Exaproxy

* Fix: handle blocking exception when writing on messagebox
reported by Tim Watts
* Fix: a few issues reported by pycharm
* Fix: honour configuration directing the proxy to use the source IP address
provided over the proxy protocol
* Fix: error sending requests to the backend decider processes when running as
an ICAP server
* Fix: error enumerating network interfaces that do not have IP addresses
reported with fix by asthomas
* Fix: do not block waiting for a response from the redirector process since we can
end up with a deadlock
reported by Emanuele Rocca
* Improvement: free memory that could be held by the client coroutine during long lived connections

* Fix: issue with handling of max-forwards
patch by Tim Watts
* Fix: gracefully shutdown when trying to read over a closed pipe for communication
with another process
reported by Emanuele Rocca
* Fix: issue 9, pip installation

* Feature: ICAP is the new black
* Feature: Separate process to manage workers
* Improvement: Smaller, faster core reactor ( only moving data around )
* Improvement: better logging
* Improvement: internal APC API ( should you care :p )
* Improvement: automatically pick the best reactor by default
* Improvement: kqueue reactor
patch by: Marek Obuchowicz
* Improvement: exaproxy.http.mask to remove the presence of any previous proxy
patch by: Emanuele Rocca
* Improvement: Forked children managed by a dedicated process (as Squid wants to do)
* Change: do not fail back to select if the request reactor is not available
* Fix: client tracking code was reporting bad information
* Fix: on exit try to flush the last logs
* Fix: badly tracking processes
patch by: Tim Watts
* Fix: references to closed sockets
* Fix: correctly reports version on help
* Fix: duplicate output
* Fix: And many more David did not record tsss tsss ...

* Improvement: added the opened connection where the client never spoke to the graphs
* Improvement: can timeout new HTTP connection which does not send data (new connections idle timeout)
* Improvement: let pass invalid pre-RFC 2616 syntaxes
* Improvement: record number of requests seen
* Backward compatibility: changed the name of the clients. and servers. keys
* Fix: do not assume that our synchronous reads return all the data we ask for
* Improvement: add required X-Forwarded-For header
* Improvement/Fix: fix scalability issue for large number of open connections ( over 30k )

* Fix: new requests on an existing connection would destroy the buffer and cause us to wake up at every loop.
* Fix: correctly report our verion on the webpage we serve
* Fix: do not fail if python was compiled without IPv6 support
* Improvement: allow to unbind and rebind like haproxy
* Improvement: implement SIGQUIT as a way to quit once all current connections are closed
* Improvement: allow to choose an IP for outgoing connections, permit over 2^16/2 client connections
* Improvement: allow to control^Wbreak the proxy using some python code on the web page
* Improvement: prevent connection to the local host to prevent firewall bypass
* Improvement: allow to stop / restart listening on port and soft-stop (stop when all connections close)
* Improvement: allow to specify exclusions for the block on localhost
* Improvement: limit connect methods to only port 443, 981 and 7000 as default (user option)
* Improvement: different counters for ipv4 and ipv6
* Improvement: new web interface without javascript, with more options (debug, ...)
* Improvement: better monitoring (60 seconds and 60 minutes graphs)
* Backward compatibility: exaproxy.daemon.connections was renamed to exaproxy.http.connections
* Backward compatibility: running.transfer keys have been renamed
* Backward compatibility: all json "running." keys have been renamed
* Backward compatibility: renamed many web page (URL should never be scraped)

* Many bug fixes including (but not exclusively)
* Fix: make sure the program always have data to display in the web interface
* Fix: dynamically adding/removing processes using SIGUSR1/SIGUSR2 now works
* Fix: bug with SIGTRAP
* Fix: prevent harmful options with select reactor
* Fix: logging values are different between logging and syslog, was causing issues
* Fix: a forever loop on chunk decoding
* Fix: actually seperate usage logging from messages generated by the proxy
* Fix: return 400 to HTTP/1.0 request with no header lines
* Fix: bug handling timeout during connection establishment
* Fix: check that hostnames do not have parts between dots that are more than 255 characters long
* Fix: ensure we do not read a request immediately after having read a previous request on the same connection
* Fix: full rewrite of the chunk parser
* Improvement: do not attempt to resolve hostnames that do not contain a dot
* Improvement: changed the worker thread queue to be non blocking
* Improvement: change exaproxy.redirector.minimum and exaproxy.redirector.maximum using the web interface
* Improvement: allow the usage of port for syslog usage record
* Improvement: better management of the worker creation and deletion
* Improvement: do not perform DNS lookup if the hostname does not have a . (search in the URL bar)
* Improvement: prevent memory exhaustion
* Many more changes ...