Use Travis for npm publishing

[tivac]

New release process:

1. Programmer enters npm version <type-of-change or new version> -m "v%s"

Everything else is fully automated, here's what will happen.

1. Tests are run
2. Linting is run (but doesn't fail build)
3. Version number in package.json is incremented
4. New bundles are generated using updated version
5. git add called on bundle output
6. package.json and updated bundles are committed to git
7. previous commit is tagged using new version number
8. git push --follow-tags pushes up new version commit & tag to github
9. Travis sees new release, starts build
10. Travis generates new bundles before running tests
11. Travis runs tests
12. Travis lints files (but can't fail build)
13. If build fails, abort
14. Build succeeded, so travis will commit back any changes to the repo (but there won't be any)
15. Travis sees that this commit has a tag associated with it
16. Travis will use the encrypted npm creds in .travis.yml to publish a new version to npm

😩 phew, that's a lot of work that a human never has to do again. This is basically the process I use to release most of the packages I maintain on npm and it's super nice to not really have to think about it. So long as you use npm version (the -m "v%s" isn't required but looks nicer as a tag name IMO) everything else just goes w/o much effort.

There's a little bit of duplicated effort between the local process and travis, it's mainly to ensure that the commit tagged v1.0.0-rc.8 actually points to the commit that bumped package.json and m.version in mithril.js/mithril.min.js to the new version. Otherwise there'd be a weird mismatch that I think people would find confusing.

[tivac]

Not quite working, since the build output didn't change it looks like Travis doesn't actually run the deploy task. Need to investigate further.

Here's the build that "worked" but stopped running, https://travis-ci.org/lhorie/mithril.js/builds/191845435

[tivac]

It was due to my "early out" calling exit 0 which ended the travis bash instance. Yikes, fixing.

[tivac]

@lhorie just a heads-up.

I believe this is fixed as of 59dfbf0, sorry about that. BASH scripting isn't something I have a ton of experience with so it takes a bit of flopping around usually.

I don't want to bump the version again & re-publish because both rc9 or rc8.1 would be ridiculous, especially given that there'd be no actual content change from rc8. I could unpublish rc8 since we're within the 24-hour window and try again but I don't think that is a great plan either.

So this code will just hang out, and we'll need to remember to use npm version when it comes time to cut a rc9.

[lhorie]

Thanks @tivac

I think we should add the release process somewhere more official so we don't need to search back for this PR

[tivac]

Agreed, will document tonight. I think I can also have Travis amend the previous commit so I can remove the need for the locally built copy. Gonna play with that in another repo first though.