Make normal fixtures work with "yield"

[nicoddemus]

Did a quick experiment to see how hard would it be to implement #1461 to suggest working on it during the sprint... astonished to find out it was way too easy. 😲

Just ignore the yieldctx parameter and decide on the spot if we are a "yield fixture" based on whether or not the function is a generator does the trick nicely.

Almost all of the changes are related only to change the tests of yield_fixture to also test with plain fixture to prove this is working.

Missing:

• Clean up the code (remove the need to pass yieldctx around);
• Update docs and CHANGELOG;

Opening the PR to discuss if we should move this forward or not. @RonnyPfannschmidt, @hpk42, @The-Compiler what you guys think?

[coveralls]

Coverage decreased (-0.02%) to 92.407% when pulling a0bbad6 on nicoddemus:issue-1461-merge-yield-fixture into b5bd4d9 on pytest-dev:features.

[nicoddemus]

(quick note: AppVeyor is failing because of the "doctesting" env, which is already fixed in master)

[RonnyPfannschmidt]

Looks neat, surprisingly simple, just wondering if this is to be considered a breaking api change art semver

[The-Compiler]

So I know this is a WIP, just adding this so we don't accidentally forget: Those lines aren't needed anymore now 😉

EDIT: Or should be changed to work again, as I pointed out in #1586 (comment)

[The-Compiler]

Awesome! 👍

@RonnyPfannschmidt What would be the breaking change? I can only imagine two things:

• A yield_fixture without a yield would now work instead of error out. We can change that to the old behaviour.
• A fixture with a yield would now work instead of erroring the test. I don't think that's a backwards-incompatible change, as hopefully nobody would keep tests around which fail in that way 😉

[RonnyPfannschmidt]

@The-Compiler true

[nicoddemus]

Nice, I will continue to work on the PR then (clean it up and docs, mostly).

About the documentation, I plan to consider @fixture + yield as the recommended way of writing teardown fixture code and update all examples to use @fixture and yield where appropriate (those which don't have tear down code will be kept using return). So yield_fixture will be a footnote in the docs mentioned only as a historical fact. How do you guys feel about that?

A yield_fixture without a yield would now work instead of error out. We can change that to the old behaviour.

I see. We just have to consider that to keep this behavior we need to keep yieldctx around... I was hoping to remove that as this will cleanup the code quite a bit. Do you feel we really need this behavior?

[The-Compiler]

About the documentation, I plan to consider @fixture + yield as the recommended way of writing teardown fixture code and update all examples to use @fixture and yield where appropriate (those which don't have tear down code will be kept using return). So yield_fixture will be a footnote in the docs mentioned only as a historical fact. How do you guys feel about that?

Sounds great to me!

I see. We just have to consider that to keep this behavior we need to keep yieldctx around... I was hoping to remove that as this will cleanup the code quite a bit. Do you feel we really need this behavior?

Hmm, that's a good point. On a second thought I guess we can do without it, as hopefully people will get away from using yield_fixture anyways.

[nicoddemus]

Cleaned up the code and updated docs! 😁

[coveralls]

Coverage decreased (-0.02%) to 92.405% when pulling a91d035 on nicoddemus:issue-1461-merge-yield-fixture into b5bd4d9 on pytest-dev:features.

[nicoddemus]

This change is already on master, cherry-picked here to see green builds. 😁

[coveralls]

Coverage increased (+0.004%) to 92.433% when pulling e2040f4 on nicoddemus:issue-1461-merge-yield-fixture into b5bd4d9 on pytest-dev:features.

[coveralls]

Coverage increased (+0.004%) to 92.433% when pulling bdc2996 on nicoddemus:issue-1461-merge-yield-fixture into 9232389 on pytest-dev:features.

[nicoddemus]

Guys, would you like to merge this now or would you rather wait until the sprint?

[The-Compiler]

I'm a bit late to the party, but LGTM, and thanks for the great work! I think this definitely simplifies things a lot. 👍