Turn off ng under OSX-CI like we do for linux.

Review Request #3067 - Created Nov. 3, 2015 and submitted

John Sirois
2488, 2489
We always should have been doing this, but a recent string of ng related
failures brought the old CI configuration issue to light.

 .travis.osx.yml | 1 +
 1 file changed, 1 insertion(+)
This will only be exercised once on master.
Matt Olsen
John Sirois
John Sirois
Review request changed

Status: Closed (submitted)

Kris Wilson

.travis.osx.yml (Diff revision 1)

is there a reason why Linux CI intentionally does not use nailguns?

it seems like turning this off leaves us prone to undetected nailgun bugs in the integration suite - had this been /on/ in the Linux tests the commit that regressed OSX CI would have never landed as-is. with this on in OSX CI, it's now masking at least one known bug.

should we perhaps be heading in the other direction here in favor of comprehensive integration coverage?

  1. Not exactly sure except that it requires configuration - which on, which off - we only have 2GB in the container IIRC.  So its very much simpler to turn them all off.  If you follow change history, before the custom pants.ini, there was a -d switch in ci.sh that was used to turn off nailguns going way back.
    I think better than relying on tangential IT is writing one that actually uses the code purposefully in a test.  That can never be accidentally turned off.
  2. I agree, I would much rather see testing that explicitly exercises this code.

  3. gotcha, and I mostly agree - it just seems like it'd be a good safety net for current state. the downside of a purposeful test is of course that they must be written first vs the approach of larger swipes. the latter was definitely useful in this particular situation.