class TwitterStream(TwitterStreamCall):
+class TwitterStream(TwitterCall):
"""
- The TwitterStream object is an interface to the Twitter Stream API
- (stream.twitter.com). This can be used pretty much the same as the
- Twitter class except the result of calling a method will be an
- iterator that yields objects decoded from the stream. For
- example::
+ The TwitterStream object is an interface to the Twitter Stream
+ API. This can be used pretty much the same as the Twitter class
+ except the result of calling a method will be an iterator that
+ yields objects decoded from the stream. For example::
twitter_stream = TwitterStream(auth=OAuth(...))
iterator = twitter_stream.statuses.sample()
for tweet in iterator:
...do something with this tweet...
- The iterator will yield tweets forever and ever (until the stream
- breaks at which point it raises a TwitterHTTPError.)
-
- The `block` parameter controls if the stream is blocking. Default
- is blocking (True). When set to False, the iterator will
- occasionally yield None when there is no available message.
+ The iterator will yield until the TCP connection breaks. When the
+ connection breaks, the iterator yields `{'hangup': True}`, and
+ raises `StopIteration` if iterated again.
+
+ Similarly, if the stream does not produce heartbeats for more than
+ 90 seconds, the iterator yields `{'hangup': True,
+ 'heartbeat_timeout': True}`, and raises `StopIteration` if
+ iterated again.
+
+ The `timeout` parameter controls the maximum time between
+ yields. If it is nonzero, then the iterator will yield either
+ stream data or `{'timeout': True}` within the timeout period. This
+ is useful if you want your program to do other stuff in between
+ waiting for tweets.
+
+ The `block` parameter sets the stream to be fully non-blocking. In
+ this mode, the iterator always yields immediately. It returns
+ stream data, or `None`. Note that `timeout` supercedes this
+ argument, so it should also be set `None` to use this mode.
"""
- def __init__(
- self, domain="stream.twitter.com", secure=True, auth=None,
- api_version='1.1', block=True, timeout=None):
- uriparts = ()
- uriparts += (str(api_version),)
-
- if block:
- if timeout:
- call_cls = TwitterStreamCallWithTimeout
- else:
- call_cls = TwitterStreamCall
- else:
- call_cls = TwitterStreamCallNonBlocking
-
- TwitterStreamCall.__init__(
+ def __init__(self, domain="stream.twitter.com", secure=True, auth=None,
+ api_version='1.1', block=True, timeout=None,
+ heartbeat_timeout=90.0):
+ uriparts = (str(api_version),)
+
+ class TwitterStreamCall(TwitterCall):
+ def _handle_response(self, req, uri, arg_data, _timeout=None):
+ return handle_stream_response(
+ req, uri, arg_data, block,
+ _timeout or timeout, heartbeat_timeout)
+
+ TwitterCall.__init__(
self, auth=auth, format="json", domain=domain,
- callable_cls=call_cls,
+ callable_cls=TwitterStreamCall,
secure=secure, uriparts=uriparts, timeout=timeout, gzip=False)