Fix and tweak integration tests

[nrc]

And emit rustfmt output so they can be debugged

[nrc]

Ping @gnzlbg - I've been trying to get more of the integration tests passing, but I can't get any useful info to debug. I've tried a few of these crates locally and they pass rustfmt-ing. Do you have any idea why the integeration tests are failing and why I don't see the output from rustfmt?

[gnzlbg]

Note that stdsimd was broken because the doc tests failed to build after a reformat. IMO we should be checking that if the library builds correctly before formatting, it should also build correctly afterwards, and that applies to all its tests, including doc tests.

Running the tests is not necessary, but the time it takes to run the tests of these libraries is negligible compared to the time it takes to build the tests. And knowing whether formatting changes alter the results of the tests in these libraries allows us to at least ping their maintainers on the nursery, or send PRs when things break.

[nrc]

Ah, I see. I'll add that back in once I've worked out what is going on with the rustfmt failures

[gnzlbg]

Many of the failures are either warnings because the crates need pull request to remove deprecated options, or because rustfmt emits warnings or errors that the user cannot do anything about.

If there is nothing that the user can do about a warning or internal error, then the diagnostic should IMO be opt-in (e.g. behind a rustfmt --debug-output or similar).

I've tried to enable these diagnostics here because even if they are opt-in, they should be fixed at some point. However a "red" allowed to fail travis build bot because of a warning can hide a change that actually breaks the build so... I did not know what the right way to proceed was here and did not wanted to add all buildbots twice...

[nrc]

What's puzzling me is that some of the errors are meant to be opt-in with --error-on-unformatted and when I run locally I don't see the errors, but running on CI is doing something different to my local setup.

[gnzlbg]

Maybe is a different rustfmt being picked up? I had that problem when I enabled the cargo cache on travis, so I disabled it. Or maybe your local rustfmt is different than the one here?

[nrc]

I figured it out! The directories here are nested under rustfmt, whereas my local tests were not, that means they see rustfmt's rustfmt.toml which opts into the debug mode stuff. Testing now without that...

[nrc]

r? @gnzlbg

[gnzlbg]

This LGTM. A way to fix #2274 would be to run cargo test before the formatting, and if that fails just return 0 instead of 1, e.g., by adding

cargo test --all	
if [[ $? != 0 ]]; then	
      return 0	
fi

before the touch rustfmt.toml line. That should make stdsimd pass for the time being, since it currently doesn't build (but we should still leave it as allow_failure).