Proposal to revise forest graph and integration practices for JDK 9

[Alan Bateman]

On 23/11/2013 22:08, Omair Majid wrote:
> :
> Could you elaborate on what is a 'breakage'? Would, say, a changeset
> that (unintentionally) makes hotspot fail to build on arm using zero
> constitute a breakage? Would a patch that fails causes non-closed builds
> to fail constitute breakage?
>
> Any ideas about what would constitute a 'prompt' response? Obviously for
> a patch that breaks all builds, prompt might be a few hours.
>
The jdk8/tl forest is one of the busiest integration forests today and 
when there is a build breakage (OpenJDK or Oracle builds) then folks 
tend to pounce on the issues quickly. There are many examples in the 
last year to point at. Frequently red faced individuals resolve mistakes 
before anyone notices (oops, forgot to "hg add" new files), sometimes it 
can be hours or days before a breakage is noticed (breaking boot cycle 
builds for example). So I would say that we've always had a culture that 
fixes build issues quickly. Clearly there will be cases where 
individuals see breakage because they using different tool chains or 
environments and I think the onus needs to be on those individuals to 
raise issues in a timely manner.

In any case, as you bring up the Zero port then I assume you are 
thinking about ports and build variants that many people might not be 
building on a regular basis. Once the AIX port is pushed then I can 
imagine us periodically breaking the port and it needing SAP or IBM 
engineers to fix up the issues. Having continuous builds or build+test 
infrastructure that cover all possible platforms and build variations 
would help but it might be too much to require everyone touching the 
build or doing platform or architecture code specific to keep everything 
working all of the time. Whether this means that "some animals are more 
equal than others" isn't clear to me but as OpenJDK takes on more ports 
then I can imagine this being a topic that needs further discussion.

-Alan.