This reverts commit 75a6c4b70c5873a14ae9fef1b94647fd8dfef733.

py.test is buggy and can't do it.

Whew.

updated to work with current RabbitMQ

getting there

The test system itself works, now for detailed bug fixing

and a bunch of syntax+import errors

TODO: rewrite AmqpProtocol as a context handler

instead of @asyncio.coroutine / yield from

Protocol.reader in Python3.5 is set to None by the connection_lost
method. Occasionally, this l...

To make sure all Tasks are properly cleaned up.

ATM, the rest of the code relies on connection_lost being called after
recieving EOF so let's ke...

CPython 3.6 adds a mean error message if Tasks are garbage collected
when they're still running,...

Refs #138

Due to a not-so-minor change in rabbit 3.6.7, rabbit's statistics
collection and propagation now...

As of CPython 3.6.1, drain() is not coroutine-safe so put a lock around it
to make sure it only ...

Will be needed in the next commit because write_frame will become a
coroutine.

Except the 2 publish methods, which call drain() explicitly after writing
the 3+ (method, header...

There is a danger: if different tasks try to use methods from the same
channel at the same time,...

With the previous commits, there's only one argument allowed by
_write_frame. So use it explici...

Apart from open() where the code is *exactly* the same, the other 3
methods implemented a sub-pa...

It's used in all user-facing methods, but it's silently ignored by
_write_frame()... so just dro...

Last user removed in previous commit.

AmqpProtocol.close() was quite confusing due to its use of frame.payload
as an argument to AmqpE...

For some reason, this test fails more reliably on python 3.6 than on 3.4.
In any case, the test ...

"BSD LICENSE" is nowhere near precise enough to be a proper license. So
go with the 3-clause ve...

Right now, only ~2 states (OPEN and CLOSED) are used. CLOSING will be
added later on.

Note: CO...

As per https://www.rabbitmq.com/resources/specs/amqp-xml-doc0-9-1.pdf

1.4.2.9. Method conne...

This shamelessly taken almost verbatim from websocket's protocol.py.

Just stop looking at the management API and get on with the consumption.

The core change is to call _stream_writer.close() both on a
server-initiated close and a client-...

The old test didn't actually cause a cancel, which would have caused a
dispatch error.

For some reason, calling cancel() on TimerHandle objects returned by
call_later() doesn't really...