CI is not running for PRs

[fabaff]

The CI is not running on PRs. It seems that the credentials are not valid which are used by travis.

        if raise_exc and self._error:
>           raise self._error
E           gmqtt.mqtt.handler.MQTTConnectError: code 134 (Connection Refused: Bad User Name or Password)

[Lenka42]

@fabaff Actually this is expected behaviour (or I don't know how to handle this)
Our tests create connection to Flespi MQTT broker and it requires token taken from environment variable. And in general third-party PR can have any code, which can potentially cause token leak.
So the flow is following: if we have PR which looks good, we merge it to the feature-branch first and afterwards, if tests are OK, we merge it to master.

[Lenka42]

@fabaff any ideas about how we can do better?

[fabaff]

Perhaps something like this could be helpful. If TOKEN is present then run the tests with flespi.io otherwise use a local MQTT broker.

if os.getenv('TOKEN'):
    host = 'mqtt.flespi.io'
    username = os.getenv("TOKEN")
    password = None
    port = 1883
else:
    host = os.getenv("HOST", "127.0.0.1")
    username = os.getenv("USERNAME", "")
    password = os.getenv("PASSWORD", None)
    port = os.getenv("PORT", 1883)

Not very elegant but would allow to run the tests locally before submitting a pull request. Perhaps it would require to add @pytest.mark.local to some tests to distinguish between a local MQTT broker and the one from flespi.io as a local broker usually doesn't support SSL/TLS in a default setup.