Feature/cheaper long stack trace

[papandreou]

Attach a stack to all non-oathbreakable promises so that (a patched) Bluebird will render a long stack trace.

Perf. penalty seems to be about 3-4% according to chewbacca, whereas always forcing BLUEBIRD_DEBUG=yes regresses 22-25%.

[papandreou]

Breaks when running the tests with Phantom.js, looking into it.

[coveralls]

Coverage increased (+0.001%) to 97.927% when pulling bca0d9f on feature/cheaperLongStackTrace into 79a8149 on master.

[papandreou]

Note that the full versions of the traces that are captured this way won't be identical to the ones Bluebird would do with BLUEBIRD_DEBUG=yes, but I'm pretty sure that the trimmed versions of the two will be identical.

On that note -- how about letting UNEXPECTED_FULL_TRACE=yes imply BLUEBIRD_DEBUG=yes? Then we're down to exposing one stack-related parameter, which hopefully won't ever be needed unless you're debugging Unexpected itself or a non-npm linked plugin.

[papandreou]

Implemented that idea on the branch as well.

[coveralls]

Coverage increased (+0.003%) to 97.929% when pulling 1686e61 on feature/cheaperLongStackTrace into 79a8149 on master.

[sunesimonsen]

How will this test be run?

[papandreou]

Via test/external.spec.js

[sunesimonsen]

This seems like at really good tradeoff, I also like the idea of having one debug flag - it is not a hard contract, we can change that without a major version.

[papandreou]

The drop in coverage was due to another coveralls fluke (lemurheavy/coveralls-public#750). It looks good again after I restarted it.

[sunesimonsen]

👍