Accelerating the JDK release cadence

Ben Evans benjamin.john.evans at
Tue Sep 12 05:47:39 UTC 2017

On Tue, Sep 12, 2017 at 1:11 AM,  <mark.reinhold at> wrote:
> In addition to divide-and-conquer, implementation features can be merged
> in an experimental mode at first (e.g., G1, AoT) and then, over time, be
> made non-experimental and even become the default, if appropriate, as
> they mature.

On the subject of G1, doesn't the new release cycle reopen the case
for delaying the cutover of G1 to default GC?

The G1 collector is coming on nicely, but in the field migrations to
it still fails extremely often - in my experience up to 30% of
applications are unable to move to G1 due to unacceptable performance

I've also heard from the Oracle engineers that the version of G1 that
is present in 9 is significantly different from what's been shipped in
any release of 8.

With the new 6-month release cycle, isn't this an ideal opportunity to
delay default cutover to provide some production feedback on this new
version of G1?

With 18.3 and 18.9 as possible release targets for default G1, is it
still so important to have default G1 arrive as part of Java 9?



More information about the discuss mailing list