Make zope a semi-optional test dependency
#685
Conversation
|
Would you mind inverting the logic and just make the Zope tests conditional on the presence of Zope.interface? That seems to be the more common approach nowadays anyways. |
Yeah, that's easy enough — the main difference between that approach and this one is that with this approach, it would be very hard to accidentally disable the zope-related tests, whereas if you always disable the zope-related testing whenever I figured this is a niche enough thing to want that it's better to require enthusiastic opt-in, but maybe it's not that big a deal if these tests sometimes get accidentally disabled in downstream testers? |
|
I don’t think anybody cares at all TBH. We’re just dragging it along for backward-compat. |
For the admittedly small number of people who don't care about support for `zope.interface`, it is unnecessary extra work to support even a test-only dependency on `zope`. The `zope`-specific tests are relatively contained, however, and with judicious use of pytest fixtures, we can make it so that `zope` is not a required dependency when skipping tests marked with `pytest.mark.zope`. This allows users to test everything but the `zope`-specific tests with no `zope` dependency with: $ TOX_AP_TEST_EXTRAS="tests_no_zope" tox -e <env> -- -m 'not zope'
This is a little more error-prone, but it's much more convenient to invoke, since it doesn't involve skipping the "zope" marker.
pganssle
force-pushed
the
zope_optional
branch
from
573e58d
to
5f696a0
Compare
|
OK, I've changed it to an auto-skipping fixture and added some docstrings and a changelog. If you think there's somewhere else that this should be explicitly documented, let me know. It didn't seem like it would fit much into |
|
Thanks! |
For the admittedly small number of people who don't care about support for
zope.interface, it is unnecessary extra work to support even a test-only dependency onzope.The
zope-specific tests are relatively contained, however, and with judicious use of pytest fixtures, we can make it so thatzopeis not a required dependency when skipping tests marked withpytest.mark.zope.This allows users to test everything but the
zope-specific tests with nozopedependency with:I realize that this is unusual and maybe not a maintenance burden you want to carry upstream, but it would smooth the process of importing
attrsupstream into a certain large company's monorepo... I'm fine with a response that's basically "Enough with your weird nonsense, Paul".I didn't bother with documentation or a news fragment yet because I figured there's a high likelihood that this wouldn't be accepted.
Pull Request Check List
changelog.d.