X-Git-Url: https://jfr.im/git/z_archive/twitter.git/blobdiff_plain/b29e07db2e9d55e3ed78b80613b868b40d99efd7..42b9cdeeee949f219a76d5cbbade53d4c1a8260c:/twitter/stream.py diff --git a/twitter/stream.py b/twitter/stream.py index 506b873..bf116d5 100644 --- a/twitter/stream.py +++ b/twitter/stream.py @@ -12,43 +12,15 @@ import sys, select, time from .api import TwitterCall, wrap_response, TwitterHTTPError -def recv_chunk_old(sock): # -> bytearray: - """ - Compatible with Python 2.6, but less efficient. - """ - buf = sock.recv(8) # Scan for an up to 16MiB chunk size (0xffffff). - crlf = buf.find(b'\r\n') # Find the HTTP chunk size. - - 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) +PY_27_OR_HIGHER = sys.version_info >= (2, 7) +PY_3_OR_HIGHER = sys.version_info >= (3, 0) - 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(max(0, remaining - end)) - sock.recv(2) # Read the trailing CRLF pair. Throw it away. +Timeout = {'timeout': True} +Hangup = {'hangup': True} - return chunk - - return bytearray() -## recv_chunk_old() +def recv_chunk(sock): # -> bytearray: -def recv_chunk_new(sock): # -> bytearray: - """ - Compatible with Python 2.7+. - """ header = sock.recv(8) # Scan for an up to 16MiB chunk size (0xffffff). crlf = header.find(b'\r\n') # Find the HTTP chunk size. @@ -67,20 +39,38 @@ def recv_chunk_new(sock): # -> bytearray: else: # There is more to read in the chunk. end = len(header) - start chunk[:end] = header[start:] - buffer = memoryview(chunk)[end:] # Create a view into the bytearray to hold the rest of the chunk. - sock.recv_into(buffer) + if PY_27_OR_HIGHER: # 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)) sock.recv(2) # Read the trailing CRLF pair. Throw it away. return chunk return bytearray() -## recv_chunk_new() -if (sys.version_info.major, sys.version_info.minor) >= (2, 7): - recv_chunk = recv_chunk_new -else: - recv_chunk = recv_chunk_old +class Timer(object): + def __init__(self, timeout): + # If timeout is None, we always expire. + self.timeout = timeout + self.reset() + + def reset(self): + self.time = time.time() + + def expired(self): + """ + If expired, reset the timer and return True. + """ + if self.timeout is None: + return True + elif time.time() - self.time > self.timeout: + self.reset() + return True + return False + class TwitterJSONIter(object): @@ -93,37 +83,40 @@ 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 + actually_blocking = self.block and not self.timeout + 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) + sock.setblocking(actually_blocking) buf = '' json_decoder = json.JSONDecoder() - timer = time.time() + timer = Timer(self.timeout) + timeout_token = Timeout if self.timeout else None while True: + buf = buf.lstrip() # Remove any keep-alive delimiters to detect hangups. try: - buf = buf.lstrip() res, ptr = json_decoder.raw_decode(buf) buf = buf[ptr:] + except ValueError: + pass + else: yield wrap_response(res, self.handle.headers) + timer.reset() continue - except ValueError as e: - if self.block and not self.timeout: pass - else: yield None try: - buf = buf.lstrip() # Remove any keep-alive delimiters to detect hangups. if self.timeout and not buf: # This is a non-blocking read. - ready_to_read = select.select([sock], [], [], self.timeout) - if not ready_to_read[0] and time.time() - timer > self.timeout: - yield {'timeout': True} + ready_to_read = select.select([sock], [], [], self.timeout)[0] + if not ready_to_read and timer.expired(): + yield timeout_token continue - timer = time.time() buf += recv_chunk(sock).decode('utf-8') if not buf: - yield {'hangup': True} + yield Hangup break except SSLError as e: # Error from a non-blocking read of an empty buffer. - if (not self.block or self.timeout) and (e.errno == 2): pass + if not actually_blocking and (e.errno == 2): + if timer.expired(): + yield timeout_token else: raise def handle_stream_response(req, uri, arg_data, block, timeout=None): @@ -133,25 +126,12 @@ def handle_stream_response(req, uri, arg_data, block, timeout=None): 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) - -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() @@ -159,28 +139,32 @@ class TwitterStream(TwitterStreamCall): 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 iterator will yield until the TCP connection breaks. When the + connection breaks, the iterator yields `{'hangup': 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}`. This is useful if you want + your program to do other stuff in between waiting for tweets. - 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 `block` parameter sets the stream to be 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),) + uriparts = (str(api_version),) + timeout = float(timeout) if timeout else None - if block: - if timeout: - call_cls = TwitterStreamCallWithTimeout - else: - call_cls = TwitterStreamCall - else: - call_cls = TwitterStreamCallNonBlocking + class TwitterStreamCall(TwitterCall): + def _handle_response(self, req, uri, arg_data, _timeout=None): + return handle_stream_response( + req, uri, arg_data, block=block, timeout=_timeout or timeout) - TwitterStreamCall.__init__( + TwitterCall.__init__( self, auth=auth, format="json", domain=domain, - callable_cls=call_cls, + callable_cls=TwitterStreamCall, secure=secure, uriparts=uriparts, timeout=timeout, gzip=False)