Rethrow unhandled promise rejections in tests

[foolip]

In https://travis-ci.org/whatwg/misc-server/builds/293854231 this
happened without failing the build:

(node:4452) UnhandledPromiseRejectionWarning: Unhandled promise
rejection (rejection id: 1): FetchError: request to
http://resources.whatwg.org/ failed, reason: connect ETIMEDOUT
165.227.248.76:80
(node:4452) [DEP0018] DeprecationWarning: Unhandled promise rejections
are deprecated. In the future, promise rejections that are not handled
will terminate the Node.js process with a non-zero exit code.

[domenic]

I think the more correct way to do this is

test().catch(e => {
  console.error(e.stack);
  process.exit(1);
});

As written right now the process will exit the moment any rejection goes unhandled, which works poorly with your perform-all-requests-in-parallel-then-report-all-results-cumulatively strategy.

[foolip]

That does look nicer, but also results in the test exiting at the first error. I think that's OK though, to deal with network timeouts I think I'll add in some retry logic.

Any reason to use e.stack as opposed to just e? They result in the same output in my throw new Error('banan cake') test, but just e would work if something other than an Error is used.

[domenic]

Hmm, I thought it was different in some cases, but I could be wrong and just cargo-culting myself from years ago. Sticking with e seems fine.

I'm quite surprised that exits at first error given how you catch all errors and turn them into non-OK status/ok values.

[domenic]

Actually, wait, I don't understand the OP. You process.exit(1) on non-OK status. So how did that rejection get triggered in the first place without failing the build?

[domenic]

I see, I was confusing the two test scripts. You should probably make them more parallel in structure, e.g. try/catching failed fetches in both cases.

[foolip]

Rather than continue to make the testing framework more fancy I went to find a popular framework and used that: #25

This means the tests are no longer run in parallel, but I think it was totally worth it. Closing this in favor of that.