JDK submit repo

mark.reinhold at oracle.com mark.reinhold at oracle.com
Fri Mar 9 15:57:21 UTC 2018

2018/3/8 9:57:44 -0800, volker.simonis at gmail.com:
>  On Tue, Mar 6, 2018 at 7:05 PM, vladimir.kozlov at oracle.com wrote:
>> My understanding is currently it is not enough for *HotSpot* changes. You
>> still need to ask for sponsor to run additional Hotspot tests.
> At FOSDEM 2018, the OpenJDK Lead Mark Reinhold answered my comment
> that it is still not possible for external committers to push HotSpot
> changes with the words:
> "Actually it is possible!" (see 7:51min of [1])
> He further detailed:
> "If you are working on shared code and you run it trough the
> submission repo ... and all test pass you should be clear to push it."
> (see 8:16 min of [1])

The full context, for those who weren't there or who didn't watch the
video, includes two other statements by me:

  (8:08) "Consult with your local HotSpot gatekeeper, Jesper or whomever"

  (8:23) "Jesper and I were chatting about this a couple of weeks ago;
          he plans to communicate this so that people understand."

In other words, in real time I clarified my initial response ("actually,
it is possible") to say that Jesper (or someone) would propose some
concrete guidelines for how the submit repo is intended to be used.

Jesper had in fact already addressed this issue prior to FOSDEM [1],
noting that:

2018/1/24 5:21:40 -0800, jesper.wilhelmsson at oracle.com:
> This is not a complete answer, but I'm working on it.
> Currently the set of tests run in JDK submit does not include all the
> HotSpot tests that we run on all pushes in jdk/hs. My hope is that we
> will be able to sync that going forward to make the JDK submit more
> useful for HotSpot developers.
> Internally we do no longer use the JPRT pre-commit test system that
> some may have heard of in the past. Instead we use post-commit
> continuous integration (CI) that is running several tiers of
> tests. This is one step closer to allowing non-Oracle Contributors to
> push HotSpot code directly. Once the JDK submit runs all required
> HotSpot CI tiers we will be almost there.
> 2018/1/24 0:24:19 -0800, volker.simonis at gmail.com:
>> If the answer is "yes", there's still the problem that hotspot changes
>> should go into jdk/hs but the submit repo mirrors jdk/jdk so what I've
>> tested in the submit repo might be slightly different from what I will
>> actually push.
> This is another problem. I'm trying to enable more frequent
> integrations between jdk/jdk and jdk/hs, but currently there is up to a
> week delay in getting changes from hs to jdk. Again it is a question of
> when and where we run different HotSpot tiers.

These weren't full guidelines, of course, but his message hopefully made
it clear that further work was needed to get everything aligned -- more
than you might assume given the very simple interface of the submit repo.

Since FOSDEM we've made some additional internal changes that ensure that
changes pushed to the submit repo will receive adequate testing, even if
those changes affect shared HotSpot code.

There remains the problem of the submit repo being a child of jdk/jdk
rather than jdk/hs.  To address that we'll shortly set up a second submit
repo that's a child of jdk/hs.  Once that's up and working then any
HotSpot change, from any contributor, that passes the tests run against
the submit-hs repo will be clear to go directly into jdk/hs (modulo any
temporary gatekeeping restrictions, which will apply to everyone and be
announced in advance).

- Mark

[1] http://mail.openjdk.java.net/pipermail/jdk-dev/2018-January/000579.html

More information about the jdk-dev mailing list