When will jdk9/client/jdk be merged into jdk9/dev/jdk?

[Volker Simonis]

On Mon, Sep 22, 2014 at 11:51 AM, Alan Bateman <Alan.Bateman at oracle.com> wrote:
> On 22/09/2014 10:06, Volker Simonis wrote:
>>
>> Hi,
>>
>> can somebody please tell me when jdk9/client/jdk will be merged into
>> jdk9/dev/jdk? I thought this happens regularly on a weekly base, but
>> maybe that's wrong?
>>
>> I'm just asking because I pushed a fix which repaired the AIX build
>> (8057934) to the jdk9/client/jdk ropository on September 10th but it
>> hasn't been propagated to jdk9/dev/jdk until now. I there something I
>> can do to speed up this process (i.e. can I manually push a change
>> which already is in jdk9/client/jdk to jdk9/dev/jdk?)
>>
> AFAIK, changes pushed to jdk9/client are integrated into jdk9/dev every 2
> weeks. The last integration seems to have been done by Phil on September 12.
> That integration was probably based on a snapshot of the outgoing changes
> taken several days before that so my guess is that your change just missed
> the bus. If the buses are running on time then it's likely that your change
> will get to jdk9/dev later this week.
>

Thanks for the information.

> Your mail is a reminder that we need to question again the need for
> jdk9/client forest. The latency issues are a big problem, esp. when working
> with project forests that pull from master. If a change that you need is
> pushed to jdk9/client then it can take many weeks to get to the place where
> you actually need it. The other issue is the clean-up and refactoring that
> covers many areas, it becomes totally uneconomical when contributors are
> forced to split the changes.
>
> There was a good discussion here (in Dec 2013/Jan 2014) about the
> integration forests that lead to the creation of jdk9/dev and jdk9/client.
> Key questions at the time were whether changes to the client libraries
> needed manual testing and the reliability of the automated tests. Testing
> client libraries is clearly a challenge and I think it would be good to
> establish where things are currently at. More generally, we've been living
> with the current arrangement for 9 months or so and it would be a good time
> to assess if people think it is working well or not.
>

The latency is for sure a problem for down-stream consumers like us.
We usually only build jd9/dev on a daily basis as we can not afford to
build all the different team repositories.

Another problem we have is that there's no possibility (or at least I
don't know of any) to bundle related/dependent changesets. So if a do
a follow-up fix F1 fro a change F and somebody will downport or
integrate F into another repo, he won't see that he'll also needs to
downport/integrate F1. Likewise, I as the author of F1 will not be
notified of any downports/integrations of F. I'll have to keep an eye
on it manually. But this is a different problem which could be solved
perhaps by by introducing new 'depends on/' fields to the bug system?

Regards,
Volker

> -Alan
>
>
>
>