Support Rustfmt

[josephlr]

We will actually run rustfmt in a later PR

[dhardy]

Does this actually help anyone? I find it needless fiddling, debatable whether the result is an improvement, and potentially clashing with other PRs.

[josephlr]

@dhardy ideally there would be a way to just make sure that new PRs were properly formatted. That way the git blame wouldn't get messed up by switching over to this. However, rustfmt doesn't support such a thing IIRC. So the options are either:

• Not having any formatting tools for PRs
• This change (or something equivalent)

[josephlr]

As for clashing with other PRs, there's really only #14 and it's simple enough to just run Rustfmt on that PR.

[dhardy]

make sure that new PRs were properly formatted

In my book, "properly" means simply reasonably neat and consistent. Actually, I may be biased because of bad experiences with rustfmt in the past (frequently finding I have the wrong version, and it being impossible to work with old commits without bumping into formatting issues).

That way the git blame wouldn't get messed up

I don't count this very important (it's very easy to modify a line in insignificant ways).

As for clashing with other PRs

Sorry, this isn't an issue here.

I'll let @newpavlov decide since he's mostly responsible for this repo, but in my book having to do extra work to get PRs merged for very little gain is just not worth it.

[newpavlov]

I also had a bad experience with rustfmt, but I think consistent formatting is not a bad thing to have. (though personally I dislike some defaults) But I believe rustfmt failure should be allowed in CI, as otherwise it's quite irritating for contributors. We can run it manually from time to time.

I think it will be better to revert formatting changes in this PR and postpone them until other PRs get sorted out. In other words let's limit it to CI changes.

[josephlr]

I think it will be better to revert formatting changes in this PR and postpone them until other PRs get sorted out. In other words let's limit it to CI changes.

Done, the check now prints the diff to the Travis CI log, but will not fail the build. I'll do a separate CL to actually run rustfmt once all the other PR are merged/closed.

[newpavlov]

I think it will be better use allow_failures for rustfmt test (it will make it stand out more on CI page) and remove cargo tests (they are simply redundant).

[josephlr]

I think it will be better use allow_failures for rustfmt test (it will make it stand out more on CI page) and remove cargo tests (they are simply redundant).

Are the cargo tests redundant? I thought they were needed to make sure the tests pass on stable w/ Linux.

EDIT: I figured it out the tests are now running normally.

[josephlr]

This should be ready to merge. The rustfmt job fails, but the build as a whole succeeds.