py-amqp Added hypothesis property based testing

This PR uses hypothesis to check if arrays & dictionaries are serialized and deserialized correctly.

I also fixed our integer serialization and deserialization code, bytes string serialization and small string serialization on tables and arrays.

Mar 09 '19 09:03 thedrow

Yes, the examples directory should be ignored when reviewing this.

Mar 10 '19 07:03 thedrow

Seems like we have a genuine problem with strings inside dictionaries on Python 2. Even without any adjustments that were made in this branch I can reproduce this:

Falsifying example: test_table(self=<t.unit.test_serialization.test_serialization instance at 0x7fdf17cde518>, table={u'': u'\x80'})
Traceback (most recent call last):
  File "/home/omer/Documents/Projects/py-amqp/t/unit/test_serialization.py", line 126, in test_table
    assert loads(b'F', dumps(b'F', [table]))[0][0] == table
AssertionError: assert {'': '\xc2\x80'} == {'': '\x80'}
  Differing items:
  {'': '\xc2\x80'} != {'': '\x80'}
  Full diff:
  - {'': '\xc2\x80'}
  ?       ----
  + {u'': u'\x80'}
  ?  +    +

Mar 12 '19 16:03 thedrow

Since '\x80' is a unicode string when you encode it to bytes, you get \xc2\x80 but we aren't decoding back. We're encoding again.

Not sure how to resolve this issue.

Mar 12 '19 16:03 thedrow

Authentication fails with the addition of support for small strings in tables and arrays. I'm not sure why though.

Also, tox-docker doesn't seem to work anymore which is super strange.

Mar 13 '19 14:03 thedrow

Seems like there's a bug in RabbitMQ with short strings in tables. It is allowed in the spec.

Mar 13 '19 15:03 thedrow

I found the source of the problem: When RabbitMQ authenticates using the AMQPLAIN method it expects a longstr instead of a shortstr. See issue https://github.com/rabbitmq/rabbitmq-server/issues/1914.

Mar 14 '19 09:03 thedrow