If we don't want timeouts we must set the socket's timeout to None

[thedrow]

See https://docs.python.org/3/library/socket.html#socket.socket.settimeout.

This commit was triggered by celery/celery#4876 (comment).
It could fix the issue as well.

[thedrow]

With this change, the integration tests hang on Travis.

@matusvalo Any idea why?

[matusvalo]

@thedrow, see my comments in the patch. The problem is that when you are using publisher confirm, the publisher checks before publishing if broker sent a message to him. This relies on timeout=0 which this patch has changed semantics to block forever.

[thedrow]

So what should we do to fix this issue exactly?

[matusvalo]

What would change on PR is:

1. remove internal interfaces changes which breaks current code - having_timeout() method.
2. Fix all timeouts in public interfaces...

py-amqp/amqp/channel.py

Lines 1686 to 1688 in b36bf88

	def _basic_publish(self, msg, exchange='', routing_key='',
	mandatory=False, immediate=False, timeout=None,
	argsig='Bssbb'):

py-amqp/amqp/connection.py

Line 507 in b36bf88

def blocking_read(self, timeout=None):

In this way it should work. Maybe we can add also integration tests for each timeout input?

[matusvalo]

This line of code breaks the following:

py-amqp/amqp/channel.py

Lines 1764 to 1769 in b36bf88

	if capabilities.get('connection.blocked', False):
	try:
	# Check if an event was sent, such as the out of memory message
	self.connection.drain_events(timeout=0)
	except socket.timeout:
	pass

[matusvalo]

The def having_timeout() method should be not changed because:

1. it is internal API
2. we are using it as mentioned above.