X-Git-Url: https://jfr.im/git/z_archive/twitter.git/blobdiff_plain/57aa6d8ca95479faa0eb03488a501a04f372acf4..42b9cdeeee949f219a76d5cbbade53d4c1a8260c:/twitter/stream.py diff --git a/twitter/stream.py b/twitter/stream.py index 662da04..bf116d5 100644 --- a/twitter/stream.py +++ b/twitter/stream.py @@ -12,107 +12,126 @@ import sys, select, time from .api import TwitterCall, wrap_response, TwitterHTTPError +PY_27_OR_HIGHER = sys.version_info >= (2, 7) +PY_3_OR_HIGHER = sys.version_info >= (3, 0) + +Timeout = {'timeout': True} +Hangup = {'hangup': True} + + +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 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() + + +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): def __init__(self, handle, uri, arg_data, block=True, timeout=None): - self.decoder = json.JSONDecoder() self.handle = handle self.uri = uri self.arg_data = arg_data - self.buf = b"" self.block = block self.timeout = timeout - self.timer = time.time() - - - def recv_chunk(self, sock): - buf = sock.recv(32) - if buf: - # Find the HTTP chunk size. - decoded = buf.decode('utf-8') # Use UTF-8 instead of ASCII because the data contains JSON, a UTF-8 format. - start = decoded.find('\r\n') - if start > 0: - # Decode the chunk size. - remaining = int(decoded[:start], 16) - chunk = buf[start + 2:] # Add the length of the CRLF pair. - remaining -= len(chunk) - - while remaining > 0: - - balance = sock.recv(remaining + 2) # Add the length of the chunk's CRLF pair. - if balance: - chunk += balance - remaining -= len(balance) - # If necessary, remove the trailing CRLF pair. (This precludes an extra trip through the JSON parser.) - return chunk[:-2] if remaining == -2 and chunk[-2] == 0x0d and chunk[-1] == 0x0a else chunk - return b'' def __iter__(self): - if sys.version_info >= (3, 0): - sock = self.handle.fp.raw._sock - else: - sock = 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) - if not self.block or self.timeout: - sock.setblocking(False) + sock.setblocking(actually_blocking) + buf = '' + json_decoder = json.JSONDecoder() + 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: - utf8_buf = self.buf.decode('utf8').lstrip() - res, ptr = self.decoder.raw_decode(utf8_buf) - self.buf = utf8_buf[ptr:].encode('utf8') + res, ptr = json_decoder.raw_decode(buf) + buf = buf[ptr:] + except ValueError: + pass + else: yield wrap_response(res, self.handle.headers) - self.timer = time.time() + timer.reset() continue - except ValueError as e: - if self.block: - pass - else: - yield None - # this is a non-blocking read (ie, it will return if any data is available) try: - if self.timeout: - ready_to_read = select.select([sock], [], [], self.timeout) - if ready_to_read[0]: - self.buf += self.recv_chunk(sock) - if time.time() - self.timer > self.timeout: - yield {"timeout":True} - else: - yield {"timeout":True} - else: - self.buf += self.recv_chunk(sock) + if self.timeout and not buf: # This is a non-blocking read. + ready_to_read = select.select([sock], [], [], self.timeout)[0] + if not ready_to_read and timer.expired(): + yield timeout_token + continue + buf += recv_chunk(sock).decode('utf-8') + if not buf: + yield Hangup + 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 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): - 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): - 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() @@ -120,28 +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)