uncaughtException: fix double EVENT_RUN_END events

[juergba]

Description

Per default Mocha tries to recover from uncaughtException and to map the exception to the correct test case. It's not possible to recover gracefully if a Runnable has already passed successfully and then fails asynchronously.

Firing an EVENT_RUN_END out of a suite other than the root suite is wrong and will result in the emission of double EVENT_RUN_END events. The first event will not abort the runner and a second event will be fired by the root suite. We already had a similar issue, see #3617.

To prevent double test epilogues the reporters have been "fixed" in the past by using runner.once(EVENT_RUN_END, ...).

Following tests were ineffective:

• test\integration\fixtures\uncaught-fatal.fixture.js:

√ should bail if a successful test asynchronously fails
  1) should bail if a successful test asynchronously fails

  1 passing (8ms)
  1 failing

  1) should bail if a successful test asynchronously fails:
     Uncaught Error: global error
      [...]

  √ should not actually get run      // runner keeps going after first end event

• test\integration\fixtures\regression\issue-1327.fixture.js

testbody1
  √ test 1
  1) test 1

  1 passing (8ms)
  1 failing

  1) test 1:
     Uncaught Error: Too bad
      [...]

testbody2              // runner keeps going after first end event
  √ test 2
testbody3
  2) test 3

Description of the Change

Instead of emitting an EVENT_RUN_END event we use Suite#bail() Runner#abort() to bail the runner cleanly.

Applicable issues

[coveralls]

Coverage increased (+0.07%) to 92.734% when pulling 1fbb9f0 on juergba/uncaught-bail into eed38d7 on master.

[juergba]

@outsideris I had to adapt my former solution.

Bailing the current suite bailed only the current suite, nevertheless subsequent suites were still executed.
Aborting the runner bails the runner completely. Runner#abort() is used also in our watch feature to abort the runner before restarting a new test run.