Remove returning stored site user

[Zharktas]

Site user is lost when validation fail during package_create.

This might fix #151.

[metaodi]

To fix the build add the following lines to .travis.yml:

# the new trusty images of Travis cause build errors with psycopg2, see https://github.com/travis-ci/travis-ci/issues/8897
dist: trusty
group: deprecated-2017Q4

[metaodi]

@Zharktas what do you mean with "site user is lost"? Lost how? The method only returns the username and not a user object, and the username won't change. Or am I missing something?

[Zharktas]

If the site user has been previously fetched, https://github.com/vrk-kpa/ckanext-harvest/blob/f5df983542781b69728e1f871f4e7fd5baa86bea/ckanext/harvest/harvesters/base.py#L192 this never executes as the.

And as the validatation error happens here https://github.com/vrk-kpa/ckanext-harvest/blob/f5df983542781b69728e1f871f4e7fd5baa86bea/ckanext/harvest/harvesters/base.py#L356, the SQLalchemy detaches itself.

[metaodi]

Okay just to get it right, if the user has been previously fetched (i.e. self._user_name is set) then the following line is not executed:

self._site_user = p.toolkit.get_action('get_site_user')(context, {})

So far, everything is clear.

What I don't understand is: why is it a problem if that line is not executed? Do we need self._site_user somewhere? In most cases _get_user_name() will just return harvest (btw: the above line is also not executed if there is a harvest user on the system).
My guess by looking at the source of get_site_user() is, that the model.Session.flush() plays a role, but I'm not sure.

I don't think, that it's a good idea to call get_site_user() for every dataset in the harvester, so to save the username in a variable seems like a good idea.

[Zharktas]

I'll test tomorrow if reorganizing the code will help the issue. Either validate the dataset before any SQL magic happens or catch validation errors properly.

[Zharktas]

Closed in favor of #312