Allow nightly builds to fail

[weiznich]

If a nightly build fails a warning instead of an error is shown in the
build console

[weiznich]

Hopefully this makes #2117 obsolete.

[weiznich]

Would be great if azure pipelines would report that there were warnings. They are shown in the build status over there, but I'm not sure why they are not reported to github

[JohnTitor]

Compiletest in Travis is failed, probably this line should be also changed.

diesel/.travis.yml

Line 46 in 0b6c59e

- rust: nightly-2019-02-26

[JohnTitor]

Ugh, ONCE_INIT is deprecated in nightly-2019-08-01 so compiletest is failed again...

[JohnTitor]

This PR seems good to me so we can merge this now then we can go to #2108, or this can also include its changes (or add --cap-lints=warn to travis' compiletest).

[JohnTitor]

Some targets are still queued on GitHub but all green on Azure Pipelines, lgtm

[JohnTitor]

Can we go? Or should we wait for another review?

[weiznich]

I would like to hear at least another opinion about the use of --cap-lints here.

[Eijebong]

So IIUC, the cap-lints part is to make depreciation warning stay as warnings even with a deny(deprecated) ?

[weiznich]

So IIUC, the cap-lints part is to make depreciation warning stay as warnings even with a deny(deprecated) ?

Yes, but only for the nightly compiler. It is basically meant as a poor mans allow_failures: nightly. Ideally azure would indicate that some job generates warnings, but I've not found any way to do this yet.

[Eijebong]

That would not work on the event of something erroring for real then ? That's kind of a bummer because a dependency might be broken on nightly for some reasons and we don't want that to make the build red in that case either

[weiznich]

Right, if we want to prevent that we would need to set the this return here to true. I'm not sure if we really want to do that, because ~all of our breakage from the last time was basically rustc introducing some new lint/deprecation/… in nightly.

[Eijebong]

Yeah but I've seen stuff erroring due to errors in dependencies in the past on other projects. I guess this is not common enough to worry about especially since it's fixed quickly most of the time.

I wish rustc would add a cfg for nightly compilers so we could do #[cfg(not(nightly), deny(warnings))] or something...

I guess this is fine then 👍