"Task was destroyed but is pending!" error in tests

[benjamin-hodgson]

Some - not all - tests print Task was destroyed but is pending! for some reason.

This doesn't seem to happen during the integration tests, which suggests that it's a problem with the fixture in the unit tests not tearing itself down properly.

$ run-contexts -v test/connection_tests.py:WhenRespondingToConnectionStart
When responding to connection start
  PASS: it should send start ok
----------------------------------------------------------------------
PASSED!
1 context, 1 assertion
(0.0 seconds)
Task was destroyed but it is pending!
task: <Task pending coro=<read_next() running at /Users/ben/Developer/asynqp/src/asynqp/bases.py:48>>
Task was destroyed but it is pending!
task: <Task pending coro=<open_connection() running at /Users/ben/Developer/asynqp/src/asynqp/connection.py:97> wait_for=<Future cancelled>>

This might have been due to a Python update; previous Travis builds don't exhibit this error, but the same code running on my box does.

[benjamin-hodgson]

See this commit to the Python repo which introduced the error messaging. I think this change made it into Python 3.4.2 but not 3.4.1. I guess Travis updated its build environment in the last week or so.

In any case, I think this error is either a) spurious, or b) surfacing a long-standing problem with asynqp. If b) it needs to be fixed, obviously; if a) then the error should be silenced for the tests.

[benjamin-hodgson]

Temporarily fixed in commit 0159e23. Leaving this issue open because the hack needs to be resolved.

[renatomassaro]

I was having the same issue using a code very similar to the repository example, but making use of consume instead of queue.get(). I thought this could be the reason but the same error happened with queue.get(), so the example on the README has this issue on python==3.4.2.

Closing the channel before closing the connection fixed this for me. I'll submit a PR with the README change.