On 12/16/2016 02:45 AM, Augie Fackler wrote:
Right now test-check-commit.t et al can mark all the builds as broken, even
when it’s just a commit that we’ve decided to take despite some sanity
checker’s objections. Usually it’s around vendored code, though today it’s
around my unittest-ification of test-bdiff.py.
Given that these results are fairly stable across platforms, and they’re not
platform-specific, how would people feel about moving those builds to a
separate builder so we can see “lint results” separately from “correctness
results”?
If nobody is opposed, I’ll work on the buildbot config change with Kevin next
week.
It is true that check-commit complaining about thing we don't care a bit
too often. I'm really not thrilled at having a new builder target
intended to be broken on a regular basis. Instead, we should fix the
problem rather than the symptoms. I would rather have a way to mark
commit as "skipped-by-check-commit" and the tests to stay green.
For example, we having test-check-commit skip commit with a description
containing "#nocheckcommit" is simple to have.
The idea of an independent target has some other merits, running all the
tests on the builder is quite long (1h30), having a small target with
just to test-check-*.t can give use feedback in less than one minutes.
But I'm not sure it is worth putting effort in that since I expect all
reviewers to run the tests before they push so that would not be too
useful (sometime people forgot, but it is rare and sometime pyflakes is
not around, but they will install it once). In all case we should
implement check-commit exception first to keep such target happy and
passing. (And we should keep then in plateform specific tests).
Cheers,
--
Pierre-Yves David
_______________________________________________
Mercurial-devel mailing list
Mercurial-devel@mercurial-scm.org
https://www.mercurial-scm.org/mailman/listinfo/mercurial-devel
