Merging jdk9/hs-rt with jdk9/hs
mikael.vidstedt at oracle.com
Mon Mar 21 23:59:06 UTC 2016
The JDK 9 development of Hotspot is primarily done in two different
mercurial forests: jdk9/hs-rt, and jdk9/hs-comp. In June of last
year we moved all the GC development from jdk9/hs-gc to
jdk9/hs-rt, and the experience so far has been a good one. Change
propagation (from jdk9/hs-rt to jdk9/hs-gc and vice verse) is now a
non-issue, we get testing faster on the union of the changes where
previously it could take weeks to catch a GC related bug in RT testing, etc.
However, both jdk9/hs-rt and jdk9/hs-comp still integrate through a
third forest - jdk9/hs, aka. hs "main" - before the changes are
integrated to jdk9/dev. In line with the previous simplification, we
would like to suggest a further simplification of the forest structure.
Specifically, we suggest that the work currently done on the jdk9/hs-rt
forest moves directly to the jdk9/hs forest. In addition to making the
forest structure easier to understand, this would have the benefit of
removing one set of integrations (jdk9/hs <-> jdk9/hs-rt), which further
reduces cost and propagation time. It is also paving the way for
eventually integrating up to jdk9/dev more often (but that is a separate
We suggest that the experiment starts on April 15th, and goes on for at
least two weeks (giving us some time to adapt in case of issues).
Monitoring and evaluation of the new structure will take place
continuously, with an option to revert back if things do not work out.
The experiment would keep going for at least a few months, after which
we will evaluate it and depending on the results consider making it the
new standard. If so, the jdk9/hs-rt forest will eventually be retired,
with an option of looking at further reduction of forests going forward.
At least for now, we suggest that jdk9/hs-comp remains a separate forest
and that it integrates through jdk9/hs just like it does today.
Much like when we merged the jdk9/hs-gc and jdk9/hs-rt forests we would
leave the jdk9/hs-rt forest around until we see if the experiment works
out. We would also lock it down so that no accidental integrations are
made to it. Once the jdk9/hs-rt forest is locked down, any work in
flight based on it would have to be rebased on jdk9/hs.
Please let us know if you have any feedback or questions!
More information about the hotspot-dev