JEP proposed to target JDK 11: 318: Epsilon: An Arbitrarily Low-Overhead Garbage Collector

Aleksey Shipilev shade at
Thu Jan 18 14:14:40 UTC 2018

Hi Roger,

On 01/18/2018 02:56 PM, Roger Riggs wrote:
> Does the entire test suite pass with this collector?
> In spite of Andrew's point, there are numerous tests of behavior that are effects/side effects of
> collectors that will fail. Does the JCK pass with this option (the all modes rule).

Haven't tried JCK with it. Epsilon comes with its own set of bounded allocation tests.

I don't think "all modes rule" applies to every single JVM mode? The applicability seems to heavily
depend on the flavor of the UseEpsilonGC/UseNoGC option. Currently it is "develop", so it is not
even available in product builds, and therefore JCK does not apply?

My view is that elevating this to "diagnostic" is also safe, because we do not provide reliability
and compatibility guarantees for those. Ditto for "experimental", although GC guys have
interpretation of it different from the definition in globals.hpp. If we do it as "product" option,
only then JCK would start to apply, right?

> It would be bad form to commit a change that causes a large number of regular tests to fail.

Sure. And we would need to exercise judgment in what tests to apply to which features.


More information about the jdk-dev mailing list