jdarcy at redhat.com
Tue Sep 6 12:22:26 UTC 2016
> > Would it be possible to have a workflow where verified +1 vote from the
> > developer indicates that the regression tests have passed in their local
> > setup?
> I *really* hope that is the case already!!
It doesn't seem that way. I'm not even trying to address the reasons or
whether they're good; that's just an empirical observation. I'll also
add that it's not that hard to have regressions pass in developer's own
environment and fail in the "official" one if there's any difference at
all - e.g. OS or compiler version, installed packages - between them.
> I think that local testing is always a must. It is not like it is
> difficult to run the tests in a VM.
The run-tests-in-vagrant stuff makes this *super* easy.
> I admit that I do not always run the
> whole suite, but at minimal the tests for the component that is affected
> by the change I'm going to post. I wait with Verified=+1 until I
> actually did run the tests locally.
If it's a subset of tests - e.g. tests/basic - I'd say we might as well
run them on the official infrastructure and go back to giving V+1
automatically. At least then we're sure of what was run. Otherwise,
if it's a full set of tests, we have to overcome developer reluctance.
More information about the Gluster-infra