cross-browser-tests-runner / cross-browser-tests-runner / 166

93%

push

travis-ci

Build Errors Fixes

- SauceLabs browser.js hook hangs
  - Various cans were opened. One significant and seemingly related
    one was that stdout I/O is asynchronous and calling process.exit
    after this can mean the I/O never happened. This seems to also
    explain what is happening with testem where it never reads the
    output of open.js.
  - However; with eliminating things one by one the above is not
    actually the case here. It was the case of Job getting created
    for bad input. When local runs of the integration tests were fine,
    they consistently hang in builds. This matches what goes on with
    such tests. Sometimes, the 1st call to JS status can get the 'test
    error' and sometimes it does not. Now, locally we see the former
    because of network delays being more, and on builds we see the latter
    with faster network. The fix is to adapt the test to expect not only
    a failure but also the success case.
  - All the research led to some changes in browser.js where we moved
    away from process.exit to process.exitCode = 0, and ensuring there
    is nothing on the event queue (cleaned up the timer that gets set)
    and yes, node does exit on its own. So this clean-up paves way for
    other places where we have used process.exit and possibly we can
    clean all of them up.

- JSON parsing failure on Linux node 8.5 build
  - It seems something is wrong with copyFileSync on Linux. The Mac node
    8.5 build worked fine, as also seen locally. This makes sense even
    more as copyFile and copyFileSync has been introduced in 8.5 itself.
  - Added our own simpler implementation of copyFileAsync and used it in
    both places where we use this API. This is a 'watch what happens' change.

- BrowserStackLocal was not found for chmod tests
  - Yep, the test sequence does not guarantee its availability. But why
    do we not see this failure on all integration tests builds? Mystery?
    Or something to do with Travis caching faults?
  - So changed the test sequence where a successful platform open test
    comes before the chmod one and that would ensure the tunnel gets
    fetched and exists locally

- Sleep tests fail
  - May be it's the VMs. Sleep tests that force the extra delay to be not
    more than 100ms fail (earlier it was at 40ms and failed pretty often).
  - Safer thing to remove the checks that force such expectations, as they
    are not coming true on VMs. On local Mac system never saw the delays
    being even more than 10ms.

SIGINT or SIGTERM to istanbul processes?
- It was seen that the browser.js SIG* handler was not getting called,
  even though changes were made to send SIGINT to all the istanbul cover
  processes used in tests. It was seen that browser.js does not get this
  signal at all.
- It was seen that SIGTERM does the job, and the fear that istanbul coverage
  won't be collected is wrong. It fixes the above issue as well as gives us
  coverage.
- So except one place, all the changes to send SIGINT while calling 'stop'
  method on the processe were reverted to go with the default (SIGTERM).
  These places are all related to browser.js invocation alone. So all of
  these reverts are actually required.

Conf updates
- Browser configs got updated with changes reported from the platforms

441 of 606 branches covered (72.77%)

Branch coverage included in aggregate %.

1538 of 1880 relevant lines covered (81.81%)

560871.87 hits per line