[pytest-dev] pytest 5.0.0 and pluginmanager

[Bruno Oliveira]

On Mon, Jul 1, 2019 at 9:21 AM Chris Dent <cdent at anticdent.org> wrote:

> On Mon, 1 Jul 2019, Bruno Oliveira wrote:
> > On Mon, Jul 1, 2019 at 8:17 AM Chris Dent <cdent at anticdent.org> wrote:
> > Is Gabbi + pytest a plugin? If so, you could store the config object
> during
> > pytest_configure to reuse later:
>
> Thanks for the quick response.
>
> It's sort of a plugin. It uses plugin mechanisms to manage the
> collection process and manage fixtures that are scoped over a
> sequence of tests (instead of just one test).
>
> The issue is that in the deployed tests already using gabbi,
> there is no conftest.py nor python test file that is fully
> "pytest aware" from which to load the plugin. Instead the test
> file calls a function which itself loads pytest and then loads
> the plugin:
>
>      import pytest
>      pluginmanager = pytest.config.pluginmanager
>      pluginmanager.import_plugin('gabbi.pytester')
>

Hmmm strange, because for "pytest.config" to be available, it means the
user is running the "pytest" command. You can see here how pytest used to
do it:

https://github.com/pytest-dev/pytest/blob/d3549df5b94137453b832345a9b8074f518731b0/src/_pytest/main.py#L191-L192

(It is inserting a wrapper object to issue a warning, but other than that
the mechanism is there)

So in theory your plugin is running under pytest, otherwise `pytest.config`
would raise an AttributeError.

Messing about just now I see that I've got access to a metafunc and
> it has a `config` so this seems to be working:
>
>      if metafunc:
>          pluginmanager = metafunc.config.pluginmanager
>      else:
>          import pytest
>          pluginmanager = pytest.config.pluginmanager
>
> (There are older calling scenarios where metafunc will be None.)
>
> Is this a suitable workaround or am I opening myself up for future
> trouble by doing this?
>

We don't have any plans to deprecate `metafunc.config`, as doing so would
break *tons* of code, and is really part of the API and no great downsides
(as opposed to the pytest.config which was a nasty global).

If you can't make the "pytest_configure" suggestion work, this seems OK to
me.

Cheers,
Bruno
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.python.org/pipermail/pytest-dev/attachments/20190701/2d41498b/attachment-0001.html>