GitLab

(see: https://dev.twitter.com/blog/introducing-new-metadata-for-tweets)

(https://github.com/tweepy/tweepy/issues/208)
 merged from origin/http_dumper branch...

1) read newlines until a digit appears
2) read rest of digits until another new line appears
3) read next status object
4) repeat

-improved speed over trunk version
-simplify loop logic
-added condition for "self.running and not resp.isclosed()" during each read-loop stage 

Updated faster read loop code.  this implementation is nearly as fast and also allows all tweets to fire on_status() event calls.  Previously, when the stream went silent, the most recent tweet would be stuck in the read buffer and would not fire an on_status() event call.

Use a buffer rather than reading the Twitter response one byte at a time. This actually respects the buffer_size option that was already there, but unused.

https://dev.twitter.com/blog/streaming-api-turning-ssl-only-september-29th

The "raise exception" is overriding the stack trace and hiding all errors, a simple "raise" should be fine.

_read_loop really didn't seem right, but somehow kinda worked. It now works when pointed at user streams too.

searches, where Twitter will take longer than five seconds even to
send a response header.

My preferred default would be no timeout at all, but if we are to
have one then five minutes seems more reasonable than five seconds.

when using OAuth (because the nonce has been used before, and the timestamp
is no longer correct).

Improve the options handling in the streaming.Stream class, so that false values can be used. (In particular I want to be able to pass timeout=None, but one might also reasonably want to pass zero for some of the other options.)

Note: this may break some code that just passes this options
by position rather than keyword.

It should now be possible to authenticate with
the streaming API by either using BasicAuthHandler or
OAuthHandler.