do not eat exceptions thrown asynchronously from passed tests; closes #3226

[boneskull]

Currently, Mocha eats the error thrown below, because the test has already passed:

it('should bail if a successful test asynchronously fails', function(done) {
  done();
  process.nextTick(function () {
    throw new Error('global error');
  });
});

If the above happens, Mocha can't recover gracefully b/c of #3223. This PR causes Mocha to bail if the above situation is encountered. Exactly how this is reported is "best effort", based on queued tasks and timing--the reporter output may not be pretty, but you should at least get a traceback.

One thing we can do to mitigate (some) #3223-related weirdness is force the reporters to only respond to the end event once, which is what I've done. Without this, machine-readable reporters in particular--such as json or xunit--get their wires crossed. No reporter should be dumping the summary twice anyway.

Also added a couple helper functions to Runnable.prototype.

[coveralls]

Coverage increased (+0.07%) to 90.014% when pulling 8186723 on issue/3226 into ec8901a on master.

[Bamieh]

Looks good, I love the usage of once instead of on.

[boneskull]

Looks good, I love the usage of once instead of on.

yeah, but it's actually a hack around the real problem, which is that end shouldn't be emitted more than once.

[Bamieh]

@boneskull regardless if it is triggered once or multiple times, we should be using once to automatically detach the event listener.

[rsheldiii]

On upgrading from mocha 3.5.3 to 5.2.0 (latest), we noticed that our test coverage started silently skipping a bunch of tests while still passing. I bisected and found that this commit is the commit that causes it. I'm aware now that this is because one or more of our tests are throwing an error in asynchronous code, possibly after they pass, but we've got quite a few tests, so it's going to take a while to pinpoint. For now we're downgrading to 4.1.0, which still runs the entire suite. Just want to leave in case it helps someone

[plroebuck]

@rsheldiii, This commit?

Two choices for your actual problem:

• Run Mocha-5.2+ with --exit cmdline option which provides pre-4.0+ exit behavior

$ mocha --exit my-shitty-code.spec.js 

• Find and fix problems in your test suite. The wtfnode package can help with this.

$ wtfnode ./node_modules/.bin/_mocha my-shitty-code.spec.js 

[plroebuck]

Note that the change to .once for the reporter event handler also prevents mocha instance from being reusable for those using it programmatically. We need another method to recreate the internal pieces/parts (e.g., reporters, growl) that have event listeners using .once.