Proposal to revise forest graph and integration practices for JDK 9

[David Holmes]

On 25/11/2013 7:39 PM, Alan Bateman wrote:
> On 25/11/2013 01:47, David Holmes wrote:
>>
>> So this brings hotspot and tl (now dev) one step closer together. But
>> I'm unclear on the schedule for integration and testing ? I'm also
>> unclear on how eg hotspot-main PIT will work ie what forest will it
>> use for the rest of the JDK given that dev might not be stable at the
>> time of the PIT ?
>>
> I assume when you say "use for the rest of the JDK" that you are
> thinking partial builds (hotspot repo only in this case) but surely this
> has to change. That is, any testing of hotspot-main (or any hotspot
> forest) now needs to be done on a full build of hotspot-main and using
> the tests that are in the hotspot-main forest.

It can be a full build, but a full build of what? I'm assuming that a 
PIT build consists of the latest hotspot with the current contents of 
the intended push target. Now if that is a "master" repo then you have 
the issue (not new) of whether you integrate latest hotspot or latest tl 
first - until the second one you don't know if the integration will be 
successful. If the repo is just dev then as you describe below we have 
the stability issue.

Bottom line: needs to see some details please.

David
-----

> That said, I think the hotspot-main -> dev push will need consideration
> as it requires dev to be stable (for some definition of stable). Keeping
> dev stable at all times is probably the biggest challenge in this
> overall proposal. The closest thing we have to dev today is jdk8/tl and
> it is rarely stable, but neither is master (if "stable" is loosely
> defined to mean that all tests pass on all platforms in all
> configurations).
>
> -Alan.