We should also add -D warnings to the CI so it fails in case of warnings.

Uh, the page you linked to explains why that is bad idea...

I'll test the CI setup in this PR (cause I'm new to CI configuration), then force-revert, and then open a new PR so we can discuss this separately.

Don't merge yet.

Tests successful – my CI configuration changes should work as expected. I've force-reverted back to the initial commit, so this PR is ready to merge again.

The CI configuration change can be found in my branch ci-config-1 which builds on this PR.

(I'm not going to fix the mem::uninitialized() deprecation warnings right now, hoping someone else beats me to it. See #1025.)

@rukai Hmm, probably should've done more than just skim that article.

We could do -D warnings -A deprecated. I'm fairly OK with Travis catching warnings for compiler mis-features, since we'd want to fix those ASAP.

I think mem::uninitialized() should be easy to fix (see #1025), and I don't think we have other outstanding warnings (besides my warnings PR #1020), so -A deprecated is not needed.

@felixrabe The -A deprecated is mainly for future-proofing PRs so they don't fail unnecessarily, since there's a decent chance we're using APIs internally that will get deprecated sometime in the future.